Re: Windows IOPS Benchmark

Started by David Rhodes Clymer on Aug 30, 2012

open the first message on its own page

Message details and related content
On Wed, Aug 29, 2012 at 8:13 PM, Jonathan Tripathy [off-list ref] wrote:
On 30/08/2012 01:09, Jonathan Tripathy wrote:
quoted
Hi Everyone,

I'm using bcache with a RAID1 pair of SSDs (for the cache) with a
MD-RAID10 spindle array for the backing device. On top of this is LVM. This
setup is used with the Xen Hypervisor. Bcache is formatted with a sector
size of 512 bytes.

If I use an LV for a Linux DomU, I get fantastic disk performance using
fio (about 23k random write). However, when I use IOMeter in a Windows HVM
DomU (with GPLPV drivers installed), my avg IOPS is around 4000. I am using
the "default" Access Specification. Am I doing something wrong? Changing the
number of workers doesn't seem to help.

Any advice is appreciated.

Thanks
Actually nvm, I forgot to enable the dist target for each of the works. Now
I'm getting an avg iops of about 34k.

But this does leave me with a question: is the number of "workers" in
IOMeter akin to "IO Depth" in fio?
I've not used IOmeter, but I would assume that the number of "workers" would
be similar to the number of "jobs" in fio. IO depth/queue depth is the
number of IO requests that are queued for processing at any one time. So, if
you've got 4 workers, each keeping one IO operation queued at all times,
your effective IO depth would be 4, while the IO depth for each job/worker
would be 1. That's my understanding at least.

-davidc
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help