Thoughts on Virtualization

I am a regular of the ZDNet blog by Paul Murphy and thought I’d add to his thoughts on Virtualization and all the brouhaha that’s going on these days —

Virtualization? uh huh… by ZDNet‘s Paul Murphy — Virtualization is popular because it was popular – and not because there’s a practical reason to do it.

The most interesting thing I discovered in the process of working on a “high-visibility” project (ERP solution) is that most mgt-types don’t understand what Virtualization has to offer. Someone high up (high-up enough I guess) decides that Virtualization is the answer to all evils that haunt a modern datacenter. The claims are that —

  1. Virtualization reduces server sprawl
  2. Virtualization reduces power and cooling footprints
  3. It empowers the IT support organization to be agile (read build more boxes fast) and really support a dynamic business (with lots of development type activities going on)
  4. It is a cure for many problems..blah blah

But when you look at what you’re saving on the standard UNIX platforms (except Sun), the costs amount to something exorbitant. I won’t name the vendor, but it charges for everything starting from it’s multi-pathing software to Resource Mgt software to Virtualization, and they charge by the core.

Soon you start thinking, does this really buy me the cost savings by reducing server-sprawl?
Then the vendor will say, “Why look at this as a consolidation platform? Why don’t you think about the flexibility you’ll get by using this model? Moving workloads around on the fly, etc?”

The problem with that is that Workload management (called SLOs I believe) calls for very detailed and in-depth recording of metrics (what kind of loads are generated by applications, starting by categorizing by application types, etc.

So you first identify the right kinds of metrics to track. The collect the data for a reasonable period of time (say 3-4 months). Then, only after munging all that data, is it possible to say with any authority that a certain amount of resources are required for a particular workload (and build a system that can manage those resource requirements on the fly). T

his entire process might take about 1 year (from start to finish) before being a viable option (some shops I’ve been in are better equipped to do this kind of measurements than others — depending on how “modern” the IT organization usually is — does it “REALLY” employ standards such as ITIL or not, etc).

I’d say that something like Sun’s container model on the Cool-threads servers would be more appropriate for all the above criteria. Consolidation, Resource management, flexibility, etc.

  • SRM has been free with Solaris since Solaris 9.
  • Solaris 10 has the virtualization pieces completely free.
  • The hardware is cheap(er than the competition’s for sure)

Leave a Comment

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.