Re: CPU usage on debian (was: Re: Memory usage on debian)
On Sun, Jan 26, 2003 at 01:19:14PM -0600, Ron Johnson wrote:
> On Sun, 2003-01-26 at 07:37, Pigeon wrote:
> > On Sun, Jan 26, 2003 at 03:03:10AM -0600, Ron Johnson wrote:
> > > On Sat, 2003-01-25 at 21:42, Pigeon wrote:
> > > > On Sat, Jan 25, 2003 at 12:20:56AM -0800, Paul Johnson wrote:
> > > > > On Fri, Jan 24, 2003 at 10:21:12PM +0000, Pigeon wrote:
> > > [snip]
> > > > Now, I have a UDMA66 HD, which on buffered disk reads in hdparm -t
> > > > gives rates of about 28Mb/s, both with the onboard VIA controller and
> > > > a CMD680 PCI card. I also have a Quantum Viking 4.5 SCSI drive and
> > > > Initio INIC-950P (9100UW) SCSI card. This only gives me around 10Mb/s
> > > > in hdparm -t. Seems a bit slow to me. hdparm -T gives over 200Mb/s for
> > > > any drive.
> > >
> > > Well, 4.5GB SCSI drives are pretty old, and 10MB/s is SCSI-2's *exact*
> > > max speed, so maybe you're maxing out the Viking.
> >
> > I thought it was 10MHz transfer rate over a 16-bit bus (68-pin cable),
> > giving 20Mb/s. No? It does seem suspiciously close to the "quantum" of
> > SCSI speed though. Is there a decent utility that allows me to see
> > what transfer mode it's using, and tweak it, in the manner of hdparm
> > and UDMA settings? I've had a look at scsiinfo but it doesn't seem to
> > be able to tell me what speed the SCSI bus is being driven at.
>
> SCSI-1 5MB/s (8 bit)
> SCSI-2 10MB/s (8 bit)
> Wide SCSI-2 20MB/s (16 bit)
Ah, right. OK. The drive is wide; sorry, should have made that clearer.
> Besides, who ever maxes out a specification? Even if the disk is
> Wide SCSI2, to get 50% of theoretical maximum is nothing to sneeze
> at...
Well, true... but it's a suspiciously exact 50%. If it was 60%, or
even 40%, I might be less suspicious!
> > The card itself claims 40Mb/s synchronous; I had expected to get up to
> > 20Mb/s from the Viking. A faster and larger SCSI drive is next on my
> > wish list; it's good that SCSI gear is finally becoming available
> > without having to pay a fortune for it.
>
> Yeah, but the problem is that ATA/EIDE is getting better, too
Never groan at an improvement :-) In "the real world", copying big
files between the Viking and my UDMA66 IDE drive, the IDE spends more
time active by quite a lot.
It would be useful to have a profiling tool that would monitor disk
access, and give information like time spent accessing each file, how
often it was accessed, what proportion of accesses came from data
cached in main memory, what proportion required an actual disk read,
what proportion found the data in the disk drive's cache, etc. etc. to
help figure out which directories go best on which drive. Any
recommendations?
Pigeon
Reply to: