Copyright 2010 TeamQuest Corporation. All Rights Reserved.
A big enough virtual pool of resources does not come without an effort. If you don t pay
attention to the characteristics of the workloads you re hosting, there s a big risk you ll either
over-provision or starve your applications.
To be sure that workloads will operate efficiently with each other (or that the analogous
Tetris-like shapes will be capable of fitting snuggly with one another) requires some
serious analysis. You need to understand a lot about the resource requirements of the
underlying applications in order to know whether one workload might interfere with
another creating unnecessary resource contention and queuing delays. So, especially for
critical workloads, it makes sense to do some careful up front analysis rather than simply
tossing various workloads onto your systems hoping that dynamic resource scheduling and
migration will be able to quickly and efficiently fit everything together for you on the fly.
Optimized workload placement
made possible by up front planning
Estimation, as we mentioned earlier, typically entails stacking workloads on top of each
other until a predetermined threshold is reached. It is the technique used by most of the tools
available for surveying a data center and identifying virtualization possibilities. The results are
cheap and easy to obtain, but represent rough estimates. By itself estimation will not come
close to helping you reach optimal performance.
Analytic modeling is not as fast, easy, and inexpensive as estimation, but it comes close.
(By analytic modeling we specifically mean modeling using an analytic queuing network
solver. Less sophisticated analytic mathematical models are perhaps better considered as
estimation tools.) Analytic modeling is much simpler and less time-consuming to set up
than load testing (or simulation modeling), and it is much more accurate than estimation.
Load testing can be very accurate if the synthetic loads and the underlying system configurations
are representative of the production environment. However, more time and expense is required
the closer the loads and the infrastructure get to matching production. For most situations, it
7 of 8 Capacity Management is Crucial in Virtualized Environments