Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

SiSoft Sandra - ATTO Disk Benchmark

So far we were only concerned with read performance and in that category, the RAID Level 5 looks actually much better than what we had expected from HDTach originally. Reads are, however, not the Achilles Heel of Level5, writes are the main limitation in performance. To get some basic idea of how the Rocket RAID controller handles writes, we ran SiSoft Sandra FileSystem benchmark.

The results in SiSoft Sandra basically mirror what we found in Winbench99 for read performance whereas they add the extra data points for write performance, too. ATTO performance is in general shown a bit lower than both WB99 Disk inspection test and SiSoft Sandra but the results are overall comparable

       

For comparison we are showing a 2-drive RAID Level0 and a 4 drive RAID Level5 using the Rocket RAID 1640. The sequential performance in RAID0 is excellent, the buffered performance is not bad either but somewhat limited by the PCI interface and the bridge translation from parallel to serial. In the Level 5 configuration, the read transfers are even better, likewise, the buffered writes are superb for a PCI card-based configuration. However, a look at the sequential writes leaves not the slightest shadow of a doubt that this is a RAID Level5 configuration. The same goes for ATTO Disk benchmark.

One of the questions we already hinted at earlier in this article was concerning the CPU usage. Most benchmarks have built-in measuring tools, however, the degree of spread between any of the benchmarks alone casts some serious doubts on the validity of the metrics, which led us to throw out these values altogether and instead use Windows performance monitoring tool that is independent of the benchmark and, at least as far as we are concerned, delivers relatively correct estimates of the CPU usage during any of the operations. This is an important point, since the quality of a controller is also measured with respect to the CPU cycles that are used for the I/O traffic. Moreover, in RAID Level5, the XOR calculations are very intensive and, in the absence of dedicated hardware, have to be carried out in software emulation.

   

CPU usage during SiSoft Sandra hovers somewhere between 5% (burst measurements), 30% (sequential reads) and 95% (sequential writes). During any run of ATTO, CPU usage increases to a flat 100%, which really stresses the level of load the XOR calculations are throwing at the CPU.

next page:    => PCMark04 =>