The Register ®

Biting the hand that feeds IT

The Register » Hardware » Servers »

Original URL: http://www.theregister.co.uk/2007/07/23/vmware_vmmark_benchmark/

VMware learns to measure itself

By Ashlee Vance in Mountain View (ashlee.vance@theregister.co.uk)
Published Monday 23rd July 2007 04:45 GMT

Not afraid to help itself while it helps others, VMware today rolled out a public version of its homegrown virtual server benchmarking tool.

VMware has been shopping VMmark in beta form since last year and now feels confident enough to set the test suite free. Server makers and end users are expected to use VMmark as a way of gaging the performance of certain workloads - namely file serving, mail serving, databases, java applications and transaction processing - running inside of virtual machines. With any luck, VMmark will add some customer-friendly metrics to a market woefully lacking in vendor vs. vendor and application comparisons.

Of course, just about every move VMware makes is meant to further its cause as the center of the x86 virtualization universe. So, it's not surprising that we find VMware working with benchmarking experts at the Standard Performance Evaluation Corporation to make a very VMmark-like suite an industry-standard.

Rivals Microsoft and SWsoft - along with the likes of Sun, AMD, Intel, HP, Dell and Red Hat - have joined VMware and SPEC to form a subcommittee meant to develop this new benchmark.

XenSource is notably absent from the group and has long argued (http://blogs.xensource.com/simon/?p=12) that it has a major performance lead over VMware due to the lightweight nature of the Xen hypervisor.

With VMmark, however, VMware has tried to shift the discussion away from the raw performance of a hypervisor layer to how well actual applications run on a fully-stacked, physical server.

"Customers will typically measure the overhead of virtualization software running on a particular box," Andrea Eubanks, a senior director at VMware, told us. "From the perspective of understanding how their workloads will run on different hardware, this is a very primitive way to measure performance."

The VMmark test suite includes load generators for the various workloads described earlier and includes SPEC's own SPECjbb and SPECweb code for testing Java and order entry. Half of the benchmarks run on Linux, while the other half run on Windows. Customers will find the Linux-friendly software wrapped up in handy virtual appliances. Microsoft's licensing restrictions blocked a similar deal for the Windows code, so some self-assembly is required.

Servers makers such as Sun (http://blogs.vmware.com/performance/2007/05/sun_uses_vmmark.html) and Dell (http://www.vmware.com/community/thread.jspa?messageID=660198) have already started issuing VMmark scores.

Eubanks is looking for the SPEC work around a standard virtualization benchmark to be completed by the second half of 2008.

It's hard to get excited about benchmarks. While helpful, they often end up making matters more confusing for customers, as vendors tweak, prod and poke their boxes to attain the best possible - often unrealistic - scores.

That said, customers have largely been left to fend for themselves in the virtualization game. As the market leader, it's natural for VMware to take charge and gives OEMs a helping hand with the measurement process. The end result should be a bunch of server and chip makers doing their best to craft kit that runs virtual servers really well.

An industry standard benchmark remains the more ideal end goal, but you'll have to live with VMware and VMware alone until at least 2008.

There's more information on VMmark due to post here (http://www.vmware.com/products/vmmark/overview.html).®

© Copyright 2008