PDA

View Full Version : hardware internal RAID 0 for G4



jsepeta
09-10-2001, 10:02 PM
my "new" system:
350mhz G4 / 1 gb RAM / 60 gb ATA100 / 2 * 9GB SCSI LVD
adaptec 2940 UW + softRAID from Hard Disk Toolkit 4.5
digidesign digi001 / emagic amt8

********

my question:

I'm currently running MacOS 9.1 and have recently upgraded to OSX, only to scamper back to 9.x because there's no support for the adaptec card, my digital audio interface, my midi interface, and of course, ProTools and Logic Audio Platinum. I would like to stick with SCSI RAIDs because I already have 3 9gb UW SCSI drives and 2 9gb LVD drives. However, I've seen some impressive times with the new ATA100 RAIDs and I was wondering if it's worth moving to ATA since any drives I would buy now are much bigger than 9gb each. however, LVD RAID seems like it would still give a performance benefit for recording & playing back digital audio. My goal is to get good performance in 9.x while knowing that any further investment will be compatible once my applications are ported to OSX.

Any recommendations for brands & models of cards? Any way I can spend under $300?

thanks in advance

ricks
09-20-2001, 10:36 PM
jsepeta,
I bought a new Quicksilver recently and it's IBM ATA100 drive gave most impressive performance. I got 61.70 sus read and 58.28 sustained write with ATTOs bench utility on the motherboard ATA bus.

Currently there are no ATA100 raid cards on the market, only ATA66. It appears that software ATA raid options give virually NO improvement over non raid.

Acard has a new ATA100 card about to hit the market that MIGHT work out as a reasonable and inexpensive option to SCSI raid. Also Sonnet will probably come out with the same hardware raid card in the very near future. Gurus would probably carry the Sonnet card since they sell the Sonnet(Acard) ATA66 raid card.

These cards are manufacturer est retail priced around $180. With the price of ATA100 Drives where they are, I suppose one could build a raid for $400 to $500 dollars depending on which drives and how big.

Don't forget cables. For ATA100 you need 80 wire 40 pin cables to meet the ATA100 spec. These are sold by the GURUS. As well as great Barracuda ATA100 drives.

You can probably get better performance from a Ultra160 raid with newer and faster drives. The cost will be higher, performance will depend on the quality of your components ie Cheetahs, ATTO Ultra3 host card, Granite cables.

If you can believe the hype from Acard on their new ATA100+ Raid card we ought to see something in the 90mb/second range. Untill I have one of the new cards(ordered,and I'm waiting) I wouldn't bet much money on it. I will post the results of the ATA100 raid when it arrives.

Rick

sauria
09-28-2001, 10:59 PM
ricks -- what is the file transfer size when you get those great results?

TZ
09-29-2001, 08:07 AM
There was a news update that Acard is skipping ATA/100 and now has ATA/133 cards. That should make for some good numbers. And Maxtor seems to have jumped the gun with IDE 133 drives.

I'm thinking that would be one way huge boost to older Yosemite systems with their ATA/33 bus interface by a long shot.

Gregory

jsepeta
10-01-2001, 02:32 AM
ended up getting 2 * IBM deskstar 60gb / 7200rpm / ATA100 drives & the Acard ATA100 card (no raid). set it up with HDToolkit 4.5's RAID0, and achieved between 39 & 40mb/s sustained read & write on my G4/350 (pci). this is about 3x better than I was getting from my old adaptec 2940UW & 2 * 9gb IBM UW drives -- which altogether cost me somewhere in the neighborhood of $1000! but it's not NEARLY as fast as your marks up in the 60's -- what am i doing wrong? for now, i can't complain -- i was able to duplicate some 4000 items on the non-raided drive in less than a minute, which is way faster than before. perhaps i'll even get ahold of osX 10.1 this week...

magician
10-01-2001, 01:29 PM
there is an easy explanation for the discrepancy in thruput.

PCI.

the PCI bus on the Sawtooth is capable of shoveling much, much more data, so it can actually exploit hardware like U160 and U3 and U320 PCI cards, and ATA/100 and ATA/133 boards. The Yosemite will be much slower, unfortunately.

we are testing an ACARD ATA/100 version put out by a company called Miglia (same guys that do the Director's Cut and some other stuff). It should be up on our website shortly. We may actually put up ACARD product itself, depending on firmware testing.

I can confirm that RAID 0 or RAID 1 using ATA hardware does not scale like SCSI. The only reason to even attempt it is if you really need mirroring or a larger virtual volume. Remember that under OS10.1, RAID 0 and RAID 1 is built into the OS, and available using the Disk Utility. No separate board is required, and the OS supports both SCSI (LVD) and ATA drives! (This could be the coolest feature yet).


http://www.macgurus.com/ubb/magician2.gif

sauria
10-01-2001, 08:06 PM
Have you tested Quicksilver PCI throughput compared to Sawtooth? It's suposed to be better than Sawtooth. Didn't they get rid of a bottleneck?

ricks
10-02-2001, 12:33 AM
sauria,
Sorry I missed your request, I was in Reno for the big horse show.

Its now been several weeks since I ran those tests, the only notes I have are the sustained numbers. I had thought I ran all 8mb tests, however, at this time I can't reproduce the results. When I ran those benchmarks, the drive was empty. Now the drive is 3/4 full and my results are;
40.53 sus read and 32.83 sustained write. I see that the first 2 MB roll in at around 60mb/s then a instant drop off to the sustained numbers for the additional 6mb. I would guess that the buffer is causing the high and that I am actually in the 30 to 40 mb/sec range.

When the new ATA133 RAID card makes it out I will again bench these drives in an empty condition being sure to record EXACTLY how and what I bench. Then I will refill and retest. Until then, your question forces me to question my methods.


I don't think the PCI bus is the cause of any difference between a B&W and a Quicksilver. The quicksilver bypasses the PCI bus with the ATA directly installed on the motherboard. I don't think the PCI bus has any bearing on ATA performance in the Quicksilver.

Rick

magician
10-02-2001, 01:51 PM
in a way, you are correct in thinking that IDE thruput has little to do with PCI bandwidth, but in consulting the Developers Note for the Sawtooth (http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/PowerMacG4/2Architecture/Block_Diagram_and_Buses.html) you will instantly see that the UltraDMA/66 bus in the Sawtooth routes to the UNI-North memory controller and PCI bridge through the KeyLargo I/O device and Disk Controller, while the PCI bus (and its cards) routes directly to UNI-N, and then to the processor.

data to and from the IDE devices, then, has a less direct route to the G4 than data coming off the PCI bus.

when i replied to the earlier post, i merely said "PCI" as my explanation, speaking and thinking in shorthand when I should have been more expansive.

there are two other factors in comparative IDE performance between the Yosemite and the improved Sawtooth.

Sawtooth uses UNI-N, while Yosemite uses Grackle. Grackle is the same PCI bridge and memory controller used in the Beige G3. UNI-N is only used in Sawtooth, and it is a primary reason why PCI bandwidth in these latest and greatest Power Macs is so phat.

Sawtooth has DMA/66 on the logic board, while Yosemite supports DMA/33.


http://www.macgurus.com/ubb/magician2.gif

[This message has been edited by magician (edited 02 October 2001).]

ricks
10-02-2001, 02:30 PM
magician,
Your amazing! Sounds like a gain could be had going to a ATA 100 or 133 PCI card then(on Sawtooth or even Quicksilver). This without considering RAID options, just through-put.

Another question, is the new Quicksilver a new logic board or just a improved sawtooth?

In looking at the developer notes I had to wonder why the AGP port is cross connected to USB1 port?
Rick

[This message has been edited by ricks (edited 02 October 2001).]

sauria
10-02-2001, 11:20 PM
Thanks ricks -- pretty good #'s

magician
10-03-2001, 03:57 PM
the QuickSilver uses a Sawtooth logic board like all other Apple G4's after the Yikes (which simply stuck a G4 ZIF on a Yosemite logic board). I'm not certain what variations, if any, are present at a low level on the newer Sawtooth. I haven't really sat down and compared an older Sawtooth with a newer one, looking at chipsets, etc. Aside from the obvious stuff (four PCI slots versus three, three SDRAM slots versus four, etc), the critical issue is that UNI-N is used on all Sawtooth main logic boards, and the newer your Sawtooth, the faster your PCI thruput.

ricks
10-03-2001, 05:53 PM
Thanks Magician,

I appreciated the direct to Apples developer server. VERY intersting stuff.

Rick

tim_s
10-08-2001, 08:48 PM
<BLOCKQUOTE class="ip-ubbcode-quote"><font size="-1">quote:</font><HR>
in consulting the Developers Note for the Sawtooth you will instantly see that the UltraDMA/66 bus in the Sawtooth routes to the UNI-North memory controller and PCI bridge through the KeyLargo I/O device and Disk Controller, while the PCI bus (and its cards) routes directly to UNI-N, and then to the processor.
<HR></BLOCKQUOTE>

Whoa there. You have to be careful about Apple's devnotes these days -- what you linked to is the Developer's Note for the current G4 Macs (Quicksilver). The page listing devnotes for all versions of every Mac is:
http://developer.apple.com/techpubs/hardware/hardware2.html

and the devnote for the original AGP G4 (Sawtooth) is:
http://developer.apple.com/techpubs/hardware/Developer_Notes/Macintosh_CPUs-G4/Original_PowerMac_G4/index.html

(You'll notice a whole bunch of different PowerMac G4 documents. Apple revved the thing quite a bit.)

As you can verify, the original Sawtooth had a version of Uni-N with a 66 MHz 32-bit PCI bus. They ran it through a PCI-to-PCI bridge to get a secondary 33 MHz 64-bit PCI bus, to which Key Largo and the PCI slots were attached. Contrast to the Quicksilver which has 33 MHz 64-bit PCI coming straight out of Uni-N to Key Largo and the PCI slots. (This is probably a reason for better PCI performance on the Quicksilvers, BTW: PCI to PCI bridges add latency, which hurts throughput on short PCI transactions.)

<BLOCKQUOTE class="ip-ubbcode-quote"><font size="-1">quote:</font><HR>
data to and from the IDE devices, then, has a less direct route to the G4 than data coming off the PCI bus.
<HR></BLOCKQUOTE>

Not true. Key Largo is nothing more than a PCI device which happens to be soldered to the motherboard rather than a plug-in card. Look at the block diagram again -- it is connected to the same PCI bus as the expansion slots.