Now For Some Real World Solaris xVM Benchmarks
By marchamilton on Sep 05, 2007
Oh, first let me quote from the last bullet on slide 16:
That was just to get a native OS performance baseline without without xVM. I'm not sure how many dCache servers are needed for 15 PB of data, but it is quite a few, and a 10-15% savings, especially for many of the sites which are maxed out on power, is one great benefit of Solaris.
As for Solaris xVM. First question everyone asks me is, "how much overhead will there be?" Some of the Linux hypervisor implementations are notoriously slow, so customers have good cause to wonder if we can do better with Solaris xVM. So with the caveate that this test is using a very early version of xVM (it is still, in fact, referred to in the benchmark results as Xen), let me quote the dCache on xVM results:
Wow! I think I have to go redo the Solaris xVM team's performance goals.
By the way, as far as I know, no one is running dCache on a NetApps server. That is because despite the name NetApps, you can't actually run applications on a NetApps box, it is a proprietary appliance that only runs the file system services that NetApps decides it wants you to run. Most of the dCache sites running Solaris are doing so on our Sun Fire x4500 data servers. You see, in this case as the name implies, the x4500 data server not only serves up your data using standard protocols like NFS, it also is a standard Solaris server that runs any of the 1000's of applications, like dCache, that run on Solaris. Not that the appliance model is a bad one. In fact, the x4500 would make a great building block for a storage server appliance. Lets just say that if Sun builds one, I bet that someone will run dCache on it.