The Benchmarks...
In the following trials, the 2400A's write cache was enabled and set to 'write through' mode (the default). The stripe size was 64k for all tests. Driver version 3.02 was used along with BIOS revision 3.50c.
Note: Shortly after testing was completed, Adaptec informed us of a new BIOS release (version 3.607) and a new driver release (version 3.03) for the ATA RAID 2400A. We could find no discernable performance difference using both the new BIOS and new drivers; therefore, we decided to stick with the original results.
WinBench Results...
We'll first take a look at single-drive scores to use as a baseline.
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - Baseline Scores
|
|
|
|
|
|
The only thing that stands out here is the Adaptec's slightly higher access time relative to most other base configurations.
WinBench RAID 0 Scores...
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 0 with 2 Drives
|
|
|
|
|
|
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 0 with 3 Drives
|
|
|
|
|
|
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 0 with 4 Drives
|
|
|
|
|
|
Overall, the 2400A's RAID 0 Diskmark scores don't increase at all relative to a single drive. But, as we've learned in previous RAID reviews, WinBench's Diskmark tests can be extremely unrepresentative of RAID performance. For that reason, we'll wait and see what IOMeter has to say about the ATA RAID 2400A's RAID 0 performance before drawing any conclusions.
Of particular interest is the 2400A's incredible STR performance. When we saw the Escalade 2400 sustaining 103MB/sec in a 4-drive RAID 0 configuration, we were quite surprised... We simply didn't think the PCI bus was capable of sustaining such a high transfer rate. Imagine our surprise when the 2400A turned in an even-more-impressive 109MB/sec.
WinBench reports a CPU utilization of 88% in the 4-drive RAID 0 configuration, just like it has done with every other RAID card we've tested so far. This is, of course, unrepresentative of reality.
WinBench RAID 1 Scores...
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 1 with 2 Drives
|
|
|
|
|
|
As was the case with the Escalade 6400 and SuperTrak100, the 2400A yields RAID 1 WinBench Diskmark scores significantly lower than that of a single drive. In both cases, we concluded that the scores were unrepresentative of each card's performance. The same is likely true for the 2400A as well; again, IOMeter may provide a more accurate performance picture.
That aside, it's interesting to note that the 2400A's STR remains unchanged relative to a single drive. This indicates that, unlike the Escalade, the 2400A isn't doing any load balancing during sequential reads. Access time is also unchanged, which by itself could lead one to conclude that the 2400A doesn't load balance random reads, either. As we'll see, however, the 2400A's RAID 1 IOMeter scores show otherwise.
WinBench RAID 01 Scores...
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 10/01 with 4 Drives
|
|
|
|
|
|
The 2400A's RAID 01 Diskmark scores are obviously unrepresentative, so we'll just ignore them. The card's STR is what we'd expect based on what we've seen so far. Its slightly higher access time in RAID 01 (compared to a single drive), however, is interesting.
WinBench RAID 5 Scores...
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 5 with 3 Drives
|
|
|
|
|
|
|
Ziff Davis WinBench 99 under Windows 2000 Professional using NTFS - RAID 5 with 4 Drives
|
|
|
|
|
|
WinBench's RAID 5 Diskmark results continue to be useless with the ATA RAID 2400A. When we first encountered this issue with the SuperTrak100, we hoped that the card itself was at fault so that it would be an isolated problem. Obviously, this isn't the case; it appears that we're going to be stuck with the problem until either the testbed changes (if it's somehow the fault of the motherboard, for example) or until a new version of WinBench is released.
Astute readers may notice something interesting about the 2400A's access times: they are, on average, about 0.5ms higher than single-drive configurations (SL6 and Ultra66) as well as the Escalade's scores. This is somewhat disconcerting- 0.5ms is fairly large overhead for random, linear reads. Does this increase in access time cause the 2400A to stumble in random IO situations? Let's take a look at some IOMeter scores and find out.
IOMeter...
