Re: IDE vs SCSI for RAID

["Stephen J. Turnbull" <turnbull@example.com>]

>>>>> "Tobias" == Tobias Diedrich <ranma@example.com> writes:

    Tobias> Stephen J. Turnbull wrote:

    >> A P120 + PCI + IDE + 32MB combo didn't give better performance

Please note that "performance" means perceived interactive response on
a workstation, not nominal statistics from benchmarks.

    Tobias> You did enable Busmaster DMA transfers using hdparm -d1,

Not on Linux in mid-1996 I didn't.  You did notice the 1.2.13, didn't you?

    Tobias> In contrast to that I get only 20MB/sec

But who cares about those numbers?  (That's a real question: "who" =
"people involved in supporting Application X".)  _I_ don't care about
burst transfer rate, which is what "high-performance" IDE is optimized
for (because it's an easily measurable damned lie), and what hdparm
measured the last time I looked (years ago).  It may be relevant to
Jc's server application, if what is important is getting longish video
clips to a few dozen humans' workstations in real time.

If it's supporting a dozen programmers all working on separate modules
of a large C app (or several of them), it'll be spraying hundreds or
thousands bursts of 1-50 KILObytes, not multiple MB, in short periods
as they rebuild.  And if you're into swap (and who isn't?) you're
talking pages (4kB, IIRC)!  SCSI is a robust solution that doesn't
require lots of tuning by the admin in that context.

Not to run down Jc, but if he needs to ask "what are the advantages of
SCSI?", is he likely to be competent to fine-tune the disk to his
application _now_?  I know I'm not.  He (and I) could _become_
competent, I'm sure---but which is cheaper, a SCSI system or our time?

The historical answer is "hardware is cheap, wetware is dear."

-- 
University of Tsukuba                Tennodai 1-1-1 Tsukuba 305-8573 JAPAN
Institute of Policy and Planning Sciences       Tel/fax: +81 (298) 53-5091
_________________  _________________  _________________  _________________
What are those straight lines for?  "XEmacs rules."