MacGurus Technical Forums

MacGurus Technical Forums (https://www.macgurus.com/forums/index.php)
-   RAID (https://www.macgurus.com/forums/forumdisplay.php?f=7)
-   -   Drive and RAID DB Updates (https://www.macgurus.com/forums/showthread.php?t=17554)

kaye 09-06-2003 10:48 AM

Responding to a suggestion to note updates I am starting this thread which will be locked.

Latest are some FW800 single drive and striped RAID at the very bottom of the DB. These are new Granite FW800 cases, 20-1032, which will be on their website for viewing Monday or Tuesday. Granite is building and proofing the pages now. Rick should have the MacGurus store open next week also and will be offering Granite products.

I like the Granite cases, easy to install a drive into, fan for cooling, front on/off switch, front LEDs for power and drive activity, FW800 cable, FW800 to FW400 cable, and USB2 cable (USB2 port on back as well as three FW800 ports) with backwards compatibility with USB1.1 and ATA-6 big drive compatibility. More features will be on the website. k

kaye 09-08-2003 07:38 PM

Updated this evening with a couple more SATA Raptor RAID entries, both Apple and SR3 along with first tests of SeriTek/1S2 revision 2 firmware which significantly boosts sustained reads in OSX.

Also more FW800 results with SR3. One minor bug with FW and SR3. After a restart or bootup in Jaguar, a window opens, or windows if more than one SR3 FW drive, to the effect that the drive is unreadable in OSX. Just click Ignore. SoftRAID is aware of this FW bug.

Just received two SATA 120GB Maxtor DiamondMax Plus 9 drives. Rick wants to sell the Maxtors. My next tests. k

kaye 09-11-2003 11:40 AM

More SATA tests with SeriTek/1S2 revision 2 firmware, this time with 120GB Maxtor DiamondMax Plus 9 8MB cache SATA drives. The drives have both a 4-pin Molex connector and a SATA power connector. Use one or the other, but not both. Very impressive performance with these SATA Maxtors.

I also tested, in striped RAID, a new version of ATTO ExpressStripe, v2.0. More features and the striped RAID in Jaguar is now as fast as Apple RAID and SoftRAID 3. The ATTO ExpressStripe benchmark utility is however still not of much use in Jaguar and now gives erroneously very low results for Sustained Reads in striped RAID. It still does not measure Sustained Writes.

In OS9.2.2, with the ATTO extention loaded, the ATTO RAID is mountable and perfectly useable, with a slight performance hit compared with SR 2.2.2. Useability in both OSes is a definite plus, but you cannot run DiskWarrior. It will make the RAID unusable. That is a definite downside. k

kaye 10-20-2003 05:12 AM

I have been conducting some tests on an external 4-bay Burly stuffed with 4x Raptors (total about 138GB after RAID0 striping with Disk Utility's RAID function) and 2x SeriTek/1S2 cards in my G4-1.25G DP MDD (FW400 dual boot). The results were disappointing to say the least, equally disappointing in both OS9.2.2 and 10.2.8, though I was warned by FirmTek that they would be. What I was told:

"It is never a good idea to stripe "across" the PCI bus: currently our card is "saturating" the PCI bus and two of them won't do any better, it could even hurt a bit. We are dealing with a 32-bit 66 MHz PCI card and 64-bit 33 MHz PCI bus. Guess what speed the entire subsystem is running? Of course, with the smallest common denominator - 32-bit and 33 MHz."

So why can my two Atto UL3D cards result in excess of 200MB/s thruput? Because tho they are 32-bit PCI compatible they are also 64-bit PCI compatible (and tho they are 33 MHz PCI compatible they are also 66 MHz PCI compatible). So on my MDD, the UL3Ds run at 64-bit 33MHz, the same as the MDD PCI bus. To follow along the ATTO path for a moment, the UL4S and UL4D retain all of those compatibilities but also 64-bit 133 MHz PCI-X compatibility.

Yesterday I took my Burly 4x RaptorRAID to Rick's Ranch and his new G5-2GHz Dual running the G5 version of 10.2.8.

G5s, both the 1.8GHz PowerPC G5 and Dual 2GHz PowerPC G5 have three open full-length PCI-X slots: one 133MHz 64-bit slot and two 100MHz 64-bit slots.

First installed 2x SeriTek/1S2 cards in the G5's 100MHz 64-bit slots, ran some one meter SATA data cables out of the G5's open 133MHz slot backplane to the Burly drives. Using QuickBench, the tests up to 100MB, results were much faster than my G4 but still disappointing. Moved one of the SeriTek/1S2 cards to the 133MHz 64-bit slot, the other still remained in the 100MHz 64-bit slot, WOW. Sustained reads now 187MB/s, sustained writes 270MB/s.

We also tested 1x SeriTek/1S2 with 1x and 2x Raptors in both the 100MHz and 133MHz slots. Results will be posted soon to the Drive and RAID DB. k

kaye 11-07-2003 11:12 AM

The Drive and RAID DB has been updated with several new entries under SCSI testing both OS9, Panther, and Jaguar 10.2.8. Panther is a huge improvement over Jaguar. I'll have more details when I get a writeup together for here. k

kaye 11-07-2003 08:50 PM

The intent of the latest SCSI DB tests was multipurpose. First, you will note that there were no previous OSX test reports. Second, the Gurus store already has the latest Seagate Cheetah X15.3 drives in various capacities, 18GB, 36GB, and 73GB, as well as Seagate Cheetah 10k.6 drives in 37GB, 73GB, and 147GB sizes. Third, the store has the Burly SCSI Enclosures in 2-bay, 4-bay, and 8-bay sizes. Fourth, the store has all of the necessary Granite Digital LVD external cables, Granite internal TPO LVD cables for the Burly, Granite terminators, and Ultimate Drive Coolers. Last, one of the intents was to keep the cost of SCSI RAID as reasonable as possible, if SCSI can be called reasonable. But when only the best will do.....

To keep costs down, as well as my own, I chose a 4-bay enclosure and added two more X15.3 18GB drives to the two I already owned (they needed a home anyway). I already had all of the necessary Granite stuff and Ultimate Drive Coolers from a Burly 8-bay enclosure which I gutted for the parts. I also have a UL3D. To get this card to work in Jaguar or Panther, the ATTO website really confused me. I asked TZ to spill his guts on what I needed. He emailed me a ton of info and links on what to do. Essentially, for a UL3D in an MDD G4 what is needed:

There are three things to download for my UL3D.
1.ATTO ExpressPCI Configuration Tool v2.50
2.Driver installer package v2.05 (xulplus205.tgz)
3.Flash image for the specific adapter (UL3D-000) to flash the firmware, which for that UL3D and G4 is 1.6.6f0.

The Flash image confused me. I was looking for a 2.x flash image. Nope. It is also easy to confuse driver and flash. You need all three downloads and these are for the UL3D. The download links on the ATTO website will tailor your download for each of the items above to type of card and G4 or G5. I also was confused by the UL3D coming in two versions the UL3D 33MHz and UL3D 66MHz. My retail UL3D sez on the box UL3D-000 which is the 33MHz version. If the box sez UL3D-066, you can imagine... There are other ways to find out, but the UL3D-066 is far less common.

My MDD is a 1.25G DP dual-boot FW400. I first tested the RAID in OS9.2.2 with older UL3D flash firmware 1.6.4f1 and SoftRAID 2.2.2, since I had found for me that it was faster than 1.6.5 firmware. The latest firmware 1.6.6f0 however is definitely faster than older firmware in OS9.2.2 and of course is necessary for Jaguar or Panther anyway.

Last, I have no idea what the problem is with Jaguar 10.2.8 very low results. I built an AppleRAID in Disk Utility for 10.3 with journaling enabled twice, same low 10.2.8 results. Built an AppleRAID with Disk Utility in 10.2.8 twice (no journaling), same poor results. Also tried SoftRAID 3.0.0 and 3.0.1 with journaling disabled, better results with SR3.0.1 in 10.2.8 but still low results. And Jaguar 10.2.8 exhibited one other very troublesome problem. With the RAID mounted, I could not shutdown (or reboot from 10.2.8 to either 10.2.8 or 10.3). Constant spinning gear on a blue screen, would not shutdown, would not even get to a reboot. I am guessing but I suspect that 10.2.8 does not know how to do the necessary cleanup of the SCSI RAID (such as write back to the RAID drives whatever is being held in their onboard drive cache). This happened always with AppleRAID regardless of whether built in 10.3 or 10.2.8. It also happened somewhat less frequently with SoftRAID 3.0.0 and 3.0.1. I also tried unmounting the RAID before shutdown/reboot in 10.2.8. No help. The only way I could consistently reboot or shutdown from 10.2.8 was to unmount the RAID and reboot to OS9.2.2. Then I could shutdown or reboot to 10.2.8 or 10.3. No such problems with Panther 10.3 as well as much better thruput. And no necessity to unmount the RAID first.

You would most likely want larger capacity Cheetahs for a serious SCSI RAID. These 4x 18GB X15.3 drives, after building the RAID in OS9.2.2 or 10.2.8 or 10.3 are a total of 68.37GB.

If you have questions, comments, criticisms, or your own experiences, start a New Topic in the RAID Forum. Next is to take the Burly to Rick for testing on his DP G5 with UL4D. k

kaye 12-19-2003 10:12 PM

Just getting started with this new drive. Keep in mind this Raptor will get faster as it breaks in. A comparison with a 36GB Raptor tested again.

Apple DU OS10.3.2/OS9 drivers.

OS9.2.2, EPT:
WD740GD, SR-69, SW-096
WD360GD, SR-57, SW-105

OS9.2.2, DB9-128MB:
WD740GD, SR-72, SW-071
WD360GD, SR-57, SW-057

OS9.2.2, QB9-10MB:
WD740GD, SR-53, SW-104
WD360GD, SR-45, SW-102

OS9.2.2, MB5 (have not retested the WD360GD yet so those are old):
WD740GD, D-5325, PD-4651
WD360GD, D-5028, PD-4133 (SoftRAID driver)
WD360GD, D-4791, PD-4099 (Apple driver)

OS10.3.2, QB2.1-100MB:
WD740GD, SR-64, SW-074
WD360GD, SR-52, SW-059

That is what I've done so far. k

kaye 12-23-2003 12:07 PM

Raptor is now incrementally faster as it breaks in. Apple DU OS10.3.2/OS9 drivers. Latest SeriTek firmware 5.0.3.L335. MDD G4-1.25GHz DP FW400.

OS9.2.2, EPT:
WD740GD, SR-71, SW-100
WD360GD, SR-59, SW-106

OS9.2.2, DB9-128MB:
WD740GD, SR-72, SW-071
WD360GD, SR-57, SW-057

OS9.2.2, QB9-10MB:
WD740GD, SR-53, SW-109
WD360GD, SR-47, SW-102

OS9.2.2, MB5:
WD740GD, D-5610, PD-4946
WD360GD, D-5042, PD-4143

OS10.3.2, QB2.1-100MB:
WD740GD, SR-64, SW-074
WD360GD, SR-52, SW-059

Updated the ATTO driver (3.0.0) in 10.3.2 for my ATTO UL3D and tested my 4x 18GB X15.3 Burly RAID using SoftRAID 3.0.2. Partitioned the drives for one large volume 58.27GB and one small volume on the slowest part of the RAID 10.01GB.

OS10.3.2, SR3.0.2, QB2.1-100MB:
58.27GB, SR-221, SW-204
10.01GB, SR-206, SW-203

I will have all of this on the Drive and RAID DB later today. k

kaye 12-24-2003 09:28 AM

Drive and RAID DB is updated for the 74GB SATA Raptor and the 4x X15.3 Burly as well as 2x X15.3 Burly with UL3D and ATTO driver 3.0.0. The 2x X15.3 config showed a greater increase because it is not so close to the PCI bus limit of my MDD G4.

2x X15.3 with QB2.1-100MB:
3.0.0 driver, SR-141, SW-145
older driver, SR-136, SW-142 k

kaye 01-27-2004 12:02 PM

Drive and RAID DB is updated for data from Boots. G5 2GHz DP, SeriTek/1S2, two 74GB Raptors internal, booting from external FireWire, OS10.3.2, Apple DU, Apple RAID, or SoftRAID 3.0.2 which was actually a bit faster.

One thing Boots found out is that both Raptors striped on the mobo SATA with either Apple RAID or SR3.0.2 gave awful striped RAID sustained read results, less than single Raptor sustained read results. SeriTek/1S2 solved that. Last, his booting problems posted in the G5 forum, it appears turned out to be his video card and is a well known problem on the Apple forums and AppleCare is shipping a replacement card. k

kaye 01-31-2004 05:42 PM

More data from Boots for the Drive and RAID DB, this time at the top of SCSI with his sweet G5-2G DP, UL4D, and 2x X15.3 and 2x X15.2 drives in various configurations. k

kaye 02-08-2004 09:24 PM

Today, Rick and I tested my Black Burly with 4x X15.3 drives Apple RAID striped. I also brought along another Burly with 4x X15.2 drives Apple RAID striped.

A little background on how well these Burlys perform on my G4s. QuickBench 2-100MB test, 10.3.2.

MDD G4-1.25GHz DP (FW400), UL3D, ATTO driver 3.0.0
4x X15.3, 2-channels, SR-221 MB/s, SW-204

(I need to test with latest ATTO driver 3.0.3.)
---------------------------------------

QS/PF G4-1.2GHz DP, UL3D, ATTO driver 3.0.3
4x X15.3, 2-channels, SR-217 MB/s, SW-201

(Slightly slower, slower DP, slower Front Side Bus, slower memory, etc.)
---------------------------------------

QS/PF G4-1.2GHz DP, UL3D, ATTO driver 3.0.3
4x X15.2, 1-channel, SR-139 MB/s, SW-144

(Set up for the G5 on 1-channel but I wanted to see what it would do on one of my G4s.)
---------------------------------------
Now to Rick's machine:

G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x X15.3, 2-channels, SR-276 MB/s, SW-282

(Huge jump over my MDD results.)
---------------------------------------

G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x X15.3, 1-channel, SR-262 MB/s, SW-232

(Not what I would have expected, given the bandwidth of the slot and the capability of a single channel on the UL4D. No explanation.)
---------------------------------------

G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x X15.2, 1-channel, SR-145 MB/s, SW-145

(Slightly faster than this same test with my QS/PF but again, as just above, no explanation.)
---------------------------------------

G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x X15.3, 1-channel, and
4x X15.2, 1-channel, 8 drives total SR-285 MB/s, SW-286

(The X15.3 drives are far superior to the X15.2 drives which appear to drag down the results. However, this is still the fastest RAID I have ever tested as well as twice the size volume to go along with the X15 reputation for maintaining thruput throughout the volume.)
---------------------------------------
Last, an update to Rick's 4x AtlasIII 10k Maxtor Burly, previously only tested in OS9 and on his QS G4-867 SP and UL3D.

G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x AtlasIII 10k, 2-channels, SR-194 MB/s, SW-201
---------------------------------------

The more I think about the G5 UL4D four drive on a single channel results, the more I think the ATTO driver, still a work in progress, is not yet optimized for that many drives to fully take advantage of the huge capability of a single channel.

Will be updating the Drive and RAID DB tomorrow. k

Costa 02-09-2004 02:07 AM

Dear Kaye,

I am referring to your last post only (haven't gone round reading all the others yet, http://forums.macgurus.com/infopop/e...on_biggrin.gif ... I apologise).

<BLOCKQUOTE class="ip-ubbcode-quote"><font size="-1">quote:</font><HR> The more I think about the G5 UL4D four drive on a single channel results, the more I think the ATTO driver, still a work in progress, is not yet optimized for that many drives to fully take advantage of the huge capability of a single channel. <HR></BLOCKQUOTE>

Mmmm..... I don't think is only a matter of software.

In fact, a G5 PCI-X bus, running at 133MHz with a bandwidth of 64-bit, should return you a theoretical throughput of 1GB of data per second.

You should see easily throughputs over the 300MB/sec barrier with those layouts.
That's why I think it isn't JUST a software problem.

What do you think?

Costa

kaye 02-09-2004 09:56 AM

Hi Costa,

One thing about the G5s, I'm sure of nothing and I do appreciate your thoughts on this matter as well as TZ's.

For instance, why does stripe RAIDing two SATA drives on mobo SATA result in very poor benchmarks? Same SATA drives on one SeriTek/1S2 is a huge jump in performance regardless of which PCI-X slot. I have no idea.

Another for instance, why is striped RAIDed FW800 drives so poor no matter what rob-Art at Bare Feats does?

In my case with the UL3D and dual boot G4s, OSX thru Jaguar never approached 9.2.2 results, Now, 10.3 improved thruput as long as you had the correct ATTO driver at that point in time. Last test on my MDD I posted was the 3.0.0 driver with 10.3.2. I never saw the 3.0.1 driver. 3.0.2 and 10.3.2 was a fairly significant decrease in thruput and it only lasted about a week before ATTO responded with the 3.0.3 driver which was a nice increase in thruput over 3.0.0.

These results are with a max of two drives per channel on my UL3D. Either my QS/PF or Rick's G5 DP with UL4D, attach four drives on one channel and thruput decreases huge compared to two drives per channel. That is why I think ATTO's drivers need more tweaking for more than two drives per channel. You are right, it could be something else. I just suspect that the ATTO drivers are a work in progress. Of course in my case the limits or saturation of the UL3D channel with more than two X15.3 drives on one channel is what limits me. But in Rick's case... It certainly is not what I expected.

But then, who can argue about the results with two drives per channel or even the four drives per channel on his G5. Overall, marvelous improvement over my G4s.

I have not tried my UL3D in Rick's machine. It is not a universal power model. Just don't want to risk his G5 DP nor my UL3D. Big cloud of smoke would ruin both of our days but his more than mine. k

Costa 02-10-2004 06:24 AM

Hi Kaye,

Got the message. Truth is that I am striving to understand and SCSI is not my strong horse (still lots to lurn about it, especially LVD technology).

Beats me how an underlining hardware which "in theory" should boot your throughput enormously can't be fully exploited by modern SCSI technology.

Now, bear with me, would you? You use a lot of abbreviations which sometimes I have difficulties in interpreting (like your "...the RAM is a keeper." sentence: I had to ask Eric for a translation http://forums.macgurus.com/infopop/e...n_confused.gif ).
What you mean with "X15.3 drives"? You mean 15,000 rpms Cheetahs?

I expect both you and Rick are using Granite Digital's cables; correct? Terminators?

I dunno, but the whole experience you are going through is as if there's a bottleneck somewhere down the line; well .... this is my humble impression.

What is Jazzbo saying about it?

Costa

TZ 02-10-2004 07:00 AM

You think the military invented all those alphabet soup acrynms alone? http://forums.macgurus.com/infopop/e...on_biggrin.gif

X15.3 = Cheetah, 3rd generation. Everyone else (Maxtor Atlas) uses "15K" - it became so popular and easy with Cheetahs that Seagate went back and started using "10K.6" for their 6th generation 10K drive model.

ATTO SCSI FAQ
www.scsifaq.com
Hitachi SCSI Checklist
Guide to RAID - Storage Review
RAID Glossary - Adaptec The Full A-Z Glossary!
About SATA - Adaptec
SCSI Basics - Adaptec

kaye 02-10-2004 05:39 PM

Eleven new entries at the top of SCSI Drives in the Drive and RAID DB. k

kaye 02-17-2004 05:22 AM

This is an update to complete the UL3D tests with ATTO driver 3.0.3 in my two G4s. Tested were my Black Burly with 4x X15.3 drives Apple RAID striped and my other Burly with 4x X15.2 drives Apple RAID striped. Old and new tests for comparison. QuickBench 2.1-100MB test, 10.3.2. Both Burly RAIDs rebuilt and optimized directories and data. Results are pretty much at the limit of these G4s.

MDD G4-1.25GHz DP (FW400), UL3D, ATTO driver 3.0.3 NEW
4x X15.3, 2-channels, SR-222 MB/s, SW-203
---------------------------------------
MDD G4-1.25GHz DP (FW400), UL3D, ATTO driver 3.0.0 OLD
4x X15.3, 2-channels, SR-221 MB/s, SW-204
---------------------------------------
---------------------------------------
MDD G4-1.25GHz DP (FW400), UL3D, ATTO driver 3.0.3 NEW
4x X15.2, 2-channels, SR-220 MB/s, SW-203
(This one surprised me with the X15.2 drives)
---------------------------------------
(No OLD test for comparison in Jaguar or Panther)
---------------------------------------
---------------------------------------
QS/PF G4-1.2GHz DP, UL3D, ATTO driver 3.0.3 NEW
4x X15.3, 2-channels, SR-219 MB/s, SW-202
---------------------------------------
QS/PF G4-1.2GHz DP, UL3D, ATTO driver 3.0.3 OLD
4x X15.3, 2-channels, SR-217 MB/s, SW-201
---------------------------------------
---------------------------------------
QS/PF G4-1.2GHz DP, UL3D, ATTO driver 3.0.3 NEW
4x X15.2, 1-channel, SR-219 MB/s, SW-202
(Again surprised by these X15.2 drives)
---------------------------------------
(No OLD test for comparison in Jaguar or Panther)
---------------------------------------
---------------------------------------
As a reminder, on Rick's machine:
G5-2.0GHz DP, UL4D in PCI-X 133MHz slot, ATTO driver 3.0.3
4x X15.3, 2-channels, SR-276 MB/s, SW-282
---------------------------------------
I will be updating the Drive and RAID DB. k

kaye 02-20-2004 08:55 AM

Drive and RAID DB updated this morning. k

kaye 03-08-2004 06:08 PM

Drive and RAID DB updated this evening. Entries for Boots and TZ at the top of SCSI and unclemac at the top of FireWire. Gents, if you see any errors, please email me. If anyone sent me data that was not included in this update, please email me. I did not mean to leave anyone out. k


All times are GMT -7. The time now is 03:51 AM.

Powered by vBulletin® Version 3.8.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 2011 MacGurus
All Rights Reserved