[music-dsp] [OT] SCSI Still Useful?
david at gardena.net
Sun Mar 18 00:23:19 EST 2001
On Saturday 17 March 2001 23:51, James Chandler Jr wrote:
> The new G4 doesn't seem noticeably faster from a user interface
> standpoint, but benchmark testing gives real high numbers, so
> hopefully it will give improvement on MIDI/Audio/Video tasks anyway
What did you expect? It's probably the video card and/or the GUI
subsystem design that's the UI bottleneck most of the time, just as
> Anyway, the Maxtor IDE drives benchmark more than twice as fast as
> the Cheetah striped RAID. The Cheetahs are run on a softraid
> maintained by the latest FWB Hard Disk Toolkit drivers.
> It is surprising to find 7200 RPM IDE drives outclassing a striped
> RAID of 10,000 RPM Cheetahs so drastically.
Not really. Transfer speed is a function of rotational speed AND
density, and as IDE drives are optimized for cost/space, they get
higher transfer rates as a bonus.
Meanwhile, they may still be outperformed by the SCSI drives on more
seek intensive benchmarks, which is where SCSI drives in general are
very strong, as they're designed for multiuser servers rather than
single user desktops.
> Am wondering if there is much point in leaving the Cheetahs and the
> 2940UW installed in the new G4?
Perhaps not, if your applications are better served with extreme
bandwidth than with low access times. (Audio applications generally
are, provided they're properly configured.)
You *may* find that the SCSI drives will deliver better # or
tracks/seek time performance in audio recording/editing applications,
but I don't know in this particular case. The latest IDE drives are
*very* fast, even when it comes to the old Akilles heel of IDE
drives; seek times...
> From earlier experience on Mac and PC, SCSI drives would offer
> performance advantages for MIDI and Audio which were not completely
> correlated to speed. It was commonplace that MIDI/Audio playback
> would be more stable with a slower SCSI drive, compared to a faster
> IDE drive.
Well, that was back when bandwidth AND seek times were almost too
high for audio regardless of what kind of drives you used. The
situation is a bit different now, at least if you can do with less
than some hundred audio tracks or so.
> One explanation was that IDE drives would hog the CPU when
> delivering those great performance numbers, and the hard drive CPU
> hogging could get in the way of good real-time Audio/MIDI
That was before EIDE busmaster DMA. This has been around for ages,
and virtually any modern IDE drive and controller should support it.
(Be aware that most OSes *disable* this feature by default, as it has
been known to fail with fatal consequences on some crappy drives.)
Another problem with IDE drives used to be that they can't disconnect
from the bus when working, like SCSI drives normally do. That
basically means you can't have more than one drive/interface.
However, it seems that this is no longer the case with some modern
IDE drives, which means that IDE striping mode software RAID becomes
even more viable, even with more than 2 drives.
> But with processors getting so fast, and IDE drives so fast
> nowadays, is this "common knowlege" still true?
No, but that's not because we have more CPU time to waste. Faster
CPUs won't help at all (except that the time you *don't* waste being
locked by the PCI bus will correspond to more FLOPS), as the PCI bus
is way too slow for CPU driven high bandwidth I/O anyway, and has
been for a long time.
Transferring N bytes without DMA is always going to take at least N /
4(bytes/word) / 33 (MHz), or possibly half that time if you have 66
MHz PCI. Doesn't sound very nice considering that some UDMA/66 drives
deliver almost a third of the theoretical maximum PCI/33 transfer
(No, I'm not talking about the interface speed, but the real
disk->controller speed. I have one DiamondMax drive and two IBM 75GXP
drives, and indeed, they can deliver more than 90% of the specified
peak rate continously, when operating on huge files. :-)
> Does my "pokey"
> Cheetah RAID have much chance of delivering better record/playback
> compared to the new IDE drives?
Probably not, unless you're going to work with very few audio tracks,
and want very fast seek + play response times. (Perhaps not even
then; I don't remember the bandwidth and seek time figures of those
drives. If the IDE drives have sufficiently high bandwidth, they'd
outperform the SCSI drives even if the latter had *zero* seek time.)
.- M A I A -------------------------------------------------.
| Multimedia Application Integration Architecture |
| A Free/Open Source Plugin API for Professional Multimedia |
`----------------------> http://www.linuxaudiodev.com/maia -'
.- David Olofson -------------------------------------------.
| Audio Hacker - Open Source Advocate - Singer - Songwriter |
`--------------------------------------> david at linuxdj.com -'
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
More information about the music-dsp