Recently I had to compose a file server for high load LAN read access. My first thought was to implement a RAID 0 over 4 harddisks. I took a dual core 2.5GHz CPU and 2GB of RAM. I attached four 500GB SATA drives to the mainboard and then I installed Debian 6 Squeeze on a small part of the first hd. The rest of it and the other drives were dedicated to the RAID 0.
After the installation and the configuration of the system, I measured the read/write performance of the RAID. I did it with the time dd combination:
time dd if=/dev/zero of=file bs=1k count=20000000
time dd if=file of=/dev/null
As you see, write actions perform better than read actions. Why is that and how can we invert this effect? I think it’s because the RAID software hasn’t to check where to write data, but it must search for it on read access.
A complete test with Bonnie++ delivers slightly different results:
Sequential Write: 290Mb/s
Sequential Read: 308Mb/s
Random Seeks: 651/s (That’s twice the seek rate of my two RAID5 servers.)
Does anybody know why the test results of Bonnie++ and dd are so different?