vm file system performance

[scottr]

Hi,
I read this article on vm filesystem performance ( http://mitchellh.com/comparing-filesystem-performance-in-virtual-machines ). The numbers are very large 7Gbytes/sec etc. The reason for this is that the vm hypervisor is caching the file ( assumption? ). However, as a general discussion do the conclusions of the article make sense? ( namely that vm filesystems are faster due to caching than the underlying OS ). The magnitude of the numbers make it an uneasy read, and it makes me uncomfortable that something is amiss, but it could just be my point of reference.

 

[Cerb]

Compared to the underlying OS, the tests really didn't do anything.

Every modern OS does demand paging for files, which will include the virtual disk images, so there will typically be some kind of double caching going on, whether it's an improvement or not. The results are also only valid with OS X as the host OS. Also, if using raw reads, and thus bypassing a good bit of caching, that would explain superior NFS performance, and very poor shared folder performance, neither of which might be the case if a realistic test scenario were used, that applied any and all available caching, save for specific applications trying to bypass such features. So, take it with lots of salt.

Iozone is also not so easily compared to other benchmarks, and specificness of the tests is a bit odd.

So, I'm really not sure what you're asking, but those tests are also not telling much. You won't find too much, though, because hosted VMs are awfully complicated to make any generic benchmarks from.

 

[scottr]

Hi,
thanks for the response. I think what I am asking, is my gut tells me the test is not actually telling me enough to conclude anything, but I just cannot put the reasons why into words. Need more think time. And like you I need a real world scenario. Which I think I will try to do tomorrow.

thanks for taking the time to respond, and you have helped to push me into more investigation.

regards