Keep this site growing - Please visit my Sponsors

Accelerate Your Mac!  - the source for performance news and reviews
The Source for Mac Performance News and Reviews
Don't forget to check out all the other site features!

IDE RAID Performance Tests:
By Mike B. (additional tests by Joe Dylik)
Published: 3/6/2000

Performance Tests | Hardware Installation | Softraid Install


Update: For more current articles on IDE RAID, see the related links below. Later articles tested the IBM 75GXP IDE drives which as of summer 2000, are the fastest IDE drives to date. The 27GB Maxtor drives used in the article below always seemed to have low write speeds. Also note that shortly after this article, Acard (and now Sonnet) sells a hardware IDE RAID card for about twice the price that may perform better. The Acard and Tempo RAID cards have switches on the card for configuring 2 drives as a single volume without needing to use Softraid drivers. Apple's drive setup sees the 2 drives as a single disk with the switches set to RAID striping on the card. However even the hardware RAID card does not offer impressive performance (compared to a single drive) based on reports, when used in pre-G3 Macs which have slower memory/PCI controllers and system bus.

After first testing IDE RAID (2 drive) in the 9600/350 with a G4/350MHz CPU Upgrade, I was very disappointed in the overall performance from IDE RAID vs a single drive. A good, fast single drive performed about as well as two drives in IDE RAID in that system. Since I knew from past testing that the Kansas motherboard design (9600/300, 9600/350, 8600/250 and 8600/300 systems) had low memory/PCI performance, I wanted to test in a more modern Mac model before passing final judgement.

Different drives than the Maxtor 27GB (ATA/66, 7200 RPM, 2MB cache) models I used may show different results. From the write performance, it looks almost as if there was no write caching on the drive. The higher density (10GB/Platter) Maxtor Plus40 drive showed literally identical Read/Write performance in tests with the onboard B&W G3, but I didn't have two of them to test for IDE RAID.

As you'll see below, even in the latest Apple G4/AGP system (as of March 2000), IDE RAID performance doesn't seem worth the extra expense of the RAID software and PCI card, especially if you have an ATA/33 or ATA/66 onboard IDE controller already. (Note that Joe's tests with two IDE controllers showed higher performance however, but that adds additional costs and uses up another PCI slot. Although the Seagate drives he used delivered more balanced Read/Write performance, their peak rates seemed lower than the Maxtor's in many cases.)

In the tests I ran, there was often little difference in single drive vs dual drive (RAID) performance, at least with the Maxtor 27GB (6800 series) 7200 RPM/2MB cache drives I used for these tests. Other drives may show different results as these seem to have much lower write performance than reads.

Listed first are my own real world and benchmark results, but below you'll find Joe Dylik's additional benchmark tests of ATA/33 vs ATA/66 PCI IDE cards using Seagate Barracuda IDE drives. He tested up to 4 drive RAID with single and dual IDE cards as well as single drive (non-RAID) tests. The results will surprise you I think. (Even Joe's results with dual cards is often baffling, as some combinations showed better performance with 2 drives than 4.)

This article includes links to my previously published guide to installing a dual drive RAID in a B&W G3 or G4 system. (See the Driveinstall and Softraid install pages linked at the top and bottom of this article).


PCI IDE Controllers Used:

  • Promax TurboMax ATA/33 card (using v1.5 firmware from Spring 1999)
  • Acard ATA/66 card using latest firmware update as of 2/15/2000


Test System:

  • Apple G4/450 AGP rev 1
  • OS 9, VM OFF, 256MB 222 SDRAM
  • ATI Rage128 AGP Graphics card (OpenGL 1.1.2/ATI 4.2 drivers)
  • Stock 20GB WD Expert (master) & 18GB WD Expert (slave) ATA/66 drives
    using onboard ATA/66 controller (both are 7200 RPM/2MB cache models)
  • OEM ZIP and DVD ROM drive
  • All the latest OS 9 updates (QT 4.1, Audio Update 1.3)
  • Multiple Users Control Panel/Extension disabled

All PCI IDE cards were tested in the middle PCI slot with ATA/66 cables. All disks were formatted in HFS+ (MacOS Extended). No boot time tests were run since striped RAID is not bootable at the current time in the new openfirmware Macs.

All RAID tests I ran used Softraid v2.2.1. This driver was also used in single drive tests with the PCI IDE controllers to compare Apple's OS 9 driver (Drive Setup v1.8.1) to Softraid's.


IDE RAID Test Results:

The following graphs show the results of tests with each of the cards and with various RAID software. Included are results with the G4's original Western Digital 20GB Expert drive connected to the onboard ATA/66 interface. You'll likely be surprised and disappointed in the results. For value, a single Plus40 (high density/high performance) Maxtor drive tested in a B&W G3 (ATA/33 onboard) delivered better sustained rates than any of the configurations shown in this article. (See the related links below for that review if you missed it.)


Real World Test Results:

The graphs below show the results of the following tests:

  1. Copying a 940MB Folder (2,551 files) from the onboard ATA/66 WD Expert 18GB slave drive. All disks were blank (empty) and freshly formatted for these tests.

  2. Duplicate the 940MB folder (eliminating the issue of the source drive as a factor)

  3. Time to load a 18MB Photoshop 5.5 file and then save it as a different name.

Since a stopwatch was used to time these tests, remember there is some human error in starting and stopping the stopwatch so consider times within 1/2 second or so to be literally identical.

920MB Folder copy and dupl tests

Fastest: The onboard IDE with OEM WD Expert drive.

Photoshop Load/Save 18MB File Results

Fastest: The onboard IDE with OEM WD Expert drive.


ATTO Tools Benchmark Results: The following are graphs of the peak and sustained read/write results from the latest ATTO Tools v2.3B Benchmark.

Peak Read and Write results

Fastest: The onboard IDE with OEM WD Expert drive.

The Acard ATA/66 had a hair faster Read Peak Rate, but far less write performance, perhaps due to the Maxtor 27GB drive apparent lack of write caching.)

Update: After updating the Turbomax to firmware v1.7.4, Peak Rates in RAID were 30.16MB/sec Read and 23.73MB/sec Writes. This is a few MB/sec higher in R/W than with the v1.5 firmware.

Sustained  Read and Write results

Fastest: Writes - The onboard IDE with OEM WD Expert drive.
Reads - Acard ATA/66 w/Maxtor 27GB (by a nose).

I didn't notice it until graphing the results at about 2AM, but the lower Turbomax RAID results compared to the same card with a single drive is odd. Due to the late hour I did not retest. And that's not a typo, the OEM WD Expert 20GB had exactly the same Read and Write rates (22.88 MB/sec).

Note how very little gain is seen from IDE RAID vs a single drive.

Update: After updating the Turbomax to firmware v1.7.4, Sustained Rates in RAID were 24.45MB/sec Read and 14.52 MB/sec Writes. (The firmware update provide a larger gain in peak rates than in sustained.) There does seem to be a 'hump' in the ATTO tools write graph that appears to indicate the drive's write cache is enabled although sustained rates are still far lower in writes than reads.


MacBench 5.0 Results:

I ran both the standard Disk/Publishing Disk tests as well as the Random/Sequential tests. The PCI IDE single drive tests were run both with Apple OS 9's Drive Setup 1.8.1 drivers and with Softraid's drivers (denoted by 'SR'). RAID tests (dual drive stripe) used Softraid only of course. All disks were empty (freshly formatted) before the test.

Macbench 5 Disk and Pub Disk tests

Fastest: The Acard ATA/66 & Maxtor 27GB
(Practically a draw with the onboard IDE/WD Expert as scores can vary a few points run/run)

Update: After updating the Turbomax to firmware v1.7.4, MacBench RAID Disk score was 2277, Pub Disk score was 1636.


Next I ran MacBench's Random and Sequential R/W disk inspection tests:

Macbench 5 Disk Random/Sequential tests

Fastest (Overall): The Acard ATA/66 IDE RAID (dual Maxtor 27GB).

Notice only the Random Write scores really favored the IDE RAID (w/Acard ATA/66) in these tests.


Other RAID Test Results by Joe Dylik:

Joe Dylik sent the results of ATTO Tools Benmark tests on an IDE RAID setup using up to 4 Seagate Barracuda IDE drives, with both one and two PCI IDE cards. Tests were run with both Softraid and FWB drivers (and in the case of single drive tests, ATTO's drivers as well). The results are very surprising. I'm not sure the ACard ATA/66 card had the most recent firmware update (released about the time Joe sent these results), but regardless his results are very interesting.

I asked Joe how he mounted the 4 IDE drives in the case:

"Mike,
I used the ATA mounting brackets from two other G4,s and mounted the disks across the bottom of the case, attaching the disks to the ATA mounting brackets, then the brackets slid into the slots at the back of the sleds on the bottom of the case. It was a tight fit, but I was able to get them in.

I also added an additional fan above the two disks mounted at the front of the case, as they seemed to be running hotter.

The disks next to the ATA controller ran cool, the middle pair of disks ran warm, and the disks furthest from the circulation fan ran hot.

The original tests were done using the full 18" ATA/66 cables, the excess cabling in the case may have also promoted the heat build up.

The cables will be trimmed to about 12" to minimize the potential heat build up because of excess ribbon cabling blocking the circulation in the case.
Joe"


Joe's Test System:

  • Apple G4/400 AGP
  • OS 9, VM OFF, 512MB SDRAM
  • ATI Rage128 AGP Graphics card (OpenGL 1.1.2/ATI 4.2 drivers)
  • OEM 27.3 GB IBM Deskstar DPTA-372730 (7200RPM/2MB cache)
  • The additional disks used in the tests are all Seagate Barracuda ST328040A ver.307 or ver.311, 28.5 GB, 7200rpm, 512KB cache models.

He said he used the latest firmware available for the Acard and Turbomax cards. (My Turbomax card had an old v1.5 (spring 1999) ROM.)

Note: Where you see 'disk position' info, slot B/C means what position the card was in the Mac, slot A being the graphics card slot, B the next one over, etc.. Interesting to note that the PCI slot affected rates more than I'd have expected in some combinations.

All the Barracudas with the exception of the one on the Apple on board controller were set to Master position. The labeling B,0 and B,2 refer to the card position in the machine and the channel on the card.

Note in the table below that using two IDE controllers dramatically increases sustained performance. Also note the odd cases where 2 drives outperformed 4 drives with some driver/card combinations. (Compare the single card/single drive results farther down - some combinations of single card/RAID delivered slightly lower performance than one card w/one drive.)

IDE RAID Table of Results


Joe also sent results of single drive tests of the Acard ATA/66 vs Turbomax. Note the more balanced Read/Write rates (but lower peak reads) than the Maxtor 6800 series drives I used. (Also compare these single card/drive results to the single card/RAID results above - with one card the RAID results were lower in some cases.)

(This table is wide so you may have to stretch or scroll your browser.)

IDE PCI Card Results (single drives)

Joe's Comments on his System and Results: Here's Joe's original mail titled 'Is the ATA/66 standard All Smoke and Mirrors?'

"Hi Mike,
Recently, I acquired a couple Acard AEC6260M ATA/66 Macintosh PCI adapter cards from Blackfire that are to be installed into a new file server that I am working on, and being that I already have a couple Promax ATA/33 PCI adapters cards, I felt a head to head test was in order.

The test machine is a G4 400 MHz with the on board Apple ATA/66 controller. It is running Mac OS system 9 and has 512 MB ram installed. The OEM hard disk is a 27.3 GB, IBM Deskstar DPTA-372730, 7200rpm, 2MB cache model. The additional disks used in the tests are all Seagate Barracuda ST328040A ver.307 or ver.311, 28.5 GB, 7200rpm, 512KB cache models.

Both of the ATA adapter cards tested (Promax and Acard) have the most current ROM installed.

All the cables used for the tests were ATA/66, 80 wire, 40 pin cables that conform to ANSI standards, with the exception of the noted tests where ATA Ė ATA/33, 40 wire, 40 pin cables were substituted to do additional performance comparisons.

When the software drivers on the hard disks were changed for testing, the system was re-started before each new benchmark performance series was run.

The test results and an additional page of 'snippets' off of different web sites are attached.

From a non-technical users point of view, the Promax is the easier of the two cards to install and configure.

When I received them a few months ago, I popped them into a Mac, connected a couple unformatted drives to the cards, and away I went, not a lick of problems.

The Acard adapters that were shipped to me from Australia had the most current flashed ROM update installed before they were released to me, but they still were problematic; when I first inserted them into the Mac with unformatted drives on them, the system would not boot, displaying symptoms similar to a SCSI conflict. I had to disconnect the hard drives, and flash a more current ROM update onto them individually before I could use the adapters.

Both cards seem to be very close in disk speed performance. The Acard controller seems to have a more evenly matched Read /Write peek and sustained speeds. The Promax exhibits slower Write than Read speeds across the various tests performed.

In Raid array tests, the Promax, FWB driver grouping, seems to be the way to go with a single card dual drive arrangement if your looking to gain a little more speed than achievable with individually formatted volumes.

In the dual card dual disk Raid array tests, the Acard, SoftRaid combination seems to be the best performer.

There is very little performance gain when setting up a Raid array as a four disks two-controller card arrangement as opposed to the dual card dual disk setup. Neither card is a dual channel card, though they both can handle two pairs of Master/Slave disks. (I didn't test ATTO's Raid software; it's a little too pricey for me to upgrade my older version, and I'm not sure of its compatibility with Mac OS 9.)

The thing that amazes me is the relatively matched performance of the two cards, even though one of them is supposed to be capable of double the signal speed when moving data across the card.

The real eye opener is when I changed the cables out to perform additional tests.

I could not record any discernable performance differences when using ATA 40-wire 40-pin cables on the drives as opposed to the required ATA/66 80-wire 40-pin cables. (See the attached ATAsnippets document.) [ I also saw no difference last year with either cable type in a B&W G3 (which has a ATA/33 controller however), noted in the B&W G3 IDE slave drive article.-Mike] I tested all the controllers (Apple on board, Promax, and Acard) with both disks and the results were similar. (Only the Apple on board controller test with the ATA cables switched is included in the test results.)

When I change the data modes on the hard disks from mode 4 ATA/66 to mode 3 ATA/33 the performance of the drives were almost identical.

Based upon all the information I can find about the ATA/33 and ATA/66 standards, THERE SHOULD BE SOME NOTICEABLE PERFORMANCE DIFFERENCES IN DATA TRANSFER SPEEDS WHEN ALL THE ATA/66 STANDARDS ARE NOT MET ON THE HOST SYSTEM.

It appears to me that the primary component to the ATA/66 standard is the 80-wire 40-pin cable whose presence is supposed to be detectable by the controller and the drive. The additional 40 ground wires in the ATA/66 cable is what is responsible for the increased signal speed across the wires that make the ATA/66 standard of data transfer rates achievable.

What gives???
Are we all being duped???
Is the ATA/66 standard all "Smoke and Mirrors"???

I called Segate tech support about my cable swapping findings; they didnít seem to have an answer. They thought maybe it might be the Mode setting on the drive though the ST328040A Barracudas supposedly shipped with Mode 4 enabled. (Some manufactures ship ATA/66 drives with Mode 4 disabled because of Bios problems with some of the older computers they are installed into. On the batch of five Barracudas that I tested, one of them shipped in Mode 3, the ATA/33 standard.) They also said that their Mac user base was very small and that they didnít have very much experience with Macs or their ROM.

I also called Apple tech support and discussed the cable swapping tests with them in regards to the on board ATA/66 controller. An incident number was opened. They havenít gotten back to me about this yet.

On to a different topicÖ

The reason for performing all of these tests was to establish disk transfer speeds for a new file server that I am setting up. Based upon my findings the G4 equipped with the Seagate Barracuda ATA drives, performs equal to and in many instances better than our G3 file server equipped with 5 external dual drive dual channel Ultra SCSI II Raid arrays both in internal and network data transfer speeds under various tested loads within our workgroup.

The G3 server is being phased out because of storage space issues. With its ten external drives it only gives us a combined 40 GB of space; not to mention the previous and potential SCSI nightmare of extending the SCSI cabling to its max limits. (The drives were acquired over a period of years while we were using MediaNet FDDI networking protocol, and were later re-purposed into one central file server as opposed to five workstation/fileserver setups commonly seen with a MediaNet arrangement.)

The costs associated with upgrading the SCSI disks on the G3 server compared to what it cost to setup the new G4 are amazing. The two Acard adapter cards and cables cost around $320. The five Segate Barracudas were purchased for about a $1000. The total combined storage space of the new G4 server is just shy of 170 GB. How much SCSI storage space can you buy for $1400??? What would 170 GB of SCSI storage cost???

In case your wondering, the power draw of the G4 server equipped with the extra drives and adapter cards is well within the capacity of the power supply unit. The power draw tests out to about a 120 to 140 watt load while running, though I did have to add an additional cooling fan mounted above the 2 drives towards the front of the case because those two drives seemed to get much hotter than the ones mounted at the opposite end of the case under the circulation fan. Also, the cables needed to be a shorter length than the standard 18-inch ATA/66 cables, thereís just not enough space for excess ribbon cabling in the case.

Thanks once again for maintaining such an informative web site.
Sincerely,
Joe Dylik
Computer Administrator
Centermark Graphics"

See the G4 onboard IDE tests page linked below for Joe's test results comparing the OEM IBM IDE drive vs. a Seagate Barracuda and the results of ATA/66 cable tests.


Performance Tests | Hardware Installation | Softraid Install

Back to XLR8YOURMAC.COM

Your Source for the best in CPU/SCSI/VIDEO card reviews, daily news, and more! 


Copyright © 2000.

No part of this site's original content is to be reproduced in any form without permission.
All brand or product names mentioned here are properties of their respective companies.

Disclaimer: Users must read and are bound by the Site Terms & Conditions of Use.