sqlio question.

Thread view:

Enable EMail Alerts

Start New Thread

Thread rating:

light_wt - 30 Jun 2008 22:57 GMT

hi,

i have a question about the sqlio

suppose i have two external drives
x:\ has got the following result
xKB sequential IOs writing
MBs/sec: 400
IOs/sec: 1800

y:\ has got
yKB sequential IOs writing
MBs/sec: 400
IOs/sec: 900

which drive is doing better on write?
is the write on y: better than x:?

secondly, how to find out the Outstanding IOs? I am using sqlio v1.5.SG.

thanks.

Reply to this Message

Linchi Shea - 01 Jul 2008 13:04 GMT

yKB was probably half the size of xKB. If you had chosen to run the tests
with the same I/O block size, the results would probably have been the same.
In other words, the two drives had similar performance with respect to
sequential writes of these block sizes.

You can spcify the numbe rof outstanding I/Os with -o command-line
parameter. You should also be able to observe the actual # of outstanding
I/Os maintained using perfmon counter I/O queue length.

Linchi

> hi,
>
[quoted text clipped - 17 lines]
>
> thanks.

Reply to this Message

light_wt - 01 Jul 2008 17:25 GMT

Linchi,

thanks for your insight. Yup, the block size is the only difference between
the two.

reason i ask is because, if one diviide the MBs/sec by IOs/sec, i will get
MB/IO. another word, it will show the how much MB will the storage transfer
in one IO. and I think the higher MB/IO will be better. is that a good
reasoning, or am I missing something?

regards to Outstanding IOs, is it identical to the disk queue lenght? if
not, what is a good number? I understand MS recommands a disk queue length of
less than 2.

thanks!

> yKB was probably half the size of xKB. If you had chosen to run the tests
> with the same I/O block size, the results would probably have been the same.
[quoted text clipped - 28 lines]
> >
> > thanks.

Reply to this Message

Linchi Shea - 01 Jul 2008 20:25 GMT

> in one IO. and I think the higher MB/IO will be better. is that a good
> reasoning, or am I missing something?

Not really. Block size per I/O is something you can control with sqlio or
something an app has control when issuing I/Os. It's more like an 'input
parameter' than an 'output value'. The key measures of an I/O subsystem's
performance are: (1)Megabytes per second--how much data can you push through
the I/O pipeline, (2) I/Os per second--how many I/Os can you do, and (3) I/O
response time (aka latency)--how fast can you do an I/O.

> regards to Outstanding IOs, is it identical to the disk queue lenght? if

It >can< be identical to the disk queue length. But since there may be
multiple layers of queues and some queues may not be directly visible to the
OS perfmon counter, it may not be identical to what the perfmon counter
measures. But in the case of sqlio, you should see the same value with
PhysicalDisk\current queue length for teh drive.

Linchi

> Linchi,
>
[quoted text clipped - 44 lines]
> > >
> > > thanks.

Reply to this Message