Recently, the Storage Review took a look at the Promise FastTrack ATA RAID controller using a pair of Maxtor DiamondMax 2880 drives. The 2880 was chosen for its outstanding performance in a single-drive configuration. The setup churned out some strange results, however. WinBench 98 results under Windows 95 for a pair of drives turned out to be no greater and in some cases actually a bit -lower- than the results posted by a single drive. Conversely, ThreadMark 2.0 scores blew through the roof. ThreadMark scores were doubly suspect since the pair of Maxtors in a mirrored configuration managed to outperform a striped setup.

Soon after the original results were posted, a test engineer from Promise contacted us, offering feedback on the results and suggesting a retest using a larger 64k stripe block size rather than the 8k recommended by the manual. We've been waiting for delivery of another pair of DiamondMax's; since we haven't yet received them, we've decided to proceed with another test changing yet another variable- the pair of drives. We purchased two Western Digital AC36400's and decided to put the FastTrack through its paces once more.

The engineer also suggested the use of "v1.10" drivers, dated 3-31-98. In the initial tests, we were using drivers dated 3-23-98, though I was under the impression that these also were revision 1.10. This time around, the tested card had a v1.03 bios, as opposed to the v1.02 chip used in the original review. Needless to say, we have many variables changing here, so control has gone out the window. As soon as we receive them we still plan to take a look at the changes using the original drives.

Installation of the controller itself was quite smooth this second time around, marred only by two consecutive defective Caviars. We did notice that the v1.03 bios seems to take much longer to initialize, perhaps 10 seconds vs the 3 seconds that one waited with v1.02. The newer bios also makes copying the contents of one drive to another when seting up a mirrored pair an option rather than an automatic process, a slight time saver.

We configured the Western Digital drives as a striped pair, yielding a single large 12.8 gig volume. 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 Western Digital 6.4GB disk driven from the PIIX4 controller on the Abit LX6 is provided for comparison.

Something's definitely different here. The striped pair of drives, unlike before, outdistanced the single unit by 13% on both the Business and High-End Disk WinMarks, a definitive improvement. Curiously, however, the transfer rate achieved by the pair when being tested by WinBench's CPU utilization was only 12.7 MB/sec, well below the 18MB/sec that the setup was able to deliver on the outermost tracks. CPU utilization in this case is still rather lofty. ThreadMark scores continued to be astonishing, posting a stratospheric 76% gain when going to a striped pair from a single drive.

Striped-pair gains are even more impressive under Windows NT, a respective 18% and 27% increase in the Business and High-End WinMarks. This time around, CPU utilization and outer track transfer rates match, both being 20 MB/sec. Under NT, its clear that Promise has trimmed down striped CPU utilization considerably, eliminating a glaring problem using the earlier driver. ThreadMark results were as amazing as ever: An 83% increase.

This time around WinBench 98 reports that the gains provided from a mirrored pair are not as great as those yielded by a striped pair, which certainly makes sense. The mirror configuration provides a 1-10% increase over a single drive. Once again, however, ThreadMark refuses to correlate. Under Windows 95 the mirror bests the striped pair while under NT the situation reverses.

Though the ThreadMark results are still quite strange, WinBench 98 reports a clear increase in speed across the board using two drives as a striped or mirrored pair. Again, its difficult to pinpoint exactly what to attribute the changes to until we take a third look using the original Maxtor 5.7's. As it stands here, though, the FastTrack delivers real gains with a pair of Western Digitals- we'll see if the results carry over to the Maxtor's with the newer bios/drivers and larger block size.