So I am running iozone to measure Random throughput and notice some weird behavior that I cannot explain and hoping someone here might shed some lights for me. When running random test for a certain blocksize. I noticed that the random write sometimes have skip value for a certain thread and then pick up again. I looked in the log and saw that the min xfer value is very small meaning the file was not completely written so that's why it's skewed the result so skipping. However, the test ran fine for the next thread. Anyone can shed me some lights as what is going on or how to go around this?
Children see throughput for 12 random writers = 13978.97 KB/sec
Min throughput per process = 731.76 KB/sec
Max throughput per process = 1854.75 KB/sec
Avg throughput per process = 1164.91 KB/sec
Min xfer = 160.00 KB
The filesize here is 1GB and this is using 32k record size at 12 threads. Below is the full results for this random test:
32 KB:
1: 4.46 MB/sec 6.41 MB/sec
3: 9.48 MB/sec 14.94 MB/sec
6: 15.65 MB/sec 29.35 MB/sec
12: 52.29 MB/sec
24: 22.81 MB/sec 87.42 MB/sec
48: ->27.41 MB/sec 128.32 MB/sec
96: 25.49 MB/sec 149.81 MB/sec
144: ->154.43 MB/sec
As you can see from the above result, thread 12 has no value since the min xfer is too low which I don't understand how / why that is.
Aucun commentaire:
Enregistrer un commentaire