April 14, 2011

    The rush to virtualize everything NOW has been in vogue for a couple of years. But warning flags were raised at TTS by keynote speaker Leonid Grinshpan, a technical consultant with Oracle. He laid out graphically how much of a performance penalty there can be due to virtualization.

    There are a lot of problems being experienced in the field with virtualization with enterprise applications, said Grinshpan.

    Big companies like Thompson Reuters, for example, even have policy against virtualization their applications due to performance issues. On that company's major application, Virtual Machines (VMs) impose an additional overhead which can reduce performance by 15 to 20 percent, and disk I/O sequential read performance can degrade by 15% - 70% depending on the architecture.

    He modeled an example of three servers running two applications. When not virtualized, response time was always below the corporate standard of less than eight seconds and typically less than four seconds. Modeling with TeamQuest also highlighted that one server could be dropped and that performance would remain acceptable for both applications. When virtualized, however, the transaction rate soared for one app to more than 30 seconds. This, said Grinshpan, was due to VMs resulting in longer queuing times. He went into detail on queuing theory and showed specific numbers to back it up. All of this will be covered in a white paper that will be posted on the TeamQuest site within a few weeks. His message: VMs cause more CPU utilization and this has to be taken into account.

    Use TeamQuest to model performance and provide the right estimates in a VM environment, he said.For enterprise applications, capacity planning will never be dead.

    One way to address the sprawl of virtualization, said Rey Rios of TeamQuest, was to become far more agile by moving up the stages of the Capacity Management Maturity Model. He recommended the audience assess themselves against the model by visiting the TeamQuest site and completing a self assessment. That shows which level your organization is currently at and what needs to be done to move up to the next level.

    According to Gartner, 60 percent of U.S. companies are at the reactive stage. To move upwards, said Rios, they need the combination of the right tools as well as the right processes and a lot of hard work.

    It is not easy to jump from one level to another as it takes time and effort, said Rios. Do it sensibly in small steps. As you see the quick wins, everyone gets on board.

    Ron Potter and Jon Hill from TeamQuest discussed another aspect of agility in a virtual world the implementation of the Capacity Management Information System (CMIS). It forms a single book of record for all capacity and performance related information for IT infrastructure components

    If you build a CMIS, people and tools know where to go to get information, said Potter.


    Category: capacity-planning