Subject: Re: Hdparm
From: Marcus Radich (marcus@darena.co.nz)
Date: Tue Apr 18 2000 - 04:38:07 EDT
At 8:28 AM +0200 18/4/2000, Patrik Schindler wrote:
>At 16:37 Uhr -0700 17.04.2000, Dale Ang wrote:
>
> >/dev/sda:
> >Timing buffer-cache reads: 175.34 MB/sec
>
>Why do you all folks add buffered reads? You are benchmarking your
>RAM transfer speed, nout your drives with this.
>
:wq! PoC
We use this to get an idea of the speed of the motherboard in a
system as we have found that slow motherboards have a profound effect
on the speed of a server. (We use hdparm as an indication only)
We have seen a system which could get 30MB/sec off the drives, but
was set up on a cheap motherboard which benched around 22MB/sec.
Needless to say, it didn't perform well under load. Replacing the
motherboard with a higher quality board saw it go up to 80MB/sec, and
the machine performed *much* better. We didn't change the processor
or the drives.
The largest netatalk-powered site we manage has 15 designers and 10
admin users with G3s and G4s, all on 100BaseT, connected to a Digital
Prioris server. This was an WinNT 'upgrade' we performed last year.
Now, this machine has a single Pentium Pro 200 processor plugged into
a outstanding motherboard (for it's age), which benched over
100MB/sec. We set up the Mylex DAC960 RAID controller with 4
Fast/wide drives in a RAID 0+1, and they benched over 26MB/sec. The
designers are all working off this server, and working with hi-res
scans! Each mac gets over 9MB/sec, and under load it holds up very
well. Even with multiple simultaneous accesses, the designers seem to
be very happy with the speed.
When performing our WinNT upgrades, we always use hdparm to give us
an idea of how good the motherboard performance is (it's just an
indication), and how well the drives will go. We can then predict
whether the system is going to perform well.
mrad01
========================================================
"Always ask the question, never assume the answer."
Marcus Radich 1999
========================================================
This archive was generated by hypermail 2b28 : Wed Jan 17 2001 - 14:30:29 EST