[Info-Ingres] Ingres 2.6 on Solaris 10: createdb slow
Roy Hann
specially at processed.almost.meat
Mon Apr 23 06:34:54 CDT 2007
"William Yuan" <yuan at ariel.its.unimelb.edu.au> wrote in message
news:mailman.143.1177319144.20015.info-ingres at kettleriverconsulting.com...
> Hi David,
>
> I tend to agree with you, but RAID5 is what the System Admin group chose
> to use on the filesystems (along with ufs). We are using high-speed 15k
> disks and the bonnie++ disk benchmarks were as follows:
I'm not familiar with Bonnie++ beyond what I've gleaned from a quick glance
at a few pages I just Googled, but something to be wary of with all
benchmarking programs--and Bonnie++ seems no different, is that they never
ever reflect the effects of queing. (The long-defunct Digital Review was
about the only publication I've ever seen that ever published benchmarks
that did consider queing properly, and it's been out of circulation for
close to 15 years.)
The disk subsystem can be screaming along magnificently, but if you are
adding IO requests to the queue even slightly faster than they are being
serviced, the users' experience of response time will degrade to a near
stand-still. (Google for the Forced-Flow Law for more info. The Denning
paper is quite good on the subject.)
Roy
More information about the Info-Ingres
mailing list