Performance - File Copy

Since we're just starting DAS coverage, I experimented a bit with a few performance measures. I tried our NAS iozone and Vista filecopy test tools and also tried HD Tune. I found that iozone, even when I used the -c switch, which in NAS across-the-network testing has the effect of removing higher-than-network-speed cached performance, provided results that were above both the 150 MB/s and 300 MB/s maximum transfer rates of 1.5 and 3.0 Gbps SATA connections until file sizes grew above 256 MBytes (Figure 4).

Figure 4: iozone performance

On the other hand, the Vista SP1 filecopy results were more realistic. Figure 5 is a summary of the write test results with the UMP configured in RAID 0 and 1 and with FAT32 and NTFS formatting. The USB 2.0 connection is clearly limiting the UMP's performance to the low 20 MB/s range. With a switch to the eSATA connection, write speeds jump to almost 108 MB/s for RAID 0, FAT 32 and 96 MB/s with RAID 0, NTFS.

Figure 5: Vista SP1 Filecopy write performance summary

The read test summary shown in Figure 6 again shows USB 2.0 connected performance in the 20 MB/s range, but slightly higher than write. The eSATA results are once more much higher, although only reaching a maximum of 83 MB/s for RAID 0, NTFS format.

Figure 6: Vista SP1 Filecopy read performance summary

These above results are all obtained with Windows drive write caching disabled because the Windows properties for the drive said that its write cache setting couldn't be modified (Figure 7). Each value is also the average of three filecopy runs.

Figure 7: Write caching can't be changed