I am aware that a large segment of the HIPPO community advocate visit based metrics, but honestly, one can get into trouble – big trouble – if one blindly adheres to “Visits are what count.” and apply visits like ketchup to every metric entree.  In web analytics there are three basic tempos – page views, visits, and visitors (actions over multiple visits). Page views represent the raw vibrance of your web site – lots of visitor actions that are difficult at times to summarize and trend.  So to gloss over the visitor churn and their propensity to reload or revisit pages, we often apply visits to metrics to put a wet blanket, metaphorically speaking, over the fire that raging in the page view traffic.

Serialization Counts

This is called serialization where we attempt to remove false or duplicate counts to report only the real counts.  By selecting visit instead of instance counts or serializing events by session ids,  we eliminate most of the false counts at sacrifice of some of the true counts.  Though this approach helps stabilize metrics, it can often make the metrics insensitive to changes in business decisions and visitor responses to those actions.

Don’t Curb Your Enthusiasm!

For example, imagine I am a business that depends upon customers placing items into shopping carts, and checking out shopping carts to make purchases.   If I only counted one shopping cart and one purchase per visit, clearly I would get different metrics than if I allowed my shopping carts and purchase metrics go free.   If I had promotions that increased shopping cart activity and purchases, my visit normalized metrics would show a less enthusiastic view than what may be raging (under the visit blanket) on the actual web site.

OK – so this may seem contrived. Who would not count every purchase? On the other hand, that should be our attitude in everything we count.  Why would I not want to count the event every time it occurs regardless if it happens multiple times within a visit or once at the end of many visits?

We Do It All the Time!

Another example, why do I want to count just entries that start a visit and not every time the visitor comes to the site from an external referrer?  The latter can occur multiple times within a visit or not occur again till many visits (These are called “Return Visits” or “Repeat Visits” not to be confused with new and returning Visitors … duh!?!).  I may be interested in how many visits include a sign-in, but why would I not also be interested how many times a visitor signs-in during a visit?  You may assume only once, but does your data confirm that assumption?

Blame It On the Solution Designer

For the solution designer, the question is then “how can I tell if the event is real and not a reflection of visitor navigation churn?”.  This involves establishing a “serialization” strategy in the client side instrumentation or for Omniture’s SiteCatalyst adding a serialization key to the event.  Too often a session cookie or session-id is too readily available to apply to collection essentially making the count a visit based metric.

What is required is that the designer take the extra effort to understand the real event being counted, and expand his repretoire of implementation tools to either declare the event only when it is true or find and establish a more appropriate serialization key. For example, may be a transaction key for a service task would be more appropriate than a visit id.

The Pain Can Be Real!

A client of mine is going through the pains of having their Key Performance Indicators transition from visit based to transaction based metrics dramatically changing their view of the business but at the same time abruptly breaking from the historical data that has been collected for years.  Like a blind person who has suddenly been given sight, there is an urge to pull out their eyes and return to their old “vision”.

Stay the Path: Real is Good!

It is too early to reap the benefits of the new vision, so we have to proceed assuming that with a vision that reflects more accurately visitor behavior, we will be able to more accurately measure effects of our  business decisions.  Whether it is marketing campaigns, changes in web design, or attempting to understand what the customer is saying (VoC), without true visitor based measures that are sensitive to visitor based actions, it is impossible to optimize performance.  The visit based metrics were like a wet blanket on the performance measures we needed to access these changes.

I’m No Einstein but …

All I am saying is perhaps a little more thought should be taken in setting up and serializing events.  Quick and non reflective solutions can truly lead to pain down the road.  Though it is our desire as analyst to make the trends evident and insights pop out from simple reports, we must still keep in mind as Einstein would say: “The solution should be simple but not too simple.”

In my line of work consulting with various customers I have been able to see common threads in how different companies approach web analytics and apply it to their business.  One puzzling aspect that becomes immediately obvious to see but takes longer to understand is how many reports seem to cover only a small portion of web behavior with most visitors and visits pushed into the general category of ‘None’.  None meaning that visitor does not belong to any of the categories for the dimension being reported.

A Puzzling Predicament

What is puzzling is: How do such reports provide useful information on visitor behavior?  The next question that comes to mind is who are those people that belong to ‘None’ or for the person that implements the report: Did I set up the metric correctly? In brief, How many of that are None should be Something?  This has lead me to the concept of “None is not an Option” reporting, meaning None is not a category of behavior but an indication of analytic failure or incompleteness.

As a consultant I can typically give advice but too often not work through the details implied in the advice.  In my current position I have had the opportunity to be the designer and implementer and put this concept to practice.  As with most theory, implementation of the theory presents its own challenges and insights.  Here I will share what I have learned.

Fragmented Segments

Because we will be dealing with implementation I will focus on a specific tool – Adobe (Omniture) SiteCatalyst 14. The concepts can be applied to Google Analytics though it has a different scheme for defining custom variables or data dimensions. Google forces one into schemes of grouping variables into limited number of tag plans (called slots) that one has to consider the strategic implications of the data being collected.

With SiteCatalyst one can quickly get into the mode of treating page properties (sprops) or event dimensions (eVars) as simple fields that one collects and then stitches together later.  The unfortunate out come of this approach is that one quickly runs out of the available number of custom variables that Omniture allows, though at first they may seem inexhaustible.  On top of this, one has the rude awakening that ability to relate these various fields in reporting has limits to number of correlations or full relations that allow one to pivot or drill down from one dimension to another.

Misplaced Expectations

When I confront the customer with the fragmented segmentation and non-strategic value of the SiteCatalyst reports, the response back is “We can get most of what we want from Discover”.  Discover is Omniture’s data warehouse exploration tool that removes most of the irritating limits of SiteCatalyst and allows one to cross correlate all dimensions.  With SiteCatalyst 15 many (not all) of these limitations go away so one will be tempted to continue this approach without further reflection.

The problem is that though Discover and SiteCatalyst 15 remove limitations on how variables can be correlated and queried, they do not address the underlying concern – fragmented segments.  These are segments that may or may not have relation to each other in that they clearly represent segments that most visitors do not belong. This shows up as a large percentage of visits or visitors that belong to “None”.  On a practical level, these variables represent a waste of a perfectly good slot that could be used to collect data that covers more visitors.  So we got to do some heavy data minding in SiteCatalyst before we can do data mining in Discover.

Analytic Capabilities as Strategic Plans

Clearly there are situations where the data collected will apply only to a small portion of visitors or for specific tasks that the visitor performs (acquisition, on boarding, marketing, etc.)  So as an analytic solution designer one is attempting to deal with a large number of different demands from different business units for data specific to their special needs. Without a strategic plan or master solution one is forced to service these request on a one by one basis leading to ad hoc solutions for each case.  This may satisfy the customer’s immediate need but rarely provides a lasting solution because there is always the next question that arises from the data.

The Fundamental Question

As a data architect (web analytic professionals are called upon to perform many roles) one has to consider once the requested data is collected how will it be used and integrated into the business process. So a good followup question back to the customer when presented a list of data fields that should be collected is: If you had this data, what would you expect to find and what decisions would you expect to make resulting from the findings?

The back and forth in addressing this question will usually lead to development of a web analytic capability that addresses a business need rather than describe an application or sequence of pages.  Since business processes are more formalized than applications, there is good chance that the resulting analytic capability can be applied in other aspects of the business and across all the web properties of the company.

How Analytic Capabilities Pop Up

Analytic capabilities are the foundation of all current web analytic tools. Web analytic providers have from the very beginning looked for common needs across all their customers to provide reports that meet those needs.  Whether it is tracking marketing campaigns to the customer’s site, tracking shopping cart activity across the various stages of the acquisition process or simply evaluating the effectiveness of forms most tools provide “canned” reports that address these needs.  More recently special reports suites have popped up to assist companies in evaluating their effectiveness in social media marketing!

Plug-In or Customize?

This is all well and good if your business fits the model represented by the analytic capability and the reports satisfies your business needs.  However in application one still needs to adapt the capability to the specific needs of the business.  So to hedge the market, web analytic providers provide custom variables and reports to allow businesses to define their own analytic capabilities.

Omniture’s offering is particularly good in this respect allowing even their standard packages such as the shopping cart to customize names of variables and events to their customer’s specific requirements.  They also provide a rather large collection of ‘plug-ins’ that add specific capabilities to their standard instrumentation (collection) script.  One that I would highly recommend is their Channel Manager plugin that results is an All Sources Report for all external referrals (paid and non-paid) to a site!

Up and Down Sides of Flexibility

With such flexibility one is provided great opportunity but with the very real downside of providing enough rope to hang yourself.  When using custom variables and reporting one must continue the discipline of identifying analytic capabilities that address general needs that can be adapted to specific business requirements.

For example, the company I am working with now has an acquisition process but does not follow a shopping cart model.  The events and special variables associated with shopping cart model can be adapted to a degree. There are products and purchases of products but the business does not refer to them as products and purchases, they have their own business terms.

Special KPIs (success events) and segments (eVars) must be defined which may provide answers to overall performance but the next questions of coarse are “Why are the KPIs what they are?”, “Can we improve them?”. Therefore, in this case, this means tying funnel reports to pathing reports and ensuring drill down (correlation) between KPIs and segments.

The result has been collection of web analytic capabilities designed specifically for the business needs but sufficient to cover all acquisition behaviors and visitors in SiteCatalyst (focused on understanding normal and daily monitoring) and Discover (focused on … well .. discovery).

How to Define and Design a Web Analytic Capability

As was mentioned above, developing analytic capabilities is a discipline that has to be carried over from the web analytic providers to the custom reporting.  As a discipline we should be able to describe the various activities involved in its development. To this end let us start by formalizing the concept of Web Analytic Capability.

A Web Analytic Capability has a well defined purpose that is usually expressed in business (not application) questions that can be addressed by the capability.  This is then translated into sets of reports and the data that must be collected to support these reports.  Another essential element that adds power to the reports are the decisions or actions that can be taken based upon the data and/or reports.  In other words, design of actionable reports.

Business Requirements without Analytic Jargon

The requirements are done independent of the specific analytic tool jargon.  In the case of Omniture, talk of eVars, sprops, success events, correlations, relations, pathing etc. must be expunged from the conversation.  Campaigns, call to action, key performance indicators, channels may be appropriate depending on the business. The discussion is with business owners that make decisions and data analysts that support those decisions. Therefore the requirements should be expressed in their business terms and translated later into web analytic jargon in the implementation.

Universal Rather Than Particular Solutions

In the solution design you are attempting to develop a capability that is “universal”. In this case universal means capabilities that cover all visitors or segments of visitors.  If the request is to track a specific link, the response should be why not track all links and provide means of selecting the specific links the customer is interested in.  If the request is to segment visitors by specific actions provide a mechanism to expand and identify other actions in the future.  In the case the of Channel Manager mentioned above, if the request is to track marketing campaigns to your site, why not track all of the sources to the site providing an All Sources Report to track everything.

In the back and forth between need and solution the result should an analytic capability that is understood by the business and the solution designer to the extent that the capability can be used to meet the business need beyond the specific concern of the immediate project.

Universality’s Relation to None

One of the implications of universality is that all visitors are categorized and None is not an option.  Even if the visitor does not belong to any of the specific segments of interest currently, place them into a segment such as ‘Unknown’, ‘Currently Not Categorized’, ‘Not of Interest at this Moment’. Anything but ‘None’ or ‘Unspecified’ which are terms used by SiteCatalyst to bucket all the counts that cannot be attributed to the specific classification or variable.

This “Unknown” category will serve two important purposes. First it will identify the scope of the specific parameter or segment.  That is to say the total number of opportunities that the data can be collected compared to the specific data of interest. Second it separates the opportunity from non-opportunity represented by ‘None’. In others words ‘None’ represents not being able to collect the desired data and hence a valuable diagnostic tool in ensuring quality data.

Even if you don’t expect to collect the specific data for all visitors there are likely cases where you do expect the data to be a necessary prerequisite for a latter user action or event.  In these cases the KPI should have no attributions to ‘None’ and if some are found you have your first clue that something is wrong and a need to track down the cause.

On the other side, the ‘Not Categorize’ category can be correlated to other dimensions allowing for potential identification of new segments that can be tracked, eventually moving visitors into other categories and providing more insight into the process being observed.

The Concept Put To Practice

By defining web analytic capabilities you will progress from simply collecting behavior data to understanding how behavior should be reflected in the data collected.  The capability is sufficiently  abstract to provide a guide for how to address not only a specific problem but the entire class the problem represents.  This in turn provides a framework for categorizing behavior and actions that is more complete and where None is not an option.

Get the Questions Straight

For example,  a frequent request starts with “We have a form that the customer must complete …”.  So one might respond with “Form Analysis” but as the discussion evolves the issues have much greater scope and questions more probing than “Where is the user having troubles and abandoning the form?” Not that it is an important question but just not the first question.  The larger question is task completion: Those who start a task (form, wizard, web flow) how many complete the task?  Or task relevancy: Is the task we gave them the one they wanted to perform?

What type of task are they performing?  Acquisition (Cart Checkout) and On-boarding (Registration) are tasks where the main objective is getting the visitor from Start to Finish as quickly and as simply as possible without diverting and distracting the customer with other tasks.  Service tasks on the other hand are dynamic where the user has more freedom to preform many actions within a task process.  These are user convenience features that need to be tracked and understood.

So looking at the task as a process where the form may be an element, the more relevant and immediate questions are:

1. How and where do users enter the task?
2. What scenarios can the user perform within the task process?
3. What has been accomplished when the task is completed?
4. Finally, where and why do users abandon the task once started?

Start with the First Question

The first question is addressed by tracking both external and internal channels so that when a user starts a task we know how they came to the site and what they viewed prior to starting the task.  One may go a step further and provide click placement tracking to understand where on the referring page the user clicked to start the task. Here the “None is Not an Option” solution is obvious: track all sources coming to the site (Paid AND Organic) and track all channels within the site.  If you are doing click placement tracking consider tracking all links.   When this is accomplished, None means a channel or action that is not tracked.

Triage Business from Analytics

The second question is more specific and takes more effort to analyze and abstract into an analytic capability.  The business owner wants to know everything and the significance of every field entered and action taken.   So triage most be done to separate what represents business concerns and what reflects user behavior.  The data for the latter can be collected with web analytics with the former best left to back end transaction engines, CRM and other business data warehouses.  The issue here for the solution designer is establishing a way to connect the data on the backend to “color” business facts with user behavior data.

Establish Realistic Expectations

Categorizing the scenarios preformed within a task (especially a service orient task) has additional challenges.  Is this first time the visitor performed the task? If they are returning how do their previous visits effect the current visit?  When is the intent of the visitor revealed – at the start, during or at completion?  If they select certain options or perform specific actions within the task, is it truly of strategic importance to capture these machinations or can this be part of what is reported at the completion of the task?

There are many more questions that can be asked but from a web analytic capability perceptive, the NINAO options are:

1. At the start of the task, have I captured everything I know about the user that will define or effect the user experience during the task process?
2. At the completion of the task, have I captured everything that was accomplished during the process hopefully available on the confirmation of the successful completion of the task?
3. For actions within a task process, how can these be best captured: Milestone (Step) events, click events, or changes in content generating page view events (even if the action does not result in an HTML page transition)?

Mockup Reports to Confirm Design

Once having defined what you believe covers all entries and exits from the process cross check this with your measures (events) and how they will be segmented and broken down.  This exercise will result in mockup reports that can be reviewed with the business owner and analysts.

In the back and forth exchange, the reports can be tuned and simplified such that what is important to understand is clear and presented succinctly in the proposed report and not encumbered by user machinations that can divert attention from what is important to understand and follow at a high level.

Now the Other Stuff

Now you can consider Form Analysis, Fallout Reports, Page Pathing Reports that will help to answer questions that will arise from the reports you have defined for task analysis.  From my experience, I provide Funnel (or Waterfall) Reports that provide a means of understanding the overall performance to business owners but then can be broken down and analyzed at the next level by data analysts using these more detailed reports.

From Solution Design to Implementation

Once solution design is in place, you can now consider how the design will be implemented in a particular web analytic tool. The solution design should identify how the data will be collected and the options for configuring the web analytic capability to a specific solution, in this case task.  The analytic capability will have to make public to the web designer or application developer what is expected for the data collection.

For example, with Form Analysis it is expected that the web designer use the standard HTML elements that make up a form.  But with the Dojo and other web toolkits this is not as straight forward so the standard web analytic provider form analysis plug-in will not work.  Even if it does, how are the forms identified for analysis and the fields named for reporting?

Application Execution Context API

This leads to identifying and formalizing the API between the web application and web analytics and how the web analytics gets access to the application execution context (AEC) on the server or client.  With standard HTML the AEC is the Document Object Model (DOM) and HTML Event Model but in most cases this not sufficient to collect all the data required so there must be mechanism to bring the relevant business data into web analytic stream.  Traditionally this is done by including static analytic tags on each page that the web analytic instrumentation collects and sends to the web analytic provider (WAP).

Dynamic Instrumentation and Data Collection

However more and more it is necessary to dynamically collect and process the data on the client and at times maintain state between page transitions in the web analytic instrumentation.  How this is addressed is a much larger topic that is beyond the scope of this post.  However it is in this environment of static tags that we often must consider our tag plans.

When considering the number and configuration of variables and properties that make up a tag plan for a web analytic capability, one has to consider how these will all be brought together in the various reports to the customer.   The application or web developer is usually willing to provide fields of data that describe a page or context for an action, but it is the web analytic implementation that must stitch these various fields into meaningful tag variable assignments.

Now the Trouble Begins

For example, in the task web analytic capability defined above, clearly each task must be identified  by a task name that can be attributed to Task Start and Task Completion events.  For SiteCatalyst this means an eVar that can be attributed to Start and Completion success events assigned to the appropriate pages – start form and confirmation page.  Now we have different segments we need to attribute to these same events but also must relate to the task name.  Of course each task name will have their own list of scenario segments as well as task completion out comes.

Now the vulgarities of SiteCatalyst come into play.  I can make the Task Name eVar have full relations and then define eVars for each of the segments (or stitch everything together is Discover where all eVars have full relations).  Then from a Task Name report I can break down the results by any of the segments (eVars) I have defined. With this approach I quickly run out of eVars.  Even if I am ‘strategic’ in my eVar assignments such that multiple tasks can share the same segment, the eVar will typically cover only a portion of the visitors and introduces fragmentation and None as an Option for those segments.

A Way Out of Trouble

Fortunately SiteCatalyst provides an alternative solution – classification.  This is a great feature that exploits the Fifth Axiom of Web Analytics: “Data that is different is counted as different.”  In this case, the Task Name eVar includes not only the Task Name but everything we know about the user at the start of the task as well as at the completion of the task!

Using Omniture’s SAINT, these variables are keys that can be classified on the backend to represent all of the segments I have identified in my solution design. So instead of dozens of eVars I have one and now have reports that drill down from the task name.  With additional discipline all the tasks can be reported and broken down in the same report!

The trick here is to declare the eVar with each event (they are after all “event variables” meaning variables specific to the event).  Then classification ties the segments that include both all events.  For example, task_name | user_senarios is declared at task Start and task_name | user_senarios | task_outcomes declared at task completion. Then through classification task_name and user_senarios that are known at the start can be attributed to task Completion along with the task_outcomes.

Now Pile On

The implementation approach is straight forward with what I call Piling On. All the parameter values that one would normally assign to different variables in a static tag (or AEC) are concatenated together (with proper separators) into one value assigned to an eVar or sprop.  The key is to make the Key differentiate all the possible variations in segments you need.

Once you have defined the classification structure in SiteCatalyst (ADMIN > conversion > classification), you can retrieve the template and key values from SAINT in a tab delimited file; separate into your desired segments via Exel (via those cleaver separators you inserted at collection), and then upload back to SAINT. SAINT then classifies all of the data collected regressively (meaning all the data all the way back to the beginning) so you can change or modify segments through all report periods.

The result is that every visitor and visit gets assigned to proper segments, the segments are correctly correlated and None is Not an Option!!!  In fact, None in this case means keys (eVar values) that have not been classified.  So occasionally you need to retrieve the unclassified keys from SAINT (using the Null field option in the Export), rinse and repeat.

Summary

“None” is a telling symptom of a problem and “None is Not an Option” encapsulates an approach to diagnosing and resolving the problem.  When reports are designed such that None is Not Option and “None” appears in a report, it becomes an important tool for diagnosing the problem.  The resulting reports not only cover all user behaviors but with discipline allow the reports to address multiple questions and with extra diligence support multiple actions.

In the case of Omniture’s SiteCatalyst, the reports are available to a wider audience than what can be provided by Discover.  Though most serious analysts work exclusively within Discover to avoid the rather arbitrary limits of SiteCatalyst, by providing strong reports in SiteCatalyst the work performed in Discover is even further empowered.

First, fragmented segments that are likely loosely coupled are now tightly correlated in both environments, allowing drill-down in SiteCatalyst that can be further broken down in Discover. Second, insights discovered in Discover can be more directly related to SiteCatalyst reports, allowing business owners to more quickly grasp and exploit the insight.  Finally, the data mining is more effective once the data has been mind, because the data has the facts (keys) already rather than stitching the facts together from potentially disparate fields.  Hence, Mind Before You Mine.

More recently Oracle CEO Larry Ellison announced Oracle’s entry into the cloud-computing arena with Exalogic Elastic Cloud, that has been dubbed “Cloud in a Box”[i]. Ironically the Cloud is supposed to eliminate the Box[ii], so technically this is an oxymoron following current definitions and trends in Cloud Computing[iii]. For example, elastic is interrupted as being able to scale computing resources to demand by incorporating more virtualized servers in the cloud and hence only pay for resources consumed. A box on the other hand is a set configuration of resources that must be purchased (as a box) and remains idle as a box until peak demand utilizes its entire potential. So how can Oracle claim to have a cloud in a box?

Cloud from a Different Perspective

To answer this, we will need to expand our perspective and look at architecture from a different point of view, perhaps wave our hands at certain points. First, the information systems space that builds boxes, which includes IBM, Oracle, Teradata, and more recently HP, has typically differentiated their products into warehouses and appliances. The latter are data boxes configured to perform one task very fast and very efficiently, and once setup and integrated into a business process, the appliance does not change. This would include transaction engines, content management, customer relation databases, and most financial databases. (As I have pointed out before, one does not go mining for profits but follows well defined accounting procedures with set data structures).

These appliances have critical roles in the business processes but often the boundary between minding the data and mining the data can be very fuzzy and broad. Though you would not mine financial data for profits, you would mine the same data for insights affecting those profits. Though you might have a content management system supporting your authoring and serving your content, you may want to feedback information as to who is viewing the content and their reaction to that content, perhaps target specific content dynamically to segments of visitors as they are discovered in real time.

Normally all this financial, transactional and presentation data is pulled off to a data warehouse where it can be integrated, aggregated and analyzed requiring great flexibility in augmenting, querying and presenting the data. Instead of fixed static queries that are optimized for retrieval or storage in an appliance, the queries are free formed and changing within a data warehouse such that a different form of optimization must take place to search and aggregate large (terabyte to petabyte) sized data warehouses. In Oracle’s case the appliance is Exadata with fast storage and access and the warehouse is Exalogic with computational power to perform functions (logic) on the data. (To make things more confusing the latter is sometimes called a “data warehouse appliance” emphasizing that the whole thing is configured as box right out of the box so to speak, but we have that covered  with the term “box”.)

Boxing in the Cloud

But this is only part of the story. Until recently, these have been separate offerings with the appliances operating in real-time and data warehouses operating in the background, typically as batch processes, at their own “more strategic” tempo. However as the data from different sources becomes more inter-related, such as forming social networks among thousands of users every minute, and requires near real-time warehouse-like processing in response, such as immediately using the new networks for post messages. The warehouse is no longer a very large database but more a massive parallel computing system that must process and access large amounts of inter linked data in real-time.

Exalogic is Oracle’s attempt at massive parallel computing (MPC) in a box configuring up to 360 processing cores[iv] to operate on a computational problem. The claim by Ellison is that as a server the box can process 1 million requests per second sustained and that two of these boxes would be sufficient to service all Facebook customers. However, there are measures more important that web hits to be considered and Facebook’s requirements illustrate this. Essentially Oracle’s Exalogic is 360 web servers built on top of an Oracle Grid RDBM. The same processors managing requests are also performing the database queries. Is there a problem with this?

Sequential Processing Giant Goes Concurrent

Oracle has practically no experience in parallel computing as it applies to database design and implementation. Oracle is after all the archetype for communicating sequential processing where grid means communications among databases where everything is shared and potentially locked down.

Interestingly their new CEO, Mark Hurd (formally ousted CEO at HP)[v], does have parallel computing exposure through Teradata as CEO of NCR. So this is an interesting twist that has not been sufficiently exposed over the HP / Oracle brouhaha. I would consider this a clue that parallel computing is at the core of the offering and not simply a box with 360 virtualized servers. Teradata on the other hand has over 30 years corporate experience applying parallel processing to computation and 20 years developing parallelized database systems based upon Share Nothing Architectures[vi] (the antithesis of what Oracle has been developing over that same period).

To me this represents a subtle but significant core shift in computing whether cloud or not. Oracle and IBM have deep pockets to acquire the expertise that Teradata has developed over the long haul. However, I would suggest that Oracle go outside of Silicon Valley because the concept of concurrent programming is counter to the current culture there. In the end, it is not the hardware but the operating system and the “programming” paradigm that can exploit the potential of the hardware that will be the breakthrough and provide the change needed to succeed.

Elastic Box: A Cloud by a Different Name

Returning to the original question with respect to elastic. Though Oracle’s offering is a box, elasticity within the box will depend on how well demand from various sources can be managed to allow flexible allocation of resources within that box. Typically architectures assume that each box is dedicated to a single function – web server, transaction engine, financial database, etc. Can all these functions be performed within the same box such that resources can quickly roll over to the function having the most demand at that moment and therefore ensuring that the box is effectively utilized while satisfying all demand? This assumes that the combined demand on all the individual boxes can be smoothed when combined into a single box.

This is the overall assumption within an elastic cloud but demand is spread over a larger number of functions and wider global demand. The flex point for cost of computing from external to internal cloud is the point where demand is well defined and smooth and the cost of owning and maintaining the equipment is much less than the utility cost of resources. In other words, when elasticity is not an over-riding factor.

In Oracles case and for others in this space, removing the need to allocate function to separate boxes and providing more flexibility in reconfiguring or expanding functions can lead to further cost savings and resilience over the traditional IT architectures. In this way the box acts as an internal cloud and can be argued to be elastic, but only if you change the currently accepted definition[vii].

One Cloud of Many Boxes

For the cloud purest that believes that the cloud is a monolithic computer, these kinds of boxes will increasingly populate the network to achieve that vision. It will only occur when the programming paradigm allows “resources” to follow “apps” much as workers follow jobs within companies. Currently we see apps allocated to resources and programmed specifically for those resources. So the shift is big – a shift that I doubt Oracle has not even started to make.

Works Cited

Abel, J. C. (2010, 7-September). Mark Hurd Lands At Oracle, HP Sues. From Wired: http://www.wired.com/epicenter/2010/09/former-hp-ceo-mark-hurd-lands-on-his-feet-at-oracle/

Beal, B. (2010, 20-September). Oracle Exalogic is the company’s “cloud in a box”. From SearchOracle: http://searchoracle.techtarget.com/news/2240022616/Oracle-Exalogic-is-the-companys-cloud-in-a-box

Bhatia, A. (2007, 10-December). Shared Nothing Architecture. From Toolbox.com: http://it.toolbox.com/wiki/index.php/Shared_Nothing_Architecture

Closson, K. (2007, 9-August). Nearly Free or Not, GridSQL for EnterpriseDB is Simply Better Than Real Application Clusters. It is Shared-Nothing Architecture After All! . From Kevin Closson’s Oracle Blog: http://kevinclosson.wordpress.com/2007/08/09/nearly-free-or-not-gridsql-for-enterprisedb-is-simply-better-than-real-application-clusters-it-is-shared-nothing-architecture-after-all/

Considine, J. (2010, 22-September). “Cloud in a Box” Gets Boxed In . From Cloud Switch: http://www.cloudswitch.com/page/cloud-in-a-box-gets-boxed-in

Dignan, L. (2010, 20-September). Oracle’s Exalogic box: Cloud washing at its best? From ZDNet: http://www.zdnet.com/blog/btl/oracles-exalogic-box-cloud-washing-at-its-best/39343

Snow, R. (2010, 5-October). Teradata Advances Personal Marketing Using Analytics. From Ventena Research: http://www.ventanaresearch.com/blog/commentblog.aspx?id=3984

Williams, A. (2010, 30-September). Why the Oracle Exalogic Cloud is Not Elastic . From ReadWrite Cloud: http://www.readwriteweb.com/cloud/2010/09/why-the-oracle-exalogic-cloud.php

OK – So its not quite breaking news. It took awhile to recover from the shock. For the moment just play along.

Google announced recently that they are suspending further development of Google Wave[i].

“Was it a social network or a collaboration tool on steroids? Nobody could quite agree, and it’s one reason for the demise of Google Wave.” Georgina Swan, PCWorld[ii]

“… admitting failure and moving on is another key lesson in managing innovation.” — Karim R. Lakhani, Harvard Business Review[iii]

“Still, I think Google’s public celebration of its recent product failures strongly suggests that its growth story is far from over.” John Keeling, The Motley Fool[iv]

I can hear the Brain going through his checklist with Pinky:

• Real time interaction among networked individuals – Check!
• Uber mash up and convergence of different social interaction metaphors – Check!
• Uber sharing and collaboration allowing networks formed in Web 2.0 to accomplish objectives in Web 3.0 – Check!
• A platform that can be expanded by third parties to provide capabilities that has yet to be envisioned – Check!
• A really really really smart spell checker[v] – Check?!
• All wrapped in an obsolete desktop metaphor with a UI that is so ugly and complex that only a geek could love it[vi] – Check! What??? Narf!!!

Geeks: A mass media they make – Not!

Anyone of these points could become the next media gestalt that changes how we operate in the world at large. However these capabilities, though requiring more computational capability, are more incremental to what has already been accomplished in Web 2.0.

Collaboration already can be done with Webex and combining email with documents is not that different from Google Groups. In the end the mashup was too complex for anyone but a techno geek to comprehend. To bring social networks into email and messaging has already been done by Facebook and MySpace with Twitter and Tumblr addressing the need to express and publish thoughts and ideas.

To return to a desktop metaphor of work groups and browsers just doesn’t seem to fit the persona of a revolution. A better spell checker and watching someone correct your text while you are typing is not sufficient change and in some ways creepy.

It is the masses of social beings (which excludes many of us geeks) accepting and adopting a technology that creates a medium. Make no doubt about it, you have to become a medium phenom if your goal is to take over the world. You can’t be like Microsoft, just coming out with new versions of the old stuff, or in Googles case tweak the search algorithm. That is the direct road to has been.

The untapped potential is the networks created in Web 2.0 can be used to accomplish things…together…in Web 3.0! Working together is not a new thing, but potentially disruptive if facilitated and accomplished on a massive scale. Unfortunately, Google is not able to exploit this potential, whereas Facebook and Twitter have been able to at least capture the public’s imagination, and Apple has been able to focus on its demands.

Google must Face the Book (and the Music)

If Google is the Architect of Cloud 2.0, then Facebook is its face. Google after all this appears as clueless as the Architect in the Matrix (don’t even consider Facebook as the Oracle … maybe Apple or Salesforce). Poor little Google caught between Facebook and Apple looks as out of place and lumbering as Microsoft and Yahoo at a search conference. Somehow Google has to defend search and get Social as Microsoft had to defend the desktop and get the Cloud. From Microsoft’s experience, it does not look good for Google nor the Silicon Valley techno-geeks that have been cheering on Google’s technological arrogance[vii]

In fact there is an ironic twist to our Pinky and the Brain metaphor. Facebook and Twitter, the scorned of the Valley Wag, are viewed as Pinky against Google as the Brain. Though he plays the part of the simpleton, it is Pinky that has the sardonic insights that cut through to the core of the Brain’s maniacal self-absorption. Will all the brain power at Google be able to work through it’s own short sightedness? The indications are no – but that’s another episode.

Don’t Count the Brain Out Just Yet

To be fair, the Brain is tenacious both in cartoon and corporate forms. Google still has the computational power and experience that Microsoft and Yahoo are only now attempting to match. The social media companies – Facebook, Twitter, Diggs – are preoccupied with meeting demand, with Facebook able to expand application of its network to different services. But can it all be done in blue with not even a glance to user experience.

To many pundits, from the Wave experience, Google has acquired capabilities that will eventually be spread to other offerings including a social offering. So perhaps let Google regroup and keep an eye on them for another breakaway plan. After all, Android is number one iDevice OS and Chrome is taking share from Microsoft IE. That’s success. Right?

I wonder what others are doing in the meantime?

Just launched on Project Hosting at Google the loafwithjam project.  LOAF stands for Live Object Application Framework and supports the Actor Model of Concurrency introduced in my last post.  I have upload the documentation for codes that are similar to what have been in operation for web analytics collection and processing successfully for many years. The intent is to initiate a Live Object in a context server when a visitor first arrives at a site and process the events of that visitor in real time.  Hence the name: Live Objects.

This is an educational project for using the Actor Model of Concurrency in implementing Java Web Services that are linearly scalable over multiple servers and processors. As I described before: This is Hadoop on steroids, Map Reduce taken to Nth degree using Share Nothing Architecture approach.

Recently, my friend Mark (from A Day Being Mark) pointed out to me that Actors have been making a big splash in the Lisp world again with the actors library for Closjure. Original research in actors was implemented in Lisp, so LOAF is Actors without the Lisp.  After what demand is there for Lisping Thespians?

The java documentation can be download from the project. I suggest starting starting your reading with the LOAF package document, which will start your introduction to Live Objects, Actors and ACID.  More on actors after I complete the posting of the series on Cloud Computing. Enjoy!

We have seen in a previous post[i] how Map Reduce is analogous to how business and product managers plan and implement projects with Gantt Charts allowing a number of tasks to proceed independently in parallel and integrating their products at the end. A question from those outside of computer science might be: why has it taken this long to figure this out? In fact, if computers are so fast and capable, why don’t they figure the quickest, most parallel way of executing multiple tasks all by themselves?

The answer represents a fascinating history of personalities, ambitions, and feuds that make up what we call computer science. First, computer science is not like any other science in that it is also an engineering discipline that often gets ahead of the science, and a business segment that often gets ahead of the engineering. As a result there is no repository of long-term memory so concepts often have to be rediscovered as the necessity of new situations arise or race to catch up with the hype. A rose by any other name would be considered something entirely different!

What has changed now is the general recognition that having thousands if not 10s of thousands of computers applied to a problem (such as search) can have significant advantages over a super-computer though the latter is tightly coupled, efficient, and designed to be super!

The web discovers Sullivan’s Law[ii], which was defined in the mid 70′s! At the same time, a group at MIT under Hewitt[iii] were developing and evangelizing the concepts of CCP derived from the Actor Model of Concurrency. Also work at Carnegie Mellon by Bennett[iv] involved development of ACID and distributed operating systems. All this was done at the same time that Dyjkstra[v] and Hoare[vi] were developing CSP. At the time, there was a great deal of exchange among these groups and the citations provided above give a flavor of the exchange at the time. In the early 80s, CSP was selected as relevant to distributed system problems at that time.

I have had the privilege of working with Sullivan, Hewitt and Bennett at different times in my career as an engineer[vii]. It has puzzled me why their work and contributions have not been more prevalent in our work today. Until recently one could argue the best approach was to build efficient algorithms and rely on Moore’s Law to double processing throughput every 2 years from more transistors on faster chips[viii]. Has anyone noticed that we have been holding at around 3 GHz for 5 years? The increased capacity has come from building multiple processors called cores on a chip. But then applications have to be able exploit parallel execution at the individual computer level. Not so difficult when there a few processors, there always a loose thread around looking for a processor, but with hundreds or thousands, one needs to stop and think.

More recently the computational capacity increase has come from distributing processors into everything from automobiles to cell phones, and interconnecting these via the Internet[ix]. Can there be anyway that this global network of processors can be harnessed to do useful work? To take the next leap in processing capacity requires harnessing distributed computing, and Map Reduce is just a gross (coarse grained) application of this concept.

Concurrency in a Distributed World

In Star Trek Next Generation (I am talking with fellow nerds here, right?), we have the Borg – a collection of assimilated races attached via cyborg appliances to a central controller that is actually the composite of all the Borg minds. All can access anything experienced by one. The Borg operate without individual identity and with relentless efficiency to absorb new races. In their organization and behavior the Borg represent one vision of distributed computing[x].

We humans, enemies of the Borg that to the Borg must be absorbed because we are so inefficient, represent another form of distributed computing[xi]. In comparison to the Borg, we seem highly inefficient, requiring a great deal of support and because we are isolated and individual, unnecessary coddling and communication to get anything done. This nicely lays out the differences and contrasts in positions (including hostilities) of the different approaches to concurrent programming.

There are the Borgist, who insists on tightly coupled processing and who build MIMD and SIMD processors with shared memory and clock synced instruction execution to squeeze every concurrent advantage with mechanistic efficiency. Then there are us humans that attempt to use tools such as computers as extensions of ourselves to be more effective or productive, and despite our individuality and independent behaviors seem to at be able to organize to perform amazing feats, such as sending someone to the moon, sequencing the human genome, building the Internet or manufacturing tooth picks. What is the essential difference?

The difference is that in the Borg model everyone shares absolute knowledge of the global state and in the human model no one at any time has absolute knowledge of the global state. In the later case everyone works with the assumption of sufficient knowledge of relevant state. This is precisely how concurrency is applied to developing distributed applications. The concept of global shared state is eliminated!

This represents one derivation of the concept of no shared state from CCP Theory. An independent derivation of the same concept (I told you that concepts have to be rediscovered) arises from attempting to unlock the shared constraints in data management systems called Share Nothing Architectures in the 90s[xii]. Yet another line of investigation comes from systems theory and economics called bounded rationality[xiii]. The latter makes clear that concurrent programming is clearly more a systems discipline than a programming discipline, which will become more clear as we continue.

Relativity in a Closed Mechanistic World View

The concept of no global shared state is gut wrenching to many in computer science on a number of different levels and for a number of different reasons. There are many consequences that result from this rule. For one, the concept of global time and ordered sequence of events has to be abandoned. Just as the transition from Newtonian to Relativistic physics absolute time is eliminated, the same is true in the transition from CSP to CCP.

Without absolute time, events in the network cannot be sequenced in absolute order everywhere in the network (unless one waits until all data is in, which can be itself an indeterminate time). But in the meantime, decisions and other events are being generated based upon incomplete knowledge of the global state. As a result, the sequence of events that are “observed” by a process is probabilistic and not guaranteed repeatable.

An example is vote counting. TV networks provide exit polling data based upon sampling; followed later with raw vote precinct counts as they are reported; and followed many weeks later with the certified recorded counts. Each count is different, but the critical threshold of who got the most votes many times can be determined from the exit poll sampling and partial precinct reported counts.

Is absolute certainty of the actual count relevant or the fact the threshold has been passed with high certainty the important fact? This is a question that is often asked in designing decision systems. For example, in designing a distributed operating system, is it important that all processes know exactly the storage available or whether a critical threshold has been crossed such that access must be serialized and tightly coordinated.

In the rare cases when results are too close that a recount is necessary such as the ones in Florida and Minnesota, one becomes exposed to the messy world of real world data with hanging chads and double entries. Ambiguity and noise in voting results is difficult to comprehend much less accept when one believes in a mechanistic closed universe where every vote is counted and the process ensures the same count every time. Whenever the ambiguities of human behavior and the verities of infrastructure are introduced, the system is never closed and the processes must take these variations into consideration. This is fundamental to how all science is performed.

In the real world, for 99.999% of the cases, the exit poll and raw count numbers are sufficient to determine the winner and proceed with high confidence from that result. Both in the real world and in concurrent programming one can insist upon an event being executed simultaneously everywhere; actions preformed only when events are properly sequenced to generate repeatable results; or respond to events in the shortest time possible at a given confidence level. Just not all at the same time.

Massive Distributed Computing in an Open Internet

For massive distributed computing to work over a global open system such as the Internet, one has to be very selective on when to insist upon global understanding of time or any other state parameter if one wants to leverage concurrency to increase computational throughput. To insist on Borg-like omniscience is neither practical nor desirable.

The result is a computer system that is programmed much as businesses are organized. Though businesses seem highly inefficient as a computational model, they do excel at making sure that all their employees (processors) are busy, and organizing them to work productively in parallel. Small start-ups of a few workers can get more done and quicker than one individual alone. Large organizations may seem to not have linear scalability but none-the-less can perform gargantuan enterprises not conceivable for an individual.

There is actually something more going on here than mere programming. The small linear programs, chubby objects, are embedded into a system that can respond to the external environment and optimize to higher-level goals. Enterprises incorporate systems and subsystems that interact at different levels of communication and coordination allowing complex processes to be distributed to simpler processes operating in parallel. The ability to adapt and optimize is a feature of self-organization that can occur when concurrent components communicate forming feedback loops. In computer science these “programs” are referred to as connectionist networks that are an out growth from artificial neural networks and require developing concurrent operating environments over a wide spectrum of couplings in both communications and shared state[xiv].

All this is accomplished by not insisting upon everyone including the president having a global view of the companies state, nor serializing access to shared assets. How this would be accomplished in programming a massively distributive system requires a different programming paradigm. One that identifies not only sequences of processing steps, but also the concurrency of processes and the dependencies of these processes to the completion of other tasks, much as how businesses define work break down structures.

Resilience Comes To The ForeFront Over Predictability

The strongest counter argument by the closed systems advocates is why develop programs that can provide only probabilistic or worst unpredictable results. The thing is we have already developed such programs without being aware we have crossed the line into open systems programming.

The current rage is data driven decisions where managers make decisions based upon analytic data and business intelligence concerning their business. The data is assumed to follow all the data integrity and consistency policies of a “global” view of the business. Companies that insist upon the highest quality data following strict accounting or for online SOX compliance place human decision makers into the process that are neither strict nor consistent. In so doing the linear programs are now part of a non-linear if not unpredictable decision process, especially if the data collected and analyzed reflects the impact of decisions made by those managers.

The fact that humans are making the decisions does not make the enterprise more stable. Until one considers the entire combination of human and automated subsystems as one system, one will not be able to access resilience of the enterprise to competition nor it’s potential for growth. From that system perspective one can see if adding automated decision processes will aid or hinder the performance of the business. At that point one has introduced feedback loops and crossed the line into open system unpredictability.

In place of predictability, one wants to know the resilience of the system to respond to unexpected events. Programs designed assuming a closed system with predictable inputs do not have much resilience when connected to an open environment. Would it not be better to build systems with an understanding of its behavior in real environment rather than expect predictable results in a fantasy environment? It is not a trick question! You would be amazed on the number that insist on the later.

Google makes it’s Move

What is this paradigm is TBD[xv]. However at this point Python can be considered as viable a candidate as C++, C#, or Objective C or even Java or Ruby. The critical question will be how finely concurrent units can be identified and how quickly these units can be integrated into systems that perform useful work. At the moment, most are considering computational units on the scale of individual servers[xvi]. However for Google, the computational units are Objects held as chubby chunks in the Big Table[xvii]. So Google has already broken away from the cloud-computing pylon, huddled around conventional notions of computation.

What can be done that will be New?

So what could be done with thousands potentially millions of servers hosting millions potentially billions of chubby worker bees just waiting to be organized to perform your every whim? Google has the servers and the engine to support chubby worker bees and has experience in some of the ways of organizing these worker bees over a number of applications including Google Analytics, Maps, Finance, and 60 other products[xviii]. Is this sufficient to propel a third wave of Web change? Well Google had a thought about that – NOT!

Works Cited

Agha, G., & Hewitt, C. (1985). Concurrent programming using actors: Exploiting large-scale parallelism . From SpringerLink: http://www.springerlink.com/content/y1wl7q3342006720/

Banerjee, U. (2009, 26-November). Cloud Computing Service: Amazon EC2 vs Google GAE. From Cloud Computing Journal: http://cloudcomputing.sys-con.com/node/1200706

Bhatia, A. (2007, 10-December). Shared Nothing Architecture. From Toolbox.com: http://it.toolbox.com/wiki/index.php/Shared_Nothing_Architecture

Boyer, R., Chen, D., Gray, A., & Lee, T. (1999, 30-November). Operating Systems for the WWW. From Google Docs: http://www.bennetyee.org/ucsd-pages/Courses/cse221.f99/OSSurveyF99/papers/boyer%2Cchen%2Cgray%2Clee.operating_systems_for_the_www.pdf

Business Dictionary. (n.d.). Bounded Rationality. From Business Dictionary: http://www.businessdictionary.com/definition/bounded-rationality.html

Closson, K. (2007, 9-August). Nearly Free or Not, GridSQL for EnterpriseDB is Simply Better Than Real Application Clusters. It is Shared-Nothing Architecture After All! . From Kevin Closson’s Oracle Blog: http://kevinclosson.wordpress.com/2007/08/09/nearly-free-or-not-gridsql-for-enterprisedb-is-simply-better-than-real-application-clusters-it-is-shared-nothing-architecture-after-all/

Dijkstra, E. W. (1975). Guarded commands, nondeterminacy and formal derivation of programs. From ACM: http://doi.acm.org/10.1145/360933.360975

Fay Chang, J. D. (2006, November). Bigtable: A Distributed Storage System for Structured Data. From Google Research Publications: http://labs.google.com/papers/bigtable.html

Hewitt, C., & Baker, H. (1977, 10-May). Laws for Communicating Parallel Processes. From MIT: http://mit.dspace.org/bitstream/handle/1721.1/41962/AI_WP_134A.pdf?sequence=1

Hoare, C. (1978, August). Communicating sequential processes. From ACM: http://portal.acm.org/citation.cfm?doid=359576.359585

Knorr, E., & Gruman, G. (2009, 5-May). What cloud computing really means. From Info World, Cloud Computing: http://www.infoworld.com/d/cloud-computing/what-cloud-computing-really-means-031

Kraft, T. (2010, 02-July). Clouds of Hadoop: How Map Reduce Changed the World . From Mind Before You Mine: http://mindbeforeyoumine.com/2010/07/02/clouds-of-hadoop/

Kraft, T. (2010, 25-June). Send In the Clouds: Old Metaphor Gets New Life. From Mind Before You Mine: http://mindbeforeyoumine.com/2010/06/25/send-in-the-clouds/

Kraft, T. T. (1989). Summary Proposal for the Development of an Architecture Neutral Distribution Format based upon Communicating Concurrent Processes. Neurocomputing , 1(2), 23-28.

Meadows, D. H. (2008). Thinking in Systems – A Primer – . (D. Wright, Ed.) River Junction, Vermont: Chelsea Green Publishing.

Moore, G. (2010). Moore’s Law: Made real by Intel innovation. From Intel: http://www.intel.com/technology/mooreslaw/

Paolo Fogliata, M. T. (2008, 18-August). Intelligence-ready network infrastructure: An ecosystem to control third-party intelligence distribution close to nomadic users. From Wiley Online Library: http://onlinelibrary.wiley.com/doi/10.1002/bltj.20306/abstract

Patton, P. C., & Hersh, C. I. (2003, April). Asynchronous Non-Preemptive Tasking for Parallel and Distributed Systems. From A Quantitative Methods/Computer Science White Paper: http://140.209.112.90/WhitePapers/QMCS/ANT.html

Ricadela, A. (2007, 16-November). Computing Heads for the Clouds. From Bloomberg Businessweek: http://www.businessweek.com/technology/content/nov2007/tc20071116_379585.htm

Sullivan, H., & Bashkow, T. R. (1977). A large scale, homogeneous, fully distributed parallel machine, I. From ACM: http://doi.acm.org/10.1145/800255.810659

Swoyer, S. (2008, 08-July). Is Cloud Computing the Next Disruptive Technology? From ESJ: http://esj.com/articles/2008/07/08/is-cloud-computing-the-next-disruptive-technology.aspx

World Domination Plan 137:
Phase 3: Distract before Conquering

While everyone is building out their Hadoop farms[i] and adopting Map Reduce to establish elastic computing[ii], Google is busy building (and patenting) some strange new technologies with names from a Dick Tracy line up – Gears, Big Table and Chubby Locks. Also it is busily building the biggest mash up heretofore envisioned called Google Wave allowing individuals across the web to message and collaborate in real time … at a keystroke level[iii]!

With Map Reduce, Google was able to process the entire content of the web[iv]. However, map reduce is a batch technology that gains concurrency advantage by mapping resources to perform dedicated parallel tasks on batches of data[v]. In parallel computing parlance one might call this Same Task Multiple Data (STMD), where not only is the same task performed on multiple computers, but within a computer the task processes multiple instances (rows) of data in batch. The dual architecture of Multiple Tasks Same Data (MTSD) is also conceivable with map reduce but a more rare parallel computing strategy[vi].

So what does it take to move beyond this batch non-real-time limitation and be able to process any task, any time, anywhere at massive demand scale? The answer is Massive Distributive Computing (MDC). This requires the transition from Communicating Sequential Processes (CSP)[vii] to Communicating Concurrent Processes (CCP)[viii]. Don’t let the big words bother you. They will be all explained in a moment, but first what is concurrency.

Sequence in a Concurrent World

Remember in a previous post[ix] we likened demands on storage to that of a cash register in a super market. By separating storage into the equivalent of aisles, we can get concurrent processing advantage to reduce queues of customers, the analog to processes accessing storage. The assumption is that any of the cashiers with their registers can service any customer. Once the customer is at the register, performing checkout, the cashier should complete the checkout, without interruption, before moving on to the next customer.

The technical term is that checkout is an atomic transaction[x]. If the checkout was interrupted so that the cashier had to checkout a more important customer (called pre-emption), besides angering the pre-empted customer, the cashier must rollback to the state prior to checkout, as though the customer had never started the process (rollback on pre-emption). So when the peeved customer leaves the store in a tizzy, neither he nor the store has lost any money in the transaction. A similar thing would happen if at the end, the customer discovered she had no money to complete the transaction (rollback on exception).

Though pre-emption and rollback are relatively rare in a supermarket, these are frequent occurrences in processing data where there is large number of processes sharing resources such as storage or state information such as current storage capacity. To handle these cases, each process must receive a lock, similar to taking a number at the deli to get service[xi]. The customer is free to do something else but usually just waits to be called.

Waiting on the Internet Highway

So our first clue that Google is moving to massive distributive computing is Chubby Access Locks[xii]. Chubby is a mechanism that allows a process anywhere in the world to obtain a lock (take a number) on a resource anywhere else in the network[xiii].

In a typical database management system the shared data elements can be very fine grained (down to individual data elements). For a distributed system, the overhead of obtaining locks on individual elements is not practical, so blocks of data are more coarse grained or chunkier, hence locks for sequencing access to chubby chunks of data.

Not the obese chunks represented in a batch file processed by Map Reduce, but pleasantly zaftig such as perhaps an individual row of data in a bulk file. Furthermore we don’t have to wait for an accumulation of chubby chunks to make up one obese batch to get concurrency advantage in increase throughput. We can process the data chunks as they arrive – asynchronously!

The discussion so far is typical of the concerns covered under Communicating Sequential Processes (CSP). The communication comes from the mechanism the notifies a process when it can safely proceed in it’s access of a shared resource and sequential (or serialized) access to the resource is the consequence of the communication. This also defines the communications between processes. The process rendezvous with the resource process and communicate back and forth (called a protocol) until the initiating process is done with the resource and relinquishes the lock.

Though almost all books on concurrent programming begin with CSP, the result is not parallel execution but serial execution of tasks that share a common resource. As we found out, sequential processing reduces the concurrency advantage over performing processes in parallel without locks. To reduce sequencing and increase concurrent leverage and likewise increase throughput performance, one needs to reduce the need for locks as well as the occurrences of pre-emption and rollback. This is covered in the discipline of Communicating Concurrent Processes (CCP) that is based upon a different model of concurrency[xiv].

Rest Stops on the Internet Highway

Relational database systems such as Oracle or MySQL are archetypes of CSP systems where even with grid implementations they can quickly extinguish any concurrency advantage. These implement the Relational Model for data independence[xv]. Their objective is to break chunks of data into it’s basic elements such that a value is stored in one place only and referenced in all other places through unique identifiers called reference keys or indices. In this way a change in a data value propagates consistently to all references of that value.

From a concurrent processing view this is analogous to running a taxi cab company, where every time a taxi comes in, it is broken down into it’s component parts and stored in racks (tables). Then when a cab is sent out, it must be reassembled from the parts available in the racks. In most cases, this is not an attractive business model[xvi]. It is tempting to apply this model to software or data, where for example, newer parts can be swapped in or same parts used for different purposes (constructing more buses than taxis) to meet demand. But even in these cases it not appropriate to insist upon the objects (the chunks of data constructed from data elements) to be broken down into parts for storage each time. Especially if the object is just resting in cache, ready to be called upon soon to perform more work. Another strategy has to be considered.

So the second clue that Google has developed an MDC system is Big Table[xvii]. Big Table deals with chubby chunks of structured data that can be anything (but typically represents an object state) held in a flat table distributed similar to files in GFS[xviii]. It is a concurrency enabler in CCP[xix]. At least that is the potential. To realize the potential requires an entirely different method of defining and developing apps. The company that gets this right rules the world!

Works Cited

Agha, G., & Hewitt, C. (1985). Concurrent programming using actors: Exploiting large-scale parallelism . From SpringerLink: http://www.springerlink.com/content/y1wl7q3342006720/

Announcing Amazon Elastic MapReduce . (2009, 2-April). From Amazon News: http://aws.amazon.com/about-aws/whats-new/2009/04/02/announcing-amazon-elastic-mapreduce/

Burrows, M. (2006, 7-November). The Chubby Lock Service for Loosely-Coupled Distributed Systems. From Google Research Publications: http://labs.google.com/papers/chubby.html

Collmeyer, A. J. (1972). Implications of data independence on the architecture of database management systems. From ACM: http://doi.acm.org/10.1145/800295.811495

Dean, J., & Ghemawat, S. (2004, 15-December). MapReduce: Simplified Data Processing on Large Clusters. From Google Research Publications: http://labs.google.com/papers/mapreduce.html

Fay Chang, J. D. (2006, November). Bigtable: A Distributed Storage System for Structured Data. From Google Research Publications: http://labs.google.com/papers/bigtable.html

Hewitt, C., & Baker, H. (1977, 10-May). Laws for Communicating Parallel Processes. From MIT: http://mit.dspace.org/bitstream/handle/1721.1/41962/AI_WP_134A.pdf?sequence=1

Hoare, C. (1978, August). Communicating sequential processes. From ACM: http://portal.acm.org/citation.cfm?doid=359576.359585

Kraft, T. (2010, 02-July). Clouds of Hadoop: How Map Reduce Changed the World . From Mind Before You Mine: http://mindbeforeyoumine.com/2010/07/02/clouds-of-hadoop/

Kraft, T. (2010, 25-June). Send In the Clouds: Old Metaphor Gets New Life. From Mind Before You Mine: http://mindbeforeyoumine.com/2010/06/25/send-in-the-clouds/

Sullivan, H., & Bashkow, T. R. (1977). A large scale, homogeneous, fully distributed parallel machine, I. From ACM: http://doi.acm.org/10.1145/800255.810659

Wikipedia. (n.d.). Paxos Algorithm. From Wikipedia: http://en.wikipedia.org/wiki/Paxos_algorithm

Yahoo! Advances Hadoop From Science to the World’s Largest Internet Deployment to Mainstream . (2010, 29-June). From Yahoo! News: http://labs.yahoo.com/news/426

From my blog analytics, I can actually see the eyes rolling as my audience quickly skirts my Cloud Computing posts to go to the old stuff (back when I was funny). In some ways I understand the ambivalence. I perhaps like you are not overtaken by awe with virtualization and utility computing that defines cloud computing today. I would call this more cloud hosting than computing. My urge is to go beyond this in an attempt to grasp the computing part of the cloud. Understand what new capabilities may be facilitated and computations performed. So why should you care?

I know. Why be concerned with the future when we barely understand the present (from an analytics perspective). What happens to our data as more customers go mobile? How are we suppose to deal with social media and mashup tools such as Kickin or Cooliris or Yoono, which was the topic of my first blog post “How Web Analytics Sees Reality”? How do we combine brick and mortar marketing with on-line marketing? Is it all new or the old repackaged with a new name? Is Viral Marketing just another name for publicity stunts?

My natural instinct when I seem to be bobbing up and down like a cork on an ocean in a hurricane of change is to attempt to see the larger picture in play. What is turbulence (noise) and what is a meaningful trend? I found out recently (from reading Donella Meadows’ “Thinking in Systems – A Primer -”) that this is an aspect of “Thinking in Systems”, looking at events as part of larger system processes with longer time-lines. As we move from reactive analytics (what happened and why?) to predictive analytics (what will happen and how much?), this is an instinct and discipline that should be nurtured by all in analytics.

But returning to Cloud Computing and why we as analysts should care and have some understanding, I will give three reasons. As you may be aware, I am not big with posts that give lists of simple answers that can be quickly absorbed and just as easily forgotten. In this case will make exception and try not to spin the topic into simplistic babble.

1. IT is Part of Our Job Description

Someday if not already you will find yourself across the table with someone in IT (at Yahoo! the particularly stubborn ones were aptly called Paranoids) whose job description is to maintain the status quo while your objective will be to change the quo. He will have no reason or motivation to understand what you are trying to do, so it will be up to you be the bridge from your problem into what IT is concerned with.

You will not necessarily have to write the code or even design the solution. But you will need to cross the bridge and present options that IT would be interested in understanding and following up with action. Cloud Computing and its concern with scalability, availability along with elasticity and utility presents a lot of options that must be understood by all parties. This includes us data wranglers, since data movement and processing are significant aspects effecting these options. Leading us to reason two:

2. Analytics is typically the First User of Cloud Computing

Some might argue pornography as the first user but all the clicks to all those sites had to be counted by a much larger entity. Web analytics has always had to scale larger than its largest customers. EBay is only a portion of Omniture’s traffic. They don’t have to serve all the content but they do need to record all the transactions.

Having started in this business at the beginning of the switch to Web Analytics as a Service, I have had first hand experience in meeting collection and processing demand. This meant implementing map reduce before the term was even invented. In my case, we developed appservers based upon the actor model of concurrency – Map Reduce on steroids. It was the same struggle for all the other analytic companies. If you want to go really deep, I have published the documentation for the system I developed called LOAF with JAM.

Even the large companies I worked for (Yahoo, Overture, EBay), processing and warehousing data is a central aspect of their business model. This is one of the reasons that Yahoo! has gone “big” on Hadoop server farms having lashed together some of the largest outside of Google.

Even Google Analytics takes a small niche analytics company and hosts them in Google’s cloud to become the commodity provider of web analytics to most of the world. This has not abated! I recently meet with a startup with relatively few customers (couple 100) but yet has collected several petabytes of data that must be mined and reported. All these are possible through the evolving discipline of distributed (i.e. cloud) computing.

Now most of us can have a good career simply trying to get IT to put Web Analytic Provider tags on our sites. At some point we may be wildly successful and actually get the business to look at the data and base decisions upon the insights derived from that data. At which time you will be cubby holed with IT to get new capabilities on line.

At some further point, the business is going to out grow the standard offerings and attempt to develop in-house analytic capabilities and tools. This most likely will also become the business’s introduction to cloud computing platforms such as Hadoop or data warehouses such as Teradata or Interwoven or Customer Experience Management such as Tealeaf. As the web analytics person you will be in the middle of this, maybe even the cause of it.

3. Cloud Computing is the Reality We Are Attempting to Measure

In McLuhan-esque terms – Cloud Computing is the Medium and as we all know, the Medium is the Message. Who knew? You might have thought that Google Search was the message, or Facebook or Twitter or Foursquare or Wrrl or maybe Farmville. These are but ephemeral manifestations enabled by the underlying medium. Ultimately it is the medium that changes us and not specific babbling that crosses the medium.

The message is not Justin Bieber but the fact that an entity that happens to be Justin Bieber can consume 3% of all Twitter’s infrastructure and that with all the freedoms that we can exercise, we as a mass can coalesce around an entity such as Justin Bieber.

Justin Bieber is the Britney Spears of Twitter, who was the Madonna of Search, who was the Cher of MTV, who was a hippie.

The cloud has manifested itself in Web 2.0 first as search and then as social networking and micro-blogging. It is now mobile and in the future will become something else. It will be advancements in computing more and faster that will drive the medium to other manifestations in Web 3.0.

Is Cloud Hosting sufficient to change our reality by enabling all the X as a Service providers to finally have a viable business model or does the Cloud have to actually do work? It appears that an entire economy is growing around apps empowered by the cloud. What is the potential of these AppStores harnessing cloud computing to shape our world? These are questions I am trying to understand.

The reason why I believe it is important to understand is that our first duty in analytics is to understand the reality we are trying to measure and then determine how we track and measure that reality. Advancements in cloud computing will be the enabler of all the realities we will be tracking in the future. Perhaps this is a more fundamental perspective than most would want to take, but I hope that my contributions will reduce the effort for my readers.

What is Coming!

It has been an interesting journey, which started in mid summer. I thought I saw a narrative arc for researching and organizing the material for understanding cloud computing, but then events intervened and converted it into more of a corkscrew.

It started with Google’s dropping Wave in August, which to me was the most ambiguous mash-up to date. It illustrated what could be accomplished with Google’s massive computing power. Why it failed is a proverb of our times, cautioning us on our punditry about the future. More recently is Larry Ellison’s brouhaha with HP over Mark Hurd and Oracle’s announcement of their “Cloud in a Box”. Don’t want to go all Glen Beck but all these events are related.

The result is a white-paper that consists of over 70 references and 40 citations that covers the conventional view of cloud computing [here] and what Google is doing that goes beyond conventional approaches [here]. I will admit that I do have an agenda or primary thesis: that cloud computing is about both virtual hosting AND organizing these virtual computation units into massive distributed computing networks. In other words, cloud computing is about computing not merely hosting.

So there is a post that covers Massive Distributed Computing and its features that change radically how we approach computing in the future [here]. Finally we take this new found perspective for a test drive looking at the surprising failure of Google Wave and Oracles assertion of a Cloud in a Box.

You may also want to take a look at those post that you might have avoided – “Send in the Clouds” and “Clouds of Hadoop”.

If anyone would like a copy of the complete white paper in Word format, please email me and I will provide a link with no further hassle or obligations. Otherwise, I hope you enjoy the series and find some value in the references and narrative.

Cloud computing is a concept best understood within the context of IT[i]. Its primary effect on business is the potential significant saving in capital expenditures (CAPEX) and potentially maintenance (OPEX)[ii] of IT’s hosting assets. This is due to leveraging the scale of large server farms, and the recent ability to allocate slices of CPU, memory and storage on demand within these farms[iii].

This is sometimes referred to as elastic computing or utility computing. Elastic in the sense that additional resources can be allocated to meet periods of peak demand[iv] and utility in that one pays only for resources used[v]. So this would be a boon for startups and for new ventures within established companies to launch new products and services with significantly reduced startup expenditures and time to market[vi].

Scalability and Availability At the Core

For large Internet companies with stable traffic the benefits quickly diminish as the enterprise can build at a greater long term cost savings what is referred to as internal clouds, which in similar fashion integrate all resources as a single maintained entity through virtualization[vii]. In this sense cloud computing, whether external or internal, refers to computing and infrastructure resources that are configured for high scalability and availability.

Scalability has already been discussed in a previous post[viii] and is a significant aspect of the second surge in cloud computing or Web 2.0. Availability is a related concept dealing with strategies for completing processes when computers and communications fail. Typically the strategies require duplicating processes on multiple computers in parallel and deciding at the completion of each process the correct result from these multiple sources[ix]. Duplicating processes is a form of concurrent programming and follows the same principles as scalability, including Sullivan’s Law[x] – it’s the high availability and not efficiency that is the critical factor[xi].

An Enterprise Platform with Broad Application

Cloud computing as defined above is neither a strikingly new business model nor novel technology breakthrough. After all, free or low cost hosting has been supported from the very beginning. What has changed is the experience of large Internet companies such as Google, Amazon, EBay, and Yahoo and more recently Facebook and Twitter in dealing with the scalability of servicing high consumer demand and streamlining the hardware / software platform cost[xii]. This has resulted in better leveraging of engineering and IT efforts such that the resulting platforms and infrastructure are more abstract and flexible to support a wider array product service demands[xiii].

This in turn enables and moves forward various development and support strategies, such as Web Services (WS), Software as a Service (SaaS), Platform as a Service (PaaS), Infrastructure as a Service (IaaS) and Managed Services, which are sectors that have been around for sometime but have yet to reach mainstream acceptance[xiv]. The expectation is that these will become more widely accepted and adopted successfully by more business models going forward[xv].

The Dark Side of the Cloud

As close to utopia as this would seem, cloud computing presents significant issues that must be addressed to successfully deploy applications to the cloud. The first and most significant is that cloud computing as discussed in most articles has little to do with computing and more to do with virtual hosting – getting resources into the cloud. Hosting issues that are relatively easy to address in self-hosted scenarios become an order of magnitude greater in virtual hosting environments. These include control, security, reliability, and performance monitoring[xvi]. Of course, new visualization technologies and standards have been developed to address these but require new IT procedures and tools that can be incompatible with standard operating procedures in a self-hosting scenario[xvii]. So the full implication of the change has to be considered along with business value proposition.

From Virtual Hosting to Cloud Computing

As important and fascinating as virtual hosting is to the evolution of the Internet, it is not by itself sufficient to ensure success or propel a Web 3.0 gestalt. Even with virtual resources with potential unlimited scalability and promised three sigma availability, ones actual results may differ. For success in realizing this potential and promise, the primary effort comes down to web architecture[xviii]. This means actually designing the service to be elastic and highly concurrent. We will find in this series of posts that this will require a new computing paradigm that is actually well understood and practiced by everyone except computer scientist.

To understand the implications of this paradigm, we will need to enlarge our scope of cloud computing beyond utility hosting to encompass how web site implementations and actual computing will change[xix]. This is what we will address in the next post and will begin with Google, who is the defacto architect of the current Cloud[xx] and appears to be moving in different direction from conventional approaches being implemented by Amazon, Yahoo, and Microsoft[xxi]. How can writing apps in Python be the basis of world domination? If we have learned anything from our history (as irreverent as it has been)[xxii], this is a clue we should probably not ignore.

Works Cited

Banerjee, U. (2009, 26-November). Cloud Computing Service: Amazon EC2 vs Google GAE. From Cloud Computing Journal: http://cloudcomputing.sys-con.com/node/1200706

Burrows, M. (2006, 7-November). The Chubby Lock Service for Loosely-Coupled Distributed Systems. From Google Research Publications: http://labs.google.com/papers/chubby.html

DataPlex. (2010, 7-January). Cloud Computing Issues. From Dataplex.com: http://www.dataplex.com/blog/index.php/2010/01/07/cloud-computing-issues/

Dedicated Server School. (2009, 16-August). Business Benefits of Cloud Computing. From Dedicated Server School: http://www.serverschool.com/cloud-computing/business-benefits-of-cloud-computing/

Dugal, I. S., Alexander, S., & Hellman, J. (2010, 15-July). Software as a Service. From WikiInvest: http://www.wikinvest.com/concept/Software_as_a_Service

Fogarty, K. (2010, 14-July). Cloud Computing: Today’s Four Favorite Flavors, Explained. From ITWorld: http://www.itworld.com/saas/113387/cloud-computing-todays-four-favorite-flavors-explained

Gedda, R. (2010, 24-February). Open source helps Facebook achieve massive app scalability. From CIO: http://www.cio.com.au/article/337284/open_source_helps_facebook_achieve_massive_app_scalability/

Grehan, R. (2008, 8-August). Inside Amazon Web Services. From Info World: Cloud Computing: http://www.infoworld.com/d/cloud-computing/inside-amazon-web-services-421

High Scalability. (2008, 3-September). Some Facebook Secrets to Better Operations. From High Scalability: http://highscalability.com/some-facebook-secrets-better-operations

Houghton, J. (2010, 25-July). Back to the Future: Monitoring the Cloud. From Cloud Computing Journal: http://cloudcomputing.sys-con.com/node/1475427

Houghton, J. (2010, 16-June). Cloud Computing Journal. From Business Value and Cloud Computing: http://cloudcomputing.sys-con.com/node/1433147

Houghton, J. (2010, 7-July). Virtualization Journal. From Virtualization Does Not a Cloud Make: http://virtualization.sys-con.com/node/1455031

Knorr, E., & Gruman, G. (2009, 5-May). What cloud computing really means. From Info World, Cloud Computing: http://www.infoworld.com/d/cloud-computing/what-cloud-computing-really-means-031

Kraft, T. (2010, 02-July). Clouds of Hadoop: How Map Reduce Changed the World . From Mind Before You Mine: http://mindbeforeyoumine.com/2010/07/02/clouds-of-hadoop/

Kraft, T. (2010, 25-June). Send In the Clouds: Old Metaphor Gets New Life. From Mind Before You Mine: http://mindbeforeyoumine.com/2010/06/25/send-in-the-clouds/

Marshall, D. (2008, 02-August). GoGrid wants to be the new face for cloud computing. From InfoWorld Visualization: http://www.infoworld.com/d/virtualization/gogrid-wants-be-new-face-cloud-computing-096?r=677&r=572

Patton, P. C., & Hersh, C. I. (2003, April). Asynchronous Non-Preemptive Tasking for Parallel and Distributed Systems. From A Quantitative Methods/Computer Science White Paper: http://140.209.112.90/WhitePapers/QMCS/ANT.html

Ricadela, A. (2007, 16-November). Computing Heads for the Clouds. From Bloomberg Businessweek: http://www.businessweek.com/technology/content/nov2007/tc20071116_379585.htm

Rothschild, J. (2009, 13-October). Why are Facebook, Digg, and Twitter so hard to scale? From High Scalability: http://highscalability.com/blog/2009/10/13/why-are-facebook-digg-and-twitter-so-hard-to-scale.html

Schonfeld, E. (2010, 7-September). As Digg Struggles, VP Of Engineering Is Shown The Door. From TechCrunch: http://techcrunch.com/2010/09/07/digg-struggles-vp-engineering-door/

Staten, J. (2010, 02-August). To Get Cloud Economics Right, Think Small, Very, Very Small. From James Staten’s Blog: http://blogs.forrester.com/james_staten/10-08-02-get_cloud_economics_right_think_small_very_very_small

One can be easily lulled into believing it is simple and straight forward. Though McLuhan’s presentation may seem flippant, the sociology in the background is serious and complex. Indeed it is the background or context of technology that at first seems unmoved and unchanging that does eventually change in ways that might be missed because the changes appear unrelated. But it is in understanding both the obvious implications of a technology and the not so obvious effects on culture and society that transforms technology into media in McLuhan’s view.

For example, an obvious and much anticipated characteristic of the internet is that it “makes the world much smaller”. One might expect in the free exchange of ideas and knowledge that the world as whole would become more unified. But as we look at society, it is anything but unified. Can the Internet as a medium be responsible for this?

As I look at the problem from the 4 Laws of Media, the results are startling and not what I initially expected. To review briefly: the four laws represent four aspects of media that must always be present. To apply, one must not only consider what is manifest and technical but also latent and behavioral in the background.

The result is a tetrad consisting of four co-equal aspects of the mediums effect on human behavior:

So now apply these rules to a medium such as the Internet (before the Web). The Internet (as invented by DARPA and Al Gore) enhances thought both in retrieving information that nurtures thought and expressing thoughts to help others. It obsolesced books or publications that inform thought but at a much slower rate and at higher cost preventing just anyone off the street from participating (except as a reader). The Internet retrieves the Forum in Athens where individuals came to share thoughts and dialogue.

Normally one would think of the Internet as a medium of communication with dialog as a form of communication. Cell telephones are a form of communication medium that enhances the voice, the Internet goes beyond voice and records & recalls what has been “voiced”. It is therefore an extension of the minds creative abilities to recall, extend and amplify thought.

This is important when considering the original characteristics and intentions and how the medium when taken to extremes reverses these characteristics. The expectation was that the Internet would bring democracy or equality to thought allowing persons anywhere in the world to participate in the public forum. The hope was for unity of thought that transcended specific cultures or traditions – world-wide culturalization. A scientist in Botswana would be on equal footing to one at MIT or Oxford.

The expected unity has not materialized. That is because the reverse of thought is thoughtlessness or those things that imprison the mind. Instead of amplifying just good and reasonable thoughts, the Internet allows marginal thoughts that would be normally isolated and pushed to the “fringe” of society to likewise find and bring together the few supporters to amplify dogma, bigotry, lies, superstition, and racism just as quickly. In a world of false equivalency, it seems only fair that every thought whether based on truth or lies to be treated equally.

Instead of unity, we have a fragmented American as well as World culture where individuals can believe and operate in their own “realities” with their own “truths” and find others that believe the same. So the fact that in this country no “reality” can get more than 20% to 30% confirmation in polling, is this because of the Internet? Your thoughts please.

To check this further, I have applied the Tetrads to Facebook, Twitter, and the Blogosphere. Amazingly they are very different media though they seem to be doing similar things communication wise. What are your views?

Not to belabor the obvious, yes this is important for anyone attempting to measure “behavior” on the Internet because we are attempting grasp the effect (sociology) of technology on behavior. So get out your crayons & paper and start drawn up some Tetrads!