Here’s the details of the performance testing listed below, hardware etc and the raw IOZone numbers.
Client Details:
- Dual-Core AMD Opteron(tm) Processor 1220 (Sun 2100)
- 2.8Ghz, 2GB Memory, 1Gbps network, RHEL4
Server Details:
- Intel Xeon 5120, 1.86Ghz (dell 1950)
- 1GB Memory, 1Gbps network, RHEL5
- Perc5i raid controller setup in 2 7-disk raid 5 volumes
- Disk: Dell MD1000 w/ 750Gb SATA 7200rpm drives
Untuned results (4G test, 4k records) in KB/s
random random bkwd
KB reclen write rewrite read reread read write read
4194304 4096 93137 68662 28869 29300 23118 57440 23186
Tuned results:
random random bkwd
KB reclen write rewrite read reread read write read
4194304 4096 95360 71281 108200 112603 67761 69528 68227
Notice the huge gains in read performance 28MB/s to 108MB/s for sequential and 23MB/s to 67MB/s for non-sequential.
Here’s the theoretical max based on running iozone on the server. It looks like nfs is adding overhead on rewrites and possibly random reads and writes with sequential reads/writes bottlenecking at the 1Gbps connection
random random bkwd
KB reclen write rewrite read reread read write read
2097152 4096 302379 234327 174385 196905 102032 144843 99866