PDA

View Full Version : Express Stripe vs. SoftRaid



Byron
05-23-2002, 04:12 PM
I've been fascinated lately by the Guru's Raid Database, but am "missing" seeing some numbers using ATTO's Express Stripe instead of SoftRaid, especially since the raid volumes created by Express Stripe are usable both in OS 9 and X. Most especially because when I ran the benchmarks myself, the results were "interesting"...

First, the hardware specs: G4 450 (AGP graphics), ATTO UL3S, 2X 18.4 GB Atlas III 10 K drives.

Benchmark utility: ATTO ExpressPro Tools, Gurus endorsed settings (8 MB maximum transfer, 2 MB sample, no disk cache).

To get something to compare with, I initialized the drives in OS 9 using Apple's drivers. The results for a *single* drive were:
PR: 148.00 MB/s
SR: 114.62 MB/s
PW: 52.83 MB/s
SW: 51.96 MB/s

Then I booted into OS X, deleted the Apple volumes, and set up two two-drive Express Stripe raid volumes (a 10GB "scratch" volume made up of the outer tracks, and a 24 GB "graphics" volume made up of the rest of the drive).

And THEN I booted back into OS 9 and ran the EPT benchmarks. Here's the results:

SCRATCH
PR: 146.69 MB/s
SR: 144.44 MB/s
PW: 71.66 MB/s
SW: 70.24 MB/S

GRAPHICS
PR: 146.55 MB/s
SR: 144.50 MB/s
PW: 71.94 MB/s
SW: 70.59 MB/s

It's interesting to note that there's not a dime's worth of difference between the reported speeds of the outer tracks volume and the inner tracks volume.

It's also interesting that the sustained read results are almost exactly what Rick got with SoftRaid using two Atlas III 10k's on one channel, in both a Quicksilver and a B&W. He did get sustained write results of 100.43 MB/s in the Quicksilver, but "only" 52.40 in the B&W.

So I guess Express Stripe ain't *too* shabby while we're waiting for SoftRaid for OS X http://www.macgurus.com/ubb/message.gif

-- Byron
"Born a Junior"

TZ
05-23-2002, 05:53 PM
Byron,

Keep us updated and informed. I've been waiting for ExpressStripe to get to "1.1" or something as there are some noted issues it seems. If it is reliable, stable, and if Disk Warrior etc can repair volumes (or Disk Utility DFA).

The 53MB/s on B&W's is a 'known' roadblock. Only way to get past is to use a true 66 MHz controller in the 66 MHz video PCI slot - in which case I was able to get sustained writes 115MB/s with a pair of X15s. However, Adaptec has yet to come out with an OS X driver I can use so it is back to ATTO UL3S.

Disk Basher, Intech QuickBench X, and ExpressStripe (read only results under OS X) don't give a clear picture, but one thing that is certain, OS X copies files so much faster, that for now, I gave up using RAID - and safer probably.

I know three people have tried and used ExpressStripe even if only temporarily when it came out and for testing. Now that ATTO wants to charge $99 that may cool the heels of others that might have but turned off by the price as long as Apple's works, is supported, and is bound to improve. Can't test Apple's from OS 9 but I think using the three native OS X benchmarking utilities, Apple's will yield better numbers.

Gregory

ricks
05-23-2002, 07:42 PM
Hi Ya Byron,

I have been horribly busy in my business last few days, forgive me for not being very involved the last couple of days.

I agree that the numbers are interesting. I never tested Express Stripe after the OS9 compatibility was available. Your testing is very interesting since you went back and tried a couple of things that we didn't.

Amazing how AtlasIIIs can read so high and get so-so writes. I've been wondering how to up the writes for a long while now.

If you want your results in the DB please email me screenshots of the benchtest if you would. Include:

*Hardware particulars
???*Host Card
???*Drive Model number and name
???*Computer model/speed

*Software particulars
???*OS version
???*Software RAID utility
???*Drivers if different from the raid utility

You've done most of this already, be great to add it to the DB for all to see. You might also test with MacBench5 to see what the Disk and Publishing Disk numbers are. That utility is available from the Gurus ftp site if you need it.

When I mount a Drive or RAID Array I am testing I tend to name the mounted volume so I know what the configuration was IE: 2xAtlasOSXraid or 2xAtlasExpRAID0 or 1AtlasOSXdriver

That makes it easy when you do a screenshot to remember what each screenshot was of since the name of the drive/array is on the benchtest window. I had to do this in self defense as we started getting up over a hundred screenshots in the folder.

Thanks for the great post. Very informative, I look forward to having time tonight or this weekend to get back to it and compare with the DB numbers and see just what you were seeing.

I'll post back when I have time later on.

Rick

Byron
05-24-2002, 12:41 AM
Rick -- Okay, this weekend I'll retest and make screenshots with the drives named as you suggested (the ones I already made show my own rather-pedestrian drive names).

But now it occurs to me that for true comparison purposes, I should have also done a test series using the SoftRaid drivers, so I'll do that too. Oh, well... the weather promises to suck this weekend here in anything-but-sunny Denmark.

Gregory -- I agree with you about OS X's speed possibly making Raid redundant http://macgurus.com/infopop/emoticons/icon_smile.gif

Still, I'd like to see some testing standards developed with the "correct" (read "useful") parameters, because you're also right about the through-a-glass-darkly picture we get using the various X utilities. Gad, can anyone tell me what those numbers mean?

Byron
"Born a Junior"

ricks
05-24-2002, 02:04 AM
Byron,

You certainly don't have to re-do any screenshots, I'm happy if they get named so we all know what they were by 'saving as' or using a photo editor to add the text. I was just making a suggestion that seems to work for me. Testing drives is addicting and the better your system, the more useful the data is after a couple of months when the brain no longer remembers which was which.

Have fun. Thanks.

Rick

TZ
05-24-2002, 07:35 AM
I decided to stripe a couple X15's using ExpressStripe (ES?) and see. Ouch! Terrible. Got better with plain volumes that stripped. But under OS 9 it did show much better sustained reads (100MB/s) but always the 50MB/s for writes - B&W G3, UL3S.

the 'real' test was copying 10GB in 10 minutes (non-stripe) to X15.

A 10k Cheetah though delivers nearly as good - and a "flat line" writes even on a drive that is about 50% full after weeks of use with lots of additions and deletions.

I was copying the 10GB while doing other work - now that is something I could never have done in OS 9 and why I continue to use OS X.

Disk Basher 1.1.1 seems pretty real in numbers, free, and from Apple. Probably the best to use for OS X (once Rick figures what settings to use). One nice part is that it will export and save the test results to a text file. No pretty charts to look at though.

Gregory

rhunt000
05-24-2002, 07:45 AM
I tested SoftRaid and ExpressStripe under OS 9 and got nearly identical results at ~160 Read and ~160 Write on my 1 GHz DP. I can redo that and take some screenshots. I have noticed that with 2 Cheetahs on each channel the Read and Writes are very close to one another. Back in Nov I posted results on a 533 DP with 2 Quantum Atlas 10 K II on each channel and I was getting 200 Reads and 150 Writes. Here is the post http://www.macgurus.com/ubb/Forum1/HTML/000191.html I was questioning the write speeds and Kaye stated "A long while ago, a member said that the true measure of a drive's capability is SW. IMO, he is correct, tho at first I wondered about it." So even though now I'm not getting as high of Reads, my writes are higher and I am now sold on Cheetahs.

Byron
05-26-2002, 06:13 PM
Ah... I spent a lovely rainy day formatting and reformatting drives and running benchtests (all results emailed to Ricks). It's been lots of "fun" , especially coming to a surprising conclusion:

On MY machine, with MY SCSI host adapter, with MY SCSI drives configured as a RAID 0 voume, ATTO's Express Stripe (with OS 9 enabler) is FASTER than SoftRaid in OS 9.2.2. A little bit. Using ATTO's benchmark. But SoftRaid gets higher MacBench 5.0 scores. (Gosh, I should have a Master's in waffling, huh?)

No, I didn't capitalize "MY" all the time cause I'm an egomaniac, but to stress that your results may vary (and are certainly void where prohibted, licensed, taxed,or otherwise frowned upon).

Anyway, my machine is a G4 450 (AGP graphics) "Sawtooth", 1.5 GB ram, (running OS 9.2.2 under the tests), and the drives in question are 2 x 18.4 GB Atlas III (KW18L018-02-F), connected to an ATTO UL3S single-channel SCSI host adapter.

The results below are derived from ATTO's ExpressProTools benchmarking utility (Guru's recommended settings), and MacBench 5.0 (default settings).

1 X Atlas III with Apple driver
PR - 147.76 MB/s
SR - 111.33 MB/s
PW - 51.47 MB/s
SW - 50.73 MB/s
MB5 disk - 3652
MB5 pub disk - 1636

1 x Atlas III with SoftRaid 2.2.2 driver
PR - 148.11 MB/s
SR - 111.30 MB/s
PW - 51.13 MB/s
SW - 50.28 MB/s
MB5 disk - 3763
MB5 pub disk - 1660

Note that the Apple driver gives slightly "faster" results in EPT, but SR outscores it in MacBench. Note also that there's no point in running a single drive test with a drive formatted with Express Stripe -- to use it in OS 9, you have install the same Apple OS 9 driver. (I ran the test anyway, but why waste digital ink with essentially duplicate numbers?)

2 x Atlas III with SoftRaid 2.2.2 driver, striped as RAID 0
PR - 146.32 MB/s
SR - 144.47 MB/s
PW - 64.39 MB/s
SW - 63.40 MB/s
MB5 disk - 4479
MB5 pub disk - 1731

2 x Atlas III with Express Stripe 1.0, striped as RAID 0
PR - 146.82 MB/s
SR - 144.29 MB/s
PW - 71.60 MB/s
SW - 68.80 MB/s
MB5 disk - 4006
MB5 pub disk - 1662

Can anyone explain SoftRaid's higher MacBench scores? Due to the fine tuning possible when setting up a SoftRaid striped volume?

Oh, well... this was a fun exercise, but essentially moot for my purposes since I'm pretty much always in OS X now where, as Gregory said, all drives are faster. Just to bore you with more numbers, in OS X I took a 1.2 GB folder filled with 25 Photoshop tiff files ranging in size from 40-60MB apiece, and copied it to both my IBM 60 GXP (on the stock ata bus) and to an Express Stripe RAID 0 volume made from the Atlas IIIs, and timed the transfers (with a wristwatch). Results: it took 1 minute and 10 seconds to copy it to the IBM, but ONLY 1 minute and *1* second to copy it to the striped Atlas'. Whoopdeedoo! Just think of all I can accomplish in that extra 9 seconds!

Byron
"Born a Junior'

P.S. to Gregory (if you were asking): no, DiskWarrior can't even graph, much less repair, RAID volumes made with Express Stripe1.0.