Reviews Leaderboard Database Reference Search StorageReview Discussion Reliability Survey Search About StorageReview.com Contents

Promise FastTrak ATA RAID Controller

  April 14, 1998 Author: Eugene Ra  
See also Living with the Promise FastTrak
and The Promise FastTrak, Caviar Style
and Western Digital Caviar AC36400
and A Look at Windows NT's Software Striping

Whenever I hear the term "RAID," I always conjure up images of big, expensive servers housing lots of fast, expensive disks with an advanced, expensive controller . RAID originally stood for "Redundant Array of Inexpensive Disks", although these days a more politically-correct "Independent" is used in the place of "Inexpensive." The fact was that these systems were always SCSI-based, and therefore incredibly expensive. It always remained out of reach of most individuals, used only where the utmost in performance and/or data security was needed.

Since its inception, Promise Technology has always been a leader in the ATA arena. Last summer, they released an ATA RAID controller. Their rationale? The company felt that much of the performance benefits of a RAID could be brought to end-users at a significantly lower cost using ATA rather than SCSI technology. Using a single IRQ, a FastTrak controller can support up to four drives in varying configurations. Further, the controller may be used in conjunction with another to control up to 8 drives, not to mention the fact that they can coexist with the onboard ATA controllers found on virtually all motherboards. The FastTrak supports RAID levels 0, 1, and 0/1.

RAID Level 0, AKA "data striping," uses two or more preferably identical drives interleaving operations (reads or writes) between them. This allows the drives in the array to be seen as one huge linked drive. In a two drive array, for example, the drives are formatted so the first, say, 8k is placed on drive 1, the second 8k on drive 2, the third on drive 1, fourth on drive 2, and so on. Since the drives are sharing the load of reading and writing, they can theoretically provide the sum of their individual transfer rates to greatly improve performance. The disadvantage, however, is that the failure of a single drive results in the loss of data across the entire array. Since performance and capacity are limited by the slowest/smallest drive, using identical drives is favored for efficiency's sake. A single FastTrak controller can support up to four disks in a data striped array.

The FastTrak also supports RAID Level 1, AKA "mirroring." This setup takes a pair of drives and writes identical data to both, thus allowing fault tolerance. In the event that one drive fails, the other has all recorded data still up and running. The controller also allows another drive to be installed and ready as a spare, to re-mirror the functioning drive upon the next reboot. Data read speed of a mirrored pair is enhanced through "elevator sorting" and "load balancing"- whichever drive is closer to the requested data, through either less necessary head positioning or rotation, will be used for the read. The total capacity of a mirrored pair is the same as that of the capacity of the lesser drive.

Finally, RAID Level 0/1 is supported. This is basically the mirroring of a striped pair, allowing the user to gain the speed benefit of striping while still maintaining the protection of data redundancy. Array capacity is twice the size of the smallest drive.

As a side note, the FastTrak also supports simple "spanning," where all drives are linked to be logically treated as one. The controller proceeds to write data to the first drive. When full, the next is used. There aren't any performance or protection features in this configuration, though. It allows for maximum space efficiency in configurations where all drives aren't identical in capacity (if they were, data striping would be a better choice in all ways).

The included manual is well done and easy to follow. Chapter one is basically an advertisement describing all the advanced features that your cool new ATA controller supports . Chapter two is a quick installation guide while Chapter three gets into the installation in greater detail. The fourth section describes the various options available in the FastTrak BIOS while Chapter 5 provides an introduction to RAID concepts. Finally, Chapters 6 and 7 offer detailed operating system specific installation and troubleshooting. The two year limited warranty is outlined in the appendix.

Hardware installation went very smoothly. The card is about as small as a PCI card with dual ATA channels can be- Its miniscule! I went ahead and installed it into a free PCI slot, attached two Maxtor 5.7 gig DiamondMax 2880s (chosen for the excellent individual performance at a low price) with the included 18" cables as the master of each FastTrak channel. I left both of the LX6 motherboard's ATA channels enabled, the primary being used for the Western Digital boot drive while the secondary remained connected to the ATA Toshiba XM-6202B CD-ROM. The PCI plug-and-play card chose its own interrupt upon booting. I pressed F2 to enter the controller's BIOS.

The FastTrak BIOS is well done and very easy to get around in. It includes an auto-configuration utility to automatically setup an array optimized for performance (data striping), fault tolerance (mirroring), or capacity (spanning). There's also several options to create the arrays manually (not much more complicated than the auto setup), view them, delete them, etc. When creating an array involving striping, the controller allows the user to set the block size anywhere from 1k to 1024k. There's a trade-off involved here- smaller block sizes optimize seek time while larger sizes allow better sequential transfer rate. Promise recommends 8k stripe blocks for general purpose use, and 64k blocks for video editing. I selected 8k blocks for stripe testing purposes.

I partitioned the array into a single large 11.5 gig unit. The FastTrak's BIOS supports massive volumes well beyond the 8.4 gig limit imposed by some systems that don't implement INT13h extensions. Windows 95 plug-and-play detected the card upon bootup, asking me to insert the included driver disk to fully identify the "PCI RAID Controller" that was detected. The system properties entry for the FastTrak is very bare, no DMA box to check to support busmastering, for example. A simple drive format later and I was ready to move onto benchmark testing.

I received some strange results during my first WinBench 98 run. As it turns out, the documentation speaks of supporting speeds up to 12 MB/sec for a two-drive array. A single DiamondMax 2880 can exceed this transfer rate on its outer tracks. I managed to fix the problem though by downloading the latest drivers from Promise's FTP site. This allowed the array to stretch its legs and achieve the promised (hehe) transfer rates. The averages of five trials from ZD's WinBench 98 and Adaptec's ThreadMark 2.0 under Windows 95 are presented below. Data from tests on a single Maxtor 5.7GB disk driven from the PIIX4 controller on the Abit LX6 is provided for comparison.

Test Bed - Hardware
Motherboard Abit LX6, v1.1, Bios vC7Q
Processor Intel Pentium II 266MHz, 512k cache
Memory 64MB 10ns SDRAM DIMM
Boot Drive Western Digital AC31600
ATA RAID Controller Card Promise FastTrak ATA RAID Controller
Video Card Matrox Millennium II PCI, 4 MB
Test Bed - Operating System Configurations
Operating System Windows 95, OSR 2.1 Windows NT Workstation 4.0
File System FAT 32 NTFS
Patches Applied 82371xB INF Update
REMIDEUP.EXE fix
Service Pack 3
IDEFIX-I.EXE
Display Driver MGA PowerDesk v3.80
1024x768, 24-bit color, 85 Hz, Small Fonts
MGA PowerDesk v3.31
1024x768, 24-bit color, 85 Hz, Small Fonts
Ziff Davis WinBench 98
- Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 0 (striped) Configuration
Windows 95 OSR 2.1, FAT 32
Business Disk WinMark 98 1406 KB/sec
SS/Database 1344 KB/sec
WP 1682 KB/sec
Publishing 1266 KB/sec
Browsers 1458 KB/sec
Task Switching 2788 KB/sec
High-End Disk WinMark 98 4166 KB/sec
AVS/Express 3.1 2422 KB/sec
Frontpage 97 3428 KB/sec
Microstation 95 7062 KB/sec
Photoshop 4.0 4320 KB/sec
Premiere 4.2 7816 KB/sec
PV-Wave 6.1 2698 KB/sec
Visual C++ 5.0 9632 KB/sec
Disk/Read Random Access 17.1 ms
Disk/Read Transfer Rate
Beginning 23340 KB/sec
End 16200 KB/sec
Disk/Read CPU Utilization 17.82%
Transfer Rate 19889 KB/sec
KB/sec per 1% CPU Utilization 1116 KB/sec
Ziff Davis WinBench 98
- Abit LX6, 82440LX, PIIX4 "Southbridge"
Single Maxtor DiamondMax 2880 90576D4 drive
 
Windows 95 OSR 2.1, FAT 32
Business Disk WinMark 98 1426 KB/sec
SS/Database 1372 KB/sec
WP 1738 KB/sec
Publishing 1252 KB/sec
Browsers 1508 KB/sec
Task Switching 2664 KB/sec
High-End Disk WinMark 98 4194 KB/sec
AVS/Express 3.1 2478 KB/sec
Frontpage 97 3396 KB/sec
Microstation 95 7232 KB/sec
Photoshop 4.0 3896 KB/sec
Premiere 4.2 8054 KB/sec
PV-Wave 6.1 2934 KB/sec
Visual C++ 5.0 9076 KB/sec
Disk/Read Random Access 15.9 ms
Disk/Read Transfer Rate
Beginning 13400 KB/sec
End 8260 KB/sec
Disk/Read CPU Utilization 5.85%
Transfer Rate 13379 KB/sec
KB/sec per 1% CPU Utilization 2287 KB/sec
Adaptec ThreadMark 2.0
- Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 0 (striped) Configuration
Windows 95 OSR 2.1, FAT 32
Data Transfer Rate 11.21 MB/sec
Average CPU Utilization 36.62%
KB/sec per 1% CPU Utilization 306 KB/sec
Adaptec ThreadMark 2.0 - Abit LX6, 82440LX, PIIX4 "Southbridge"
Single Maxtor DiamondMax 2880 90576D4 drive
 
Windows 95 OSR 2.1, FAT 32
Data Transfer Rate 6.46 MB/sec
Average CPU Utilization 24.41%
KB/sec per 1% CPU Utilization 265 KB/sec

Surprisingly, WinBench 98's high-level WinMarks shows no gain when comparing dual striped Maxtor's to a single drive. Promise's own packaging displays the older WinBench 97 reporting a 20% gain going from a single ATA to dual ATA drives. Significant changes appear in the low-level tests, however. The first thing that jumps out is the 16-23 MB/sec transfer rate. Sustained throughput seems to have risen to UltraATA levels (i.e., beyond 16 MB/sec) when using two drives as a striped pair, displaying a 75% gain on the outer tracks. On the downside, however, random access times seem to have risen about 12% due to the fact that both drives must get their heads into position before data can be read. Note also the CPU utilization of the array; even taking the high sequential transfer rate (STR) into account, CPU utilization is rather high.

ThreadMark results, on the other hand, soared. A dual drive array with the FastTrak posted a whopping 73% gain from a single drive configuration, the first double digit ThreadMark transfer rate score we've seen here at the Storage Review. ThreadMark's reported CPU utilization is also much more in line, actually being slightly lower per MB transferred than a single drive.

Here's the average of five trials of each benchmark under Windows NT:

Ziff Davis WinBench 98
- Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 0 (striped) Configuration
Windows NT 4.0, NTFS
Business Disk WinMark 98 1894 KB/sec
SS/Database 1748 KB/sec
WP 2106 KB/sec
Publishing 1760 KB/sec
Browsers 2344 KB/sec
Task Switching 3160 KB/sec
High-End Disk WinMark 98 4884 KB/sec
AVS/Express 3.1 2674 KB/sec
Frontpage 97 4188 KB/sec
Microstation 95 9076 KB/sec
Photoshop 4.0 4954 KB/sec
Premiere 4.2 8644 KB/sec
PV-Wave 6.1 3352 KB/sec
Visual C++ 5.0 10740 KB/sec
Disk/Read Random Access 9.4 ms
Disk/Read Transfer Rate
Beginning 26580 KB/sec
End 16100 KB/sec
Disk/Read CPU Utilization 40.20%
Transfer Rate 26615 KB/sec
KB/sec per 1% CPU Utilization 662 KB/sec
Ziff Davis WinBench 98
- Abit LX6, 82440LX, PIIX4 "Southbridge"
Single Maxtor DiamondMax 2880 90576D4 drive
 
Windows NT 4.0, NTFS
Business Disk WinMark 98 1844 KB/sec
SS/Database 1656 KB/sec
WP 1966 KB/sec
Publishing 1800 KB/sec
Browsers 2462 KB/sec
Task Switching 2670 KB/sec
High-End Disk WinMark 98 4566 KB/sec
AVS/Express 3.1 2368 KB/sec
Frontpage 97 3974 KB/sec
Microstation 95 9300 KB/sec
Photoshop 4.0 3912 KB/sec
Premiere 4.2 8338 KB/sec
PV-Wave 6.1 3736 KB/sec
Visual C++ 5.0 9396 KB/sec
Disk/Read Random Access 15.2 ms
Disk/Read Transfer Rate
Beginning 13300 KB/sec
End 8146 KB/sec
Disk/Read CPU Utilization 5.06%
Transfer Rate 13298 KB/sec
KB/sec per 1% CPU Utilization 2628 KB/sec
Adaptec ThreadMark 2.0 - Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 0 (striped) Configuration
Windows NT 4.0, NTFS
Data Transfer Rate 12.55 MB/sec
Average CPU Utilization 32.79%
KB/sec per 1% CPU Utilization 383 KB/sec
Adaptec ThreadMark 2.0 - Abit LX6, 82440LX, PIIX4 "Southbridge"
Single Maxtor DiamondMax 2880 90576D4 drive
 
Windows NT 4.0, NTFS
Data Transfer Rate 8.20 MB/sec
Average CPU Utilization 14.34%
KB/sec per 1% CPU Utilization 572 KB/sec

In Winbench 98 under Windows NT, the striped pair posted slight gains of 2-7% in the Disk WinMark tests. STR was even more astonishing, breaking 25 MB/sec even taking into account the fact that decimal bytes (1000 bytes per kilo) rather than binary (1024 bytes per kilo) bytes are being used. access time under NT, strangely enough, dropped far below that of a single drive. The drawback here, however, is the enormous CPU utilization - 4 times higher per megabyte than a single disk.

ThreadMark results also pushed into new territory, at about 12.5 MB/sec. Note here, though, the unusually high CPU utilization per MB transferred. Here too the results are disproportionately high, about twice the levels posted by an individual drive. Perhaps Promise has not done as good a job of implementing busmastering under NT as it should have.

I also went ahead and tested the FastTrak with the two Maxtors in a mirrored rather than striped pair. The results of five trials are presented below:

Test Bed - Hardware
Motherboard Abit LX6, v1.1, Bios vC7Q
Processor Intel Pentium II 266MHz, 512k cache
Memory 64MB 10ns SDRAM DIMM
Boot Drive Western Digital AC31600
ATA RAID Controller Card Promise FastTrak ATA RAID Controller
Video Card Matrox Millennium II PCI, 4 MB
Test Bed - Operating System Configurations
Operating System Windows 95, OSR 2.1 Windows NT Workstation 4.0
File System FAT 32 NTFS
Patches Applied 82371xB INF Update
REMIDEUP.EXE fix
Service Pack 3
IDEFIX-I.EXE
Display Driver MGA PowerDesk v3.80
1024x768, 24-bit color, 85 Hz, Small Fonts
MGA PowerDesk v3.31
1024x768, 24-bit color, 85 Hz, Small Fonts
Ziff Davis WinBench 98
- Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 1 (mirrored) Configuration
Windows 95 OSR 2.1, FAT 32
Business Disk WinMark 98 1488 KB/sec
SS/Database 1398 KB/sec
WP 1784 KB/sec
Publishing 1318 KB/sec
Browsers 1666 KB/sec
Task Switching 2656 KB/sec
High-End Disk WinMark 98 4156 KB/sec
AVS/Express 3.1 2506 KB/sec
Frontpage 97 3410 KB/sec
Microstation 95 7236 KB/sec
Photoshop 4.0 3740 KB/sec
Premiere 4.2 8284 KB/sec
PV-Wave 6.1 2790 KB/sec
Visual C++ 5.0 9322 KB/sec
Disk/Read Random Access 14.0 ms
Disk/Read Transfer Rate
Beginning 13400 KB/sec
End 8210 KB/sec
Disk/Read CPU Utilization 7.71%
Transfer Rate 13357 KB/sec
KB/sec per 1% CPU Utilization 1732 KB/sec
Ziff Davis WinBench 98
- Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 1 (mirrored) Configuration
Windows NT 4.0, NTFS
Business Disk WinMark 98 1924 KB/sec
SS/Database 1702 KB/sec
WP 2086 KB/sec
Publishing 1886 KB/sec
Browsers 2486 KB/sec
Task Switching 2676 KB/sec
High-End Disk WinMark 98 4856 KB/sec
AVS/Express 3.1 2754 KB/sec
Frontpage 97 4096 KB/sec
Microstation 95 9696 KB/sec
Photoshop 4.0 3820 KB/sec
Premiere 4.2 9086 KB/sec
PV-Wave 6.1 3898 KB/sec
Visual C++ 5.0 9868 KB/sec
Disk/Read Random Access 8.0 ms
Disk/Read Transfer Rate
Beginning 13400 KB/sec
End 8442 KB/sec
Disk/Read CPU Utilization 16.66%
Transfer Rate 13354 KB/sec
KB/sec per 1% CPU Utilization 802 KB/sec
Adaptec ThreadMark 2.0 - Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 1 (mirrored) Configuration
Windows 95 OSR 2.1, FAT 32
Data Transfer Rate 11.84 MB/sec
Average CPU Utilization 43.63%
KB/sec per 1% CPU Utilization 271 KB/sec
Adaptec ThreadMark 2.0 - Promise FastTrak ATA RAID Controller
w/ Dual Maxtor DiamondMax 2880 90576D4 drives
RAID 1 (mirrored) Configuration
Windows NT 4.0, NTFS
Data Transfer Rate 13.69 MB/sec
Average CPU Utilization 35.37%
KB/sec per 1% CPU Utilization 387 KB/sec

Here's where things start to get strange: Under WinBench 98 in both Windows 95 and NT, the mirrored pair posted slightly higher Business WinMark scores than the striped pair. If anything I would expect it to be the other way around! High-End WinMark scores remained virtually the same. The low-level tests followed predictions- lower access times and lower transfer rates. Note the significantly lower access time under NT that the mirrored pair provides over a single drive.

ThreadMark results in both 95 and NT were consistently higher with a mirrored configuration than a striped pair. This combined with the high ThreadMark results from the striped configuration make it impossible to credit the high results to either low access times or high sequential transfer rates. Suspicious, I decided to run the FastTrak with a single drive and see what kind of ThreadMark result I came up with . I didn't put it through the rigorous 5 trials that we normally use- one run would tell me what I needed to know.

Adaptec ThreadMark 2.0
- Abit LX6, 82440LX, PIIX4 "Southbridge"
Single Maxtor DiamondMax 2880 90576D4 drive
Windows 95 OSR 2.1, FAT 32
Data Transfer Rate 7.01 MB/sec
Average CPU Utilization 27.79%
KB/sec per 1% CPU Utilization 252 KB/sec

Yes, I was trying to see if a single drive would still score in the 11-12 MB/sec region. It didn't though. Note that 7 MB/sec is slightly faster than a single drive running off of the PIIX4 with around the same per MB CPU utilization to boot.

Let's turn to some aesthetic considerations. Two Maxtor drives running in tandem are about 3dBs louder, i.e, twice as noisy as a single unit. Since humans hear acoustic energy in a logarithmic rather than linear fashion though, it doesn't -sound- twice as loud. The pair is noticably louder than a single drive, but still quieter in seeking than most SCSI units. Though I didn't use a drive cooling fan unit with either disk, both became only slightly warm. Virtually anyone should be able to run two of these next to each other with no problem at all.

I'm somewhat at a loss for a firm conclusion regarding the FastTrak. WinBench low-level tests reported a dramatic increase in STR for a pair of mirrored ATA drives while ThreadMark results seconded the impressive performance increase. The fact that WinBench's high-level tests didn't confirm the increase along with the -even higher- ThreadMark scores posted by the mirrored pair make me wonder how much of this additional performance is real and can be noticed.

Its probably safe to say that those looking to do A/V editing on the cheap will find the FastTrak to be a compelling alternative to a stratospherically-priced SCSI subsystem. With today's 10+ gig ATA drives becoming available, capacity of a fully-decked out FastTrak system (say, 8 11.5 gig Maxtors on two controllers?) is staggering considering the low price one would pay. Sequential transfer rate, all-important in such applications, undeniably benefits enormously from an ATA RAID.

I'm a little more skeptical for uses involving mirroring and data redundancy. Yes, ThreadMark posts enormous gains even with drives in a mirrored configuration, and you do get true fault tolerance. Lets use two 9.1 gig 7200rpm Medalist Pro drives as an example. Two of these units plus the FastTrak will set you back over $900, just a little less than what the Barracuda 9LP and Adaptec 2940UW would cost. You'd have the same amount of capacity in either configuration. There would be more noise with the Barracuda. There would be more heat with the Medalist Pro array. Performance wise, the Medalist Pros would probably post higher ThreadMark results while the Barracuda delivers a more balanced score set. An Ultra-Wide SCSI controller scales beautifully off of a single IRQ to provide 15 devices. A two-controller Fastrack system can handle up to 8 devices, but that setup probably uses two IRQs (I'm skeptical about the PCI IRQ sharing) along with four cables all having to be 18 inches or less. It can be a mess. Ah, but what about reliability, the true purpose of mirroring? Though I'd maintain that a significant part of the price premium one pays for SCSI goes to much tighter manufacturing tolerances (and thus more reliability and longevity), two ATA drives mirrored may provide a more secure environment for data. But if you're -that- concerned with mission-critical stability, you should probably be looking at a mirrored pair of SCSI drives: If data integrity is really that important, cost should be secondary. Mirrored RAIDs protect only against disk failure. They're not a substitute for backup. I'd bet that most users experience some time of "soft" failure, whether it be the accidental deletion of a file or the infamous "OS Rot" that just seems to make a given installation less stable as time goes by. A mirrored pair won't protect you in this much-more-likely scenario.

If you've been following our other reviews, it should be obvious that we here at the Storage Review are very firm believers in the applicability of benchmarks and their correlation to real-world performance. All too often, it is popular to dump on benchmarks, dismissing them in trite phrases that include "lies" and "damned lies." Give the programmers at ZDBop and Adaptec some credit: They're studying high-level application performance and doing their best to replicate real-world disk activity in synthetic benchmarks. For the most part, my experiences have confirmed this. That said, I have to express skepticism at the ThreadMark results posted by the FastTrak. Tim, a contributing editor to StorageReview.com, will take the FastTrak and DiamondMax pair and live with a striped array for a week. Afterwards, he'll report back and post an article telling us whether or not the FastTrak lives up to the lofty figures that some benchmarks have produced.

Promise FastTrak ATA RAID Controller
Estimated Price: $149
Specifications

* Note: All reported test results are the average of five trials.


HOME | ARTICLES | LEADERBOARD | PERFORMANCE DATABASE | REFERENCE GUIDE
COMMUNITY | RELIABILITY SURVEY | SUPPORT SR! | ABOUT SR |

Copyright © 1998-2005 StorageReview.com, Inc. All rights reserved.
Write: Webmaster