I used IOMeter to run a performance test on the storage network at Environment Canada. There was a point when I was unsure whether or not I would be able to get all the disks to test available to one Linux machine, so I wanted to know if IOMeter results were comperable on Windows versus Linux.
I took an desktop Pentium III with a rather generic IDE harddisk and ran the same tests when booted from Windows XP Pro found on the other harddisk and a Red Hat Enterprise Linux 4 rescue CD.
| Target Type | Linux | Windows |
|---|---|---|
| IOps | 26147.4072 | 129.234693 |
| Read IOps | 12971.39475 | 64.265085 |
| Write IOps | 13176.01245 | 64.969608 |
| MBps | 334.93906 | 1.665443 |
| Read MBps | 166.137016 | 0.830396 |
| Write MBps | 168.802045 | 0.835047 |
| Transactions per Second | 26147.4072 | 129.234693 |
| Connections per Second | 0 | 0 |
| Average Response Time | 0.037145 | 7.736072 |
| Average Read Response Time | 0.037205 | 12.838305 |
| Average Write Response Time | 0.037085 | 2.689168 |
| Average Transaction Time | 0.037145 | 7.736072 |
| Average Connection Time | 0 | 0 |
| Maximum Response Time | 7.277046 | 56.640045 |
| Maximum Read Response Time | 5.331682 | 56.640045 |
| Maximum Write Response Time | 7.277046 | 37.456569 |
| Maximum Transaction Time | 7.277046 | 56.640045 |
| Maximum Connection Time | 0 | 0 |
| Errors | 0 | 0 |
| Read Errors | 0 | 0 |
| Write Errors | 0 | 0 |
| Bytes Read | 51980967424 | 221229568 |
| Bytes Written | 52814801920 | 222468608 |
| Read I/Os | 3870479 | 16328 |
| Write I/Os | 3931534 | 16507 |
| Connections | 0 | 0 |
| Transactions per Connection | -1 | -1 |
| Total Raw Read Response Time | 1.43581E+11 | 2.09624E+11 |
| Total Raw Write Response Time | 1.45376E+11 | 44390088176 |
| Total Raw Transaction Time | 2.88957E+11 | 2.54014E+11 |
| Total Raw Connection Time | 0 | 0 |
| Maximum Raw Read Response Time | 5316119 | 56640045 |
| Maximum Raw Write Response Time | 7255804 | 37456569 |
| Maximum Raw Transaction Time | 7255804 | 56640045 |
| Maximum Raw Connection Time | 0 | 0 |
| Total Raw Run Time | 2.97515E+11 | 2.54073E+11 |
| % CPU Utilization | 100 | 2.327142 |
| % User Time | 28.917035 | 0.943566 |
| % Privileged Time | 71.082965 | 1.401531 |
| % DPC Time | 0 | 0.114491 |
| % Interrupt Time | 0 | 0.52508 |
| Processor Speed | 997081000 | 1000000000 |
| Interrupts per Second | 1035.193004 | 1247.618523 |
| CPU Effectiveness | 261.474072 | 55.533655 |
| Packets/Second | 85.638654 | 83.008992 |
| Segments Retransmitted/Second | 0.003351 | 0.1025 |
So, this clearly shows that Windows has disk perfomance that is between 10-1000 times slower than Linux. Believe it? I don't. Same software, same hardware, same test, different results. Yet no credibility. I don't know what I did wrong as such, but I know that I don't trust those results. Consider that when ever you get any performance comparision, Linux versus Windows or otherwise.
| Thu, 7 May 2009 16:25:53 -0400 |
|