General Disk Performance

[Dog-One]

I was curious if anyone has run some disk transfer tests to see what kind of performance numbers you all were getting.

I created a 4GB file and did read and write tests. The numbers I got were:

READ: 4294967296B @ 28.975s = 148,230,105 B/s or 141 MB/s
WRITE: 4294967296B @ 28.426s = 151,092,918 B/s or 144 MB/s

I'm using a 4 disk Linux software RAID0, stripe size of 4K on a XFS filesystem. I ran the tests with a BFV server and Folding@Home running.

Here's the commands I used:

time dd if=/dev/zero of=/var/tmp/test bs=65536 count=65536

time cat /var/tmp/test > /dev/null


I'm pretty sure this isn't the best or even a proper way to measure disk performace, but if anyone wants to use it for comparison, feel free.

[Jman]

Nice.

hdparm can do bechmarks. /sbin/hdparm -t /dev/hda This may be of limited value for software RAID. You can also subtitute -T for -t and get reads from cache.

FYI, I can get 23.21 MB/sec with one of my disks.

[Dog-One]

I ran my tests on my single Hitachi DeskStar drive using ext2 filesystem and got 38 MB/s read and write. Then I ran it on my USB 2.0 external Western Digital drive with xfs filesystem and got 26 MB/s read and write.

What I was looking for and what I found is that with a four disk stipe set, it is possible to break the 133 MB/s PCI barrier using the onboard nForce3 chipset. That tells me, there isn't a real need to hop on the PCI Express bus to get raw disk performance. As long as you use the onboard disk controllers, is quite possible to push your disk arrays as fast as they will go without any bottlenecks in the system architecture.

It's time for Intel to come up with a new marketing strategy. :)