Odd disk performance results

Thanks, that's all very interesting.

At my end this is going on inside a tiny shuttle case, so raid'ing a few disks is definitely not an option.

I'm toying with the idea of a SCSI card and disk combination, which I guess might make quite a difference, will let you know if it happens and it makes a difference

[David (PassMark)]

The following numbers are an average of three runs,

Read: 52.0 MB/Sec
Write: 71.40 MB/Sec
R/W : 5.88 MB/Sec

The 1st run was the slowest, but as Windows increased the size of the disk cache the results improved slightly. On the final run we noticed that Windows had allocated a massive 416MB of RAM to the disk cache. (Remember that this is running Windows 2003 server, which likes to set-up large disk caches).

Looking at the graph for the Write test, we noted that the peak write speed was 450 MB/Sec. This is the speed of writing to the disk cache.

Using the advanced disk test (and using the un-cached option) we hit speeds of 82MB/Sec for Read and 79MB/Sec for Write. The interesting thing about this result is that the cache actually slowed things down (for large sequential files).

David
PassMark

That's interesting. What did you get for the seek+r/w? It seems to me that this is a better gauge for general-purpose computing as an OS's use of a disk is lots of small reads and writes. (I use the computer to write and run a range of computer programs, not to edit videos and music or play games).

[David (PassMark)]

Also, just for the record, we have just got some new results today.

With, dual Segate 80GB serial ATA (SATA) drives, stripped with RAID0, on Windows 2003 Server, we got to 60MB/sec writing and a more modest 40MB/Sec reading.

If you really want performance, RAID is the way to go.

David
PassMark

[David (PassMark)]

Disks that have been formatted with larger clusters and with FAT32 tend to get better results with this test. Cluster sizes of 16KB should be optimal for this test, becuase this is the size of the read and write operations.

The result can also depend on the size of the disk cache, DMA settings, the position of the test file on the disk, motherboard device drivers being used and the interface speed (e.g. ATA-100).

David
PassMark

[dcleal]

...forgot to add... the computer is running Win2000+NTFS, which might be a factor?

[Anonymous]

Hi

I just ran the tests on a computer, got roughly the results I expected except that the disk results were lousy. So I defragmented, and now the sequential figures are better, but the random seek performance is still very low - or at least by comparison with some of your canned benchmarks.

Any thoughts about whether these results look odd? And any suggestions about how to improve the speed for this hardware?

The parts of the set up that I think are relevant are
Shuttle SN41G2 (small form factor computer with custom motherboard)
AthlonXP2600+
1GB RAM
80 GB Seagate Barracuda, 7,200rpm, model #ST380011A

The disk results:
sequential read: 31.5
sequential write: 24.4
random seek + r/w: 2.5

The manufacturer claims a sustained transfer rate of 58 MB/s and a seek time of 8.5msec - which I'd have thought was pretty competitive. But the 2.5 for seek+r/w appears to be dire.

TIA for any suggestions
- Dave