How much throughput does a "typical" 64-track 96K session have?

[Allan Speers]

Well, I almost bought into the idea that the best ATA drives (barricudas) are as fast as scsi. If you look at the raw throughput numbers from the manufacturers, this looks to be true. In fact, there's one ATA drive that claims 150 mb/sec throughput!

gimme a break. It looks like the current "real world" throughput figures are aproximately as follows:

Fatstest ATA (WD1200JB 49 mb/sec outside tracks / 29 mb/sec inside tracks.

The new Cheetah 15.3K 75 g: 76 mb/sec outside / 51 mb/sec inside.
==================

Now for my question. Since we can now use the full potential of the UL3D card, we can fully max-out the capabilities of the new cheetah.

What I can't ascertain, is how much throughput is enough. At 44.1K, we know that 40 meg/ sec is enough to get 64 tracks w/ many edits and crossfades. we know this because that was the limitation on the Atto DC card, no matter how many drives were running, yes? (do I have that number wrong?)

I am trying to figure out how much throughput it would take to do 64 tracks w/ medium edits on a single drive. (I hate splitting sessions between drives.)

While there have been other threads on this topic in the past, I have yet to see a definitive answer.

Has anyone done, or seen, benchmarks on this?
Do we have absolute throughput numbers to use as guidelines?
================================================== ====

63 tracks @ 96K, medium edits and cross-fades, and one track in record:
What is the avg (even ballpark) throughput?

[Allan Speers]

Le bump. ..... Digi?

[Lee Blaske]

It's been explained to me that as far as what we're doing, seek time is far more important than throughput, and that throughput is usually a non-issue.

Lee Blaske

[Jules]

If I was to 96k HD, I could see myself hopping from

1 x SCSI hot swap + 1 x IDE internal drive

to

2 x SCSI hot swaps and 1 x huge internal IDE drive

But it would be nice to get a 96k maxed out sesion all on one SCSI hot swap.

[img]images/icons/smile.gif[/img]

[Digidesign ETS]

The sustained transfer rate for 64 tracks of 96k, 24 bit audio is a little over 17.5MB/sec. The formula for calculating the sustained rate is:

[Track Count * Bit Depth (in bytes; 16 bit = 2 bytes, 24 bit = 3 bytes, etc.)* Sample Rate]/1048576 (1 megabyte).

In your scenario, that's [64 * 3 * 96000]/1048576 = 17.6MB/sec (rounded up).

Keep in mind that the rates normally quoted for SCSI busses and ATA busses are burst rates from the drive's cache to either the host adapter (SCSI) or system memory (ATA), and are always much higher than sustained transfer rates.

--Digidesign ETS

[AE]

Sorry, don't have hard numbers that directly address your question. But I can say that while tracking exclusively at 24/96 since May on two internal Western Digital JB drives, in the worst case have maxed out a single drive around 45 tracks; best case, 60+.

[doug_hti]

I've studied into this, especially a lot for video. The SUSTAINED rate for basically BOTH SCSI and ATA of ONE DRIVE is equally about 35 megs/s. Both SCSI and ATA have higher burst rates...maybe 70 or so for ata and 100 or so for scsi (not sure though)...
So also in theory a 133 ATA card can handle 133 megs/sec and ultra scsi 160 can handle 160 megs/s. These numbers are definitley scewed a bit. Realistically a SCSI array (with at least 4 drives) would be about 120 megs/s sustained. . and the ATA would be about 80-90 megs/s sustained. Like someone said, it's the seek time that is really the important thing for audio because of all the small files. Video is really what needs the throughput, because uncompressed is about 24 megs/s for just one stream and they are much larger multi gig files. Any one single audio drive theoretically can handle the throughput of 64 tracks at 96k easily...but introduce a lot of tracks and edits..blah blah blah...there you go...that's where scsi will shine with a less than half seek rate.

Hope that helps a little

[Allan Speers]

Digidesign ETS,

Your figures don't at all reflect real-world demands of Protools. One would not be asking the drive for sustained reads of continuous tracks.

Real-world sustained-read figures are always much higher than random-reads. A typical session will have a lot of small files, plus fades, and of course, you must figure-in having at least one track in record at the same time.

Can't Digi give us a ballpark figure?
========================================
Quote:
" The SUSTAINED rate for basically BOTH SCSI and
ATA of ONE DRIVE is equally about 35 megs/s. "

Doug, where do those figures come from?

I have seen radically different numbers from barefeats, Macaddicts, and others. They all give higher (real world) numbers for ATA, And they all show significantly higher numbers for the top-end scsi drives, vs ATA.

Tests are always open to scrutiny, of course, so I'd very much like to see the data you refer to.

And, again, it's not sustained-reads that we need to worry about, but random-reads. (I think) [img]images/icons/confused.gif[/img]
=============================================

"as far as what we're doing, seek time is far more important than throughput, and that throughput is usually a non-issue."

Lee, that's a tough one for me to fully grasp. I can see this logic to some extent, there may be something to the fact that seek time is ANOTHER important issue. However, there is still a throughput issue. Dave Clementson posted several months ago that he sees less "too slow" errors now that he is running a UL3D set to "160." That says it clearly: Wether one drive or many drives, the old DC card setting was not high enough for a large HD session at 96K.

Another point: Drives with faster seek times will have higher RANDOM-read throughput. Again, it is not SUSTAINED-read throughput that I'm concerned about.
=========================================

AE, thanks for the hard numbers! That is some excellnt news. Actual benchmarks are always better than theoretical numbers, after all.

if I extrapolate those numbers to scsi UL-320, using the Barefeats test-ratio, it is clear that one Cheetah will do it, even on the inside tracks.

-And then again, if I were to use four large JD's, it would be realistic to never fill them up more than halfway, so that might actually be a realistic way to go. Hmmm. I thought this was going to be an easy decision.