Re: hdparm

[glhenni@cs.sandia.gov (Gary L. Hennigan)]

Alexander <vulture@abac.com> writes:
| Hi...
| 
| Umm:
| 
| /dev/hda:
|  Timing buffer-cache reads:   64 MB in  1.61 seconds =39.75 MB/sec
|  Timing buffered disk reads:  32 MB in  6.87 seconds = 4.66 MB/sec
| 
| Buffer-cache reads? Uh... explain that to me please, this particular UDMA
| can't go past 33 MB/s.

That's just the cache Linux reserves in your RAM. That's why it's
39.75MB/s. The fastest single disk in existence is rated at about
20MB/s. So the disk you give results for above is getting roughly
4.66MB/s and you're getting about 40MB/s from the RAM cache.

| But I do believe I heard of a UDMA/66 or something like that. I'm not
| using that here, though, so...

Yes, I remember hearing something about that too. Plus, there's U2W
SCSI rated at 80MB/s. Again, for a single disk access the bandwidth,
whether it's 33 (UDMA), 40 (UW SCSI), 66 (UDMA/66) or 80MB/s (U2W
SCSI), is overkill. The fastest disks manufactured are currently the
10,000RPM drives, e.g., Seagate Cheetah, and their peak performance is
20MB/s, and that's peak, which means probably only when
reading/writing data on the outter tracks would you ever get that
rate. Of course you benefit from the extra bandwidth if you have
multiple devices on that bus, say you have UW SCSI rated at 40MB/s,
then you can run two of those Chetah's simultaneously without degraded 
performance.

My point was that if you're benchmarking a disk and you get greater
than 20MB/s you're seeing the results of caching, either in RAM or on
the disk itself. Or you have some special set up, e.g., solid state
disk (they still make these?) or a RAID.

Gary


--  
Unsubscribe?  mail -s unsubscribe debian-user-request@lists.debian.org < /dev/null