Bug #2154
closed
rados: bench seq should not segfault when blocksize doesn't match write blocksize
Added by Dan Mick about 12 years ago.
Updated over 9 years ago.
Description
I misunderstood the problem to begin with. The issue is when read size does not match the prior write size (I hadn't realized there was a correlation but it makes sense).
- Subject changed from rados: bench should not segfault with -b 1024 to rados: bench should not segfault with -b < 1048576
- Subject changed from rados: bench should not segfault with -b < 1048576 to rados: bench seq should not segfault with -b < 1048576
- Description updated (diff)
- Subject changed from rados: bench seq should not segfault with -b < 1048576 to rados: bench seq should not segfault when blocksize doesn't match write blocksize
- Description updated (diff)
- Status changed from New to Resolved
loic@fold:~/software/ceph/ceph/src$ rados --pool rbd -b 4096 bench 5 write --no-cleanup --run-name try
Maintaining 16 concurrent writes of 4096 bytes for up to 5 seconds or 0 objects
Object prefix: benchmark_data_fold_20180
sec Cur ops started finished avg MB/s cur MB/s last lat avg lat
0 0 0 0 0 0 - 0
1 16 67 51 0.199177 0.199219 0.327991 0.228371
2 16 112 96 0.187461 0.175781 0.267772 0.32414
3 16 147 131 0.170542 0.136719 0.211988 0.360128
4 16 161 145 0.141577 0.0546875 0.994443 0.364593
5 16 172 156 0.121854 0.0429688 1.32801 0.422482
Total time run: 5.625777
Total writes made: 173
Write size: 4096
Bandwidth (MB/sec): 0.120
Stddev Bandwidth: 0.0802827
Max bandwidth (MB/sec): 0.199219
Min bandwidth (MB/sec): 0
Average Latency: 0.512548
Stddev Latency: 0.411049
Max latency: 2.49627
Min latency: 0.154259
loic@fold:~/software/ceph/ceph/src$ rados --pool rbd -b 1024 bench 5 seq --no-cleanup --run-name try
sec Cur ops started finished avg MB/s cur MB/s last lat avg lat
0 0 0 0 0 0 - 0
Total time run: 0.042500
Total reads made: 173
Read size: 4096
Bandwidth (MB/sec): 15.901
Average Latency: 0.00387084
Max latency: 0.00833
Min latency: 0.000879
Also available in: Atom
PDF