On 09/22/2011 04:47 PM, Michael D Day wrote:

project-planning-bounces@ovirt.org wrote on 09/22/2011 05:34:38 PM:

...

Does such benchmark ever existed (apart from kernel build)? :) As you pointed Anthony, let the ovirt project first focus on simplify the build process before we develop a competing benchmark suite to the one that we base lots of our marketing over.

Yes, there are several distributed benchmarks that test different aspects of virtualization performance. SPECVirt tests server consolidation. It doesn't test memory over-commitment. In fact, I don't think anyone who publishes a SPECVirt score using a configuration that over-commits memory because it hurts the score. There are other benchmark suites that test HPC cluster-style workloads running on a hypervisor. And networking performance. These are just some of the ones we use within IBM. They are very useful. If we would have had better benchmark suites to test memory usage you would have decided to include the balloon driver in RHEL 5.4 instead of relying solely upon KSM. And if we had used a good distributed transactional benchmark like DayTrader we would have known early on that small-packet network performance is relatively poor. Just a couple of examples.

If a benchmarking project does happen, it would need to be a self-sustaining project under the oVirt community umbrella. That means there would need to be interest in the form of independent developer resources.

DayTrader is a web application benchmark. You provide it a profile and it will simulate a bunch of clients trying to connect to your web application. It's geared toward J2EE developers as it provides to do performance testing of your J2EE application. I think the way to think of this in the context of oVirt, is a suite of tools that can be used by an ISV/Customer to validate that they're deployment of oVirt is optimized appropriately. I definitely think its worth having a perf BoF as part of the workshop (or perhaps the next workshop) to start discussing how we handle performance measurement/tuning in an oVirt world. I agree though that we need to tackle things one at a time but it doesn't hurt to have an idea of where we want to be in the long term while we do that :-) Regards, Anthony Liguori

And keep in mind I'm passing on customer feedback. (I happen to think its a good suggestion as well).

Thanks,

Mike

Mike Day IBM Distinguished Engineer Chief Virtualization Architect, Open Systems Development Cell: +1 919 371-8786 | mdday@us.ibm.com http://code.ncultra.org