I am running ceph bench mark on a 3 node cluster. I am seeing that the bandwidth is going down and latencies are going up beyond 16k. What must be going on ? Is there anything I must check? Thanks in advance for your help. Hard disk can handle it.(dd below) and it is a 1Gb network. vjujjuri@rgulistan-wsl13:~/ceph-cluster$ sudo dd if=/dev/zero of=/media/data/ddtest bs=8k count=100 oflag=direct 100+0 records in 100+0 records out 819200 bytes (819 kB) copied, 0.00778172 s, 105 MB/s vjujjuri@rgulistan-wsl13:~/ceph-cluster$ sudo dd if=/dev/zero of=/media/data/ddtest bs=16k count=100 oflag=direct 100+0 records in 100+0 records out 1638400 bytes (1.6 MB) copied, 0.00819058 s, 200 MB/s vjujjuri@rgulistan-wsl13:~/ceph-cluster$ sudo dd if=/dev/zero of=/media/data/ddtest bs=32k count=100 oflag=direct 100+0 records in 100+0 records out 3276800 bytes (3.3 MB) copied, 0.0133412 s, 246 MB/s vjujjuri@rgulistan-wsl13:~/ceph-cluster$ sudo dd if=/dev/zero of=/media/data/ddtest bs=64k count=100 oflag=direct 100+0 records in 100+0 records out 6553600 bytes (6.6 MB) copied, 0.0213862 s, 306 MB/s vjujjuri@rgulistan-wsl13:~/ceph-cluster$ rados bench -p sfdc_ssd -b 8192 10 write --no-cleanup Maintaining 16 concurrent writes of 8192 bytes for up to 10 seconds or 0 objects Object prefix: benchmark_data_rgulistan-wsl13_3978 sec Cur ops started finished avg MB/s cur MB/s last lat avg lat 0 0 0 0 0 0 - 0 1 16 4491 4475 34.9558 34.96090.004625760.00356678 2 16 8019 8003 31.2573 27.56250.005743870.00399099 3 15 11550 11535 30.0349 27.59380.003749640.00415737 4 16 14836 14820 28.9414 25.66410.005634550.00431221 5 15 17328 17313 27.0481 19.47660.007941170.00461614 6 16 20654 20638 26.8692 25.97660.003703640.00464857 7 16 24144 24128 26.9253 27.26560.004056120.00463968 8 16 27647 27631 26.9803 27.36720.002609330.00463053 9 15 31049 31034 26.9361 26.58590.005216640.00463805 10 16 34482 34466 26.9235 26.81250.009889080.00464022 Total time run: 10.006237 Total writes made: 34482 Write size: 8192 Bandwidth (MB/sec): 26.922 Stddev Bandwidth: 8.84853 Max bandwidth (MB/sec): 34.9609 Min bandwidth (MB/sec): 0 Average Latency: 0.00464095 Stddev Latency: 0.00186824 Max latency: 0.035694 Min latency: 0.00191166 vjujjuri@rgulistan-wsl13:~/ceph-cluster$ rados bench -p sfdc_ssd -b 16384 10 write --no-cleanup Maintaining 16 concurrent writes of 16384 bytes for up to 10 seconds or 0 objects Object prefix: benchmark_data_rgulistan-wsl13_3999 sec Cur ops started finished avg MB/s cur MB/s last lat avg lat 0 0 0 0 0 0 - 0 1 16 3416 3400 53.1168 53.1250.005582130.00468952 2 16 6804 6788 53.0233 52.93750.004283230.00470578 3 16 10310 10294 53.6069 54.78120.004640170.00465615 4 16 13598 13582 53.0482 51.3750.005829570.00470762 5 16 16159 16143 50.4404 40.01560.00569706 0.0049528 6 16 19616 19600 51.0352 54.01560.005133240.00489462 7 16 22904 22888 51.0828 51.3750.00464216 0.0048904 8 15 26388 26373 51.5034 54.45310.00280325 0.0048512 9 16 29098 29082 50.4835 42.32810.004756190.00494863 10 16 32401 32385 50.5954 51.6094 0.00410140.00493792 Total time run: 10.007195 Total writes made: 32401 Write size: 16384 Bandwidth (MB/sec): 50.590 Stddev Bandwidth: 16.0195 Max bandwidth (MB/sec): 54.7812 Min bandwidth (MB/sec): 0 Average Latency: 0.00493992 Stddev Latency: 0.00167482 Max latency: 0.0296512 Min latency: 0.00211289 vjujjuri@rgulistan-wsl13:~/ceph-cluster$ rados bench -p sfdc_ssd -b 32768 10 write --no-cleanup Maintaining 16 concurrent writes of 32768 bytes for up to 10 seconds or 0 objects Object prefix: benchmark_data_rgulistan-wsl13_4175 sec Cur ops started finished avg MB/s cur MB/s last lat avg lat 0 0 0 0 0 0 - 0 1 16 1224 1208 37.7451 37.750.005912320.00995742 2 16 2145 2129 33.2615 28.7812 0.0100326 0.014982 3 16 3707 3691 38.4427 48.8125 0.01267 0.0128672 4 16 4880 4864 37.9949 36.65620.00840912 0.0131463 5 16 6292 6276 39.2197 44.125 0.0107234 0.0123698 6 16 7392 7376 38.4115 34.375 0.0052234 0.0130013 7 16 8941 8925 39.8385 48.40620.00581271 0.0125307 8 16 10395 10379 40.5378 45.43750.00430514 0.0123214 9 16 11257 11241 39.0263 26.9375 0.011219 0.0127994 10 16 12757 12741 39.8106 46.8750.00634619 0.0125532 Total time run: 10.251305 Total writes made: 12758 Write size: 32768 Bandwidth (MB/sec): 38.891 Stddev Bandwidth: 14.2386 Max bandwidth (MB/sec): 48.8125 Min bandwidth (MB/sec): 0 Average Latency: 0.0128327 Stddev Latency: 0.0231052 Max latency: 0.259471 Min latency: 0.00291666 -- Jvrao --- First they ignore you, then they laugh at you, then they fight you, then you win. - Mahatma Gandhi _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com