Re: Bit off topic : It should go faster than this...

"Ralph" <rlro99_at_hotmail.com> wrote in message news:e2c49cae.0306130709.59d61be9_at_posting.google.com...

>
> Its 2 5300's each with 4 channels, so in total I have 8 channels going
> to 4 enclosures. So each 5300 controls 1 array of 26 disks in 2
> enclosures with 4 channels.

OK. So, each 5300 controls 2 enclosures. Presumably within each enclosure is the RAID 0 portion (striping), then the second enclosure gives you the RAID 1 (mirror)? If so, I'm willing to bet the striping inside each enclosure is disabled or non-operative because of the enclosure's internal controller and that's the reason for your not so optimal I/O.

> >
> Yeah to be honest I am not 100% with all the details of how all this
> stuff works. I saw a seriously IO bound system, got compaq to spec the
> fastest direct attacthed storage they could get, but since then have
> had little involvement apart from the fact that it was my idea and it
> hasn't worked as well as expected. It has done the job in that the
> batch times have gone from 20+ hours to 7, so we are hitting the
> window, its just that I'm sure there is more to come from this
> system...

It should be doing better. Good that it's within the window, but that IO power should have done better than one third of previous I/O system.

> Yeah when I asked for them to copy a 2gb file, they didn't have one so
> created two copys of the NT system stuff which approximated to 2gb
> with thousands of files. This took 32 seconds to copy. Once I had
> given them a real 2gb file it went to 24secs...

I'd have expected even more difference. Not bad, though. Yeah, makes a huge diff to have to update (or not) the MFT. Which reminds me: in your registry, make sure that HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\FileSystem\ NtfsDisableLastAccessUpdate = 1
Otherwise you're wasting I/O.

> Yeah its appreciated though, I'm really looking for reassurance that I
> am not mad/stupid.
>

Not at all. I find the I/O in NT and the PC environment somewhat arcane myself. Always have to measure and compare to make sure I'm not getting stuffed by some smart-arse vendor...

BTW: turn on the DISKPERF stuff and do some detail I/O physical and logical monitoring using the NT monitor. It shows problems with disk strings as I/O queue lengths where there shouldn't be any.

--
Cheers
Nuno Souto
wizofoz2k_at_yahoo.com.au.nospam