We’ve been having quite a few discussions with analysts over the past few months on the subject of “cloud”. The interesting thing about these discussions is the vast array of points of view from which those analysts are viewing “cloud”. Some are focused on the network aspects, others on pricing/differentiation, and some are even very focused on what “cloud” means to applications – and the organizations that will, allegedly, take advantage of the cloud as a means of application deployment.

One such analyst is Daryl Plummer of Gartner. Daryl has always been very application focused so it’s always a pleasure to speak with him and, of late, read what he has to say via his blog. (Daryl is also a cartoonist, and has turned his interests in that area on the cloud, resulting in “G-Men”. If you haven’t yet, take a gander. He’s quite talented.)

The last time we spoke to Daryl he asked “What can you do to help an organization move a monolithic application into the cloud?” That’s a fairly straightforward answer for F5, unless you specify that the organization wants to move workload into the cloud, not necessarily the entire application.

SOA IS BACK IN BUSINESS

See, the problem here is that workload is not the same thing as an application. Workload is more equivalent to, say, an activity in a business process orchestration than it is the entire process, which would equate more closely to the application.

Workload is a discrete block of application logic that is self-contained, and can be executed on its own. In structured languages we might codify this as a function, in an object-oriented language we’d likely go the route of a method, and in the land of SOA (Service Oriented Architecture) we’d call this a web service.

That’s right, folks, SOA has risen from the dead and is about to embark on a comeback tour.

Invariably applications always seem to have one or two “functions” that are fairly compute intense; these are the chunks of application logic that require more processing than others usually because they’re mathematically complex, or require a lot of analysis, or just involve churning through huge data sets. For whatever the reason, these “workloads” are expensive to run.

The belief is that these workloads are the ones that can be more effectively offloaded to the cloud. Often, these workloads are of a nightly or weekly execution nature; they aren’t run all the time and when they are running, nothing else can because it’s chewing up resources faster than housing values are dropping.

But you can’t “pull them out” of a monolithic application. The cloud wasn’t designed to assist in decomposition of monolithic applications into composite processes. It was designed, for the most part, to run applications; the two are not the same.

In order to move a “workload” into the cloud you have to decouple it from the application; you have to use the basic principles associated with SOA and decompose the application into its composite processes such that you can distribute those processes in a way that most effectively utilizes the processing power at hand – whether that’s locally or in the cloud. You can’t simply move a monolithic application into the cloud and expect the cloud provider to be able to dig into it and optimize the execution of specific processes. It just isn’t that smart.

BUT WHAT ABOUT GRID?

The concept of grid has always revolved around parallelization of processes; executing lengthy or computationally expensive tasks in parallel to reduce the amount of time required to complete. But grid requires that you separate out (decouple) the processes to be parallelized from the application. Grid isn’t necessarily smart enough either to move the distribute a specific function or operation across multiple machines in order to increase the speed of execution. At least not yet.

The problem appears to be that we’re attributing cloud and grid with attributes that are more akin to CPU scheduling than what they really are capable of doing. Yes, the use of CPU cycles is an integral part of the concept of cloud and grid, but the ability to schedule individual pieces of logic across CPUs is not something the cloud or grid is capable of doing – unless the developer uses tools and methodologies available to tell it to do so.

Which is the point of SOA, isn’t it? SOA (is supposed to, anyway) decomposes applications into discrete services so they can be distributed intelligently. If one service is reused by multiple business processes it can be replicated or moved into the cloud so that it scales appropriately to meet the demands that are placed upon it by other applications.

The problem, of course, is that decomposing monolithic applications requires resources and time. But there really is no other way to solve the problem – at least not yet. The cloud is not a huge bank of CPUs across which discrete functions can be distributed. It’s not. The cloud is a huge bank of servers and while it’s more than capable of distributing applications across those servers, it isn’t necessarily about optimizing the execution of applications across CPUs. That’s more grid, and taking advantage of grid is going to require some changes to the application, too.

Basically, if you’ve got a monolithic application you’re either (a) moving it en masse to the cloud or (b) ripping it apart into services or grid-enabled processes. Those are your options right now, take it or leave it. If you want to move “workload” into the cloud, you’re going to have to enable your applications to do so. And that means SOA or proprietary grid-enablement.

Or you can wait and see what happens next. But it’s likely that before grid meets cloud and actually creates a system capable of distributing both applications and workload – automatically – across servers and CPUs that it’ll be somebody else’s problem.

• SOA World 2009 West: Call for Papers Deadline June 30
• SOA & Cloud Computing To Intersect This Week at SOA World