Subject: Re: More on UFS performance
To: None <current-users@netbsd.org>
From: Thor Lancelot Simon <tls@cloud9.net>
List: current-users
Date: 11/30/1994 22:10:01
Oh -- one thing I forgot:
If anyone out there has access to a _really fast disk_, like a Barracuda
or perhaps a RAID 5 system, with a _really fast bus_ on it, like fast wide
SCSI, on a fast controller (what _is_ the fastest PCI SCSI controller you
can get these days? The Adaptec 2940? The NCR 53C825?), I'd love to hear
how close our "raw" values (iozone raw /dev/rsdwhatever) can get to the
manufacturer's quoted figures, or to benchmark values from other systems.
If the device-drivers or the SCSI subsystem are imposing a "speed limit"
for some reason, a lot of the logic behind what I've been writing may be
bogus, or at least we may not be getting performance as good as I think
we are.
I hope I'm not annoying anyone by going on at such length about this; I've
been installing a lot of NetBSD systems in commercial uses and it seems to
me that the more performance measurement we do, the better off everyone is.
(that is, _I_ think I have a good reason for blathering on about this, but
if y'all think I should shut up, I'll shut up. :-)
Thor