I was waiting to see if anybody had any insightful answers,
but it appears not...
> When a vendor or spec sheet says a disk frame will do 65 i/os per
> second, how do you translate that to megabytes per second?
You can't, unless every IO operation is the exact same size.
Also, that seems like a weird way to spec a disk. Usually they're
touted in terms of bandwidth or throughput, not just IOs.
Quote:> For example, EMC says the disks should do 65 i/os per second and when
> I do a "cp filename /dev/null" I get 5 megabytes per second transfer
> throughput. Do these numbers correlate with each other?
Not really. Run "iostat -xnM 1" and see what you get. Check this:
$ dd if=/dev/zero of=/var/tmp/JUNK bs=1 & iostat -xnM 1
extended device statistics
r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device
0.0 160.0 0.0 0.1 0.0 0.0 0.0 0.3 0 4 c0t0d0
Wow, 160 writes per second! Clearly my disk is far better than
yours. ;-) However, note that the actual amount written to disk is
pathetically small. Whereas...
$ dd if=/dev/zero of=/var/tmp/JUNK bs=1024k & iostat -xnM 1
extended device statistics
r/s w/s Mr/s Mw/s wait actv wsvc_t asvc_t %w %b device
0.0 126.5 0.0 13.8 4.4 2.0 34.5 15.8 100 100 c0t0d0
Fewer writes per second, but much more data written to disk.
This is why operations is not a good benchmark.
--
If mail to me bounces, try removing the "+STRING" part of the address.