Cloud Computing and Application Performance Management

By Ken Blackwell
published: Tuesday, November 24 2009

Cloud computing signifies a transformational approach in the way businesses consume and utilize IT resources. Based on the efficient use of on-demand computing, cloud computing is the logical evolution of efficiencies IT managers strive for as they manage the IT infrastructures they own today. In the future, virtualization and cloud computing will offer financially attractive options leading to reduced capital expenditures, freeing money for other investments.

But how does cloud computing affect application performance and management and, in turn, how does application performance in virtualized and cloud environments affect your end users? Where should you focus your efforts in the management of applications in these environments?

Quantum mechanics provides context ; yes, that's right - quantum mechanics

In quantum mechanics, the Heisenberg uncertainty principle states that certain pairs of physical properties, like position and momentum, cannot both be known to arbitrary precision. The more precisely one property is known, the less precisely the other can be known. In plain English, Werner Heisenberg figured out that for tiny things like a subatomic particle, it is impossible to know precisely and simultaneously where it is and how fast it is going. The more you nail one down, the more uncertainty you have in the other.

Heisenberg caused quite uproar when he first proposed the theory that became the uncertainty principle. Even Einstein challenged his theory for years because it stood in such direct contrast to the accepted principles of classical physics where both the speed and location of big things, like a truck rolling down a highway, can both be known accurately at the same time. In the end, Heisenberg was proven right and he got a Nobel Price out of it.

Heisenberg, Einstein, and scores of other physicists since them have tried for decades to meld together classical mechanics and quantum mechanics into a grand unified theory that explains the physical natures of things both big and small. To date the grand theory of relatively remains elusive.

OK... so what does that have to do with cloud computing?

It turns out that virtualized resources in the cloud are very similar to particles and waves in quantum mechanics; it is impossible to know both precisely where they are and how fast they are performing at the same time.

In cloud computing, dynamically scalable and virtualized resources are provided as a service to users and those users usually don't have knowledge of or control over the technology infrastructure in the cloud that supports them. At any given instant, a software application may have multiple layers of virtualization between it and a specific piece of hardware that is executing it. A Web application, for instance, may be clustered across multiple application servers running on a collection of virtual machines  spread across multiple physical nodes located in multiple data centers spread around the planet. To complicate things a bit more, all of these levels of virtualization and abstraction can be reshuffled at any time as the cloud readjusts resources dynamically.

The rate of change happening in enterprise IT environments has been accelerating for years, but cloud computing changes the measurement unit from mach to warp!  Measuring the performance of the stuff in the cloud starts to look more like an exercise in quantum physics than the classical physics behaviors of traditional IT computing models. The only way to get a precise read on where stuff is happening is to take a snapshot in time of a massive infrastructure. But a snapshot at an instant in time does not let you measure the speed of processing across some period of time. Conversely, trying to measure the performance of the various pieces of infrastructure involved in delivering a business service requires looking across a time window, and the infrastructure configuration can change in the middle of that window, so you can't say precisely where the processing is occurring during that time.

We still have to measure IT effectiveness in delivering business services, so where do we focus?

Most contemporary Service Level Agreements (SLAs) and Service Level Objectives (SLOs) are stated in classical terms of overall response times, throughput and availability of high level business services. They are no longer specified in quantum terms like processor and memory utilization, network throughput and system availability of the highly abstracted and virtualized components that makes up the cloud infrastructure. The result is that measuring the end-user's experience of a delivered business service, manifested in the transactions that the user initiates against the service, becomes the critical metric to monitor. Of course, an end-user perspective may be that of a human being or it may be one application component calling to another.

It is the transactions which are conceptually well above all of the abstractions and virtualizations provided in the cloud that keep acting like classical objects where you can know very precisely how fast they are executing as well as where they are executing in the logical business applications. Sure, you may not know precisely where those logical business applications are physically running at any point in time, but that is a level of detail that is not so important in managing the overall health of a business service. Think about that truck I mentioned earlier. You can know with equal precision where that truck is and how fast it is going despite the fact that you cannot precisely know the position and velocity of the subatomic particles in the truck's tires, steel and paint.

You need a constant in the world of cloud computing

In the highly virtualized and abstracted world of cloud computing, the only remaining constant is the transaction; everything else, from application to hardware, is changing so fast that measuring and analytically correlating the relationships between them is nearly impossible and will be out of date by the time the correlations are determined. The transaction is the unifying concept that provides the classical properties required for measuring service delivery effectiveness as well as the connections to the quantum components required for diagnosing issues down to the root cause. The transaction ties it all together. Heisenberg and Einstein would be proud.

Related Links:

CA