Shulou(Shulou.com)06/02 Report--

Ceph contains commands to perform performance benchmarking on the rados bench RADOS storage cluster. This command performs a write test and two types of read tests. -- no-cleanup this option is important when testing read and write performance. By default, the rados bench command deletes objects that have been written to the storage pool. Leaving behind these objects allows two read tests to measure order and random read performance.

(1) preparation to clear all file system caches

[root@node141 mnt] # echo 3 > / proc/sys/vm/drop_caches

[root@node141 mnt] # sync

(2) create a new storage pool

[root@node141 mnt] # ceph osd pool create rbd 128 128

(3) Test the write performance of rbd and add-- no-cleanup parameter

[root@node141 mnt] # rados bench-p rbd 10 write-- no-cleanup hints = 1Maintaining 16 concurrent writes of 4194304 bytes to objects of size 4194304 for up to 10 seconds or 0 objectsObject prefix: benchmark_data_node141_28363 sec Cur ops started finished avg MB/s cur MB/s last lat (s) avg lat (s) 0 000-0 116 81 65 259.999 260 0.107901 0.200157 2 16 161 145 289.976 320 0.190659 0.215092 3 16 252 236 314.632 364 0.0807696 0.193549 4 16 351 335 334.958 396 0.191582 0.185652 5 16 445 429 343.157 376 0.158618 0.178393 6 16 548 532 354.621 412 0.162274 0.178178 7 16 652 636 363.378 416 0.190742 0.173784 8 16 753 737 368.448 404 0.15392 0.170866 9 16 863 376.391 440 0.103038 0.169129 10 16 964 379.144 0.138741 0.167306Total time run: 10.1023Total writes made: 965Write size: 4194304Object size: 4194304Bandwidth (MB/sec): 382.09Stddev Bandwidth: 53.3267Max bandwidth (MB/sec): 440Min bandwidth (MB/sec): 260Average IOPS: 95Stddev IOPS: 13.3317Max IOPS: 110Min IOPS: 65Average Latency (s): 0.167422Stddev Latency (s): 0.0765119Max latency (s): 0.668499Min latency (s): 0.0569634

(4) Test, read in order of 10 seconds:

[root@node141 mnt] # rados bench-p rbd 10 seqhints = 1 sec Cur ops started finished avg MB/s cur MB/s last lat (s) avg lat (s) 0 000 0-0 1 16 257 241 963.679 964 0.0607106 0.0623843 2 16 549 533 1065.73 1168 0.0450988 0.0583824 3 16 835819 1091.76 1144 0.044447 0.057111Total time run: 3.47417Total reads made: 965Read size: 4194304Object size: 4194304Bandwidth (MB/sec): 1111.06Average IOPS: 277Stddev IOPS: 27.8747Max IOPS: 292Min IOPS: 241Average Latency (s): 0.056581Max latency (s): 0.388605Min latency (s): 0.015351

(5) Test the random read and write performance of 10 seconds.

[root@node141 mnt] # rados bench-p rbd 10 randhints = 1 sec Cur ops started finished avg MB/s cur MB/s last lat (s) avg lat (s) 0 000-0 1 16 408 392 1566.56 1568 0.038812 2 16 794 778 1554.08 1544 0.0219942 0.0396005 3 16 1211 1195 1591.93 1668 0.125987 0.0389289 415 1604 1589 1587.9 1576 0.0497586 0.0392412 5 16 2013 1997 1596.67 1632 0.00512184 0.0388821 6 16 2415 2399 1598.52 1608 0.146983 0.0389842 7 16 2806 2790 1593.56 1564 0.110503 0.0391283 8 16 3186 3170 1584.34 1520 0.0226104 0.0393014 9 15 3598 3583 1591.83 1652 0.0117805 0.0392624 10 16 3965 3949 1579.03 1464 0.0752829 0.0395739Total time run: 10.0536Total reads made: 3965Read size: 4194304Object size: 4194304Bandwidth (MB/sec): 1577.54Average IOPS: 394Stddev IOPS: 15.6308Max IOPS: 417Min IOPS: 366Average Latency (s): 0.0397174Max latency (s): 0.353027Min latency (s): 0.00414538

(6) to increase the number of concurrent reads and writes, use the-t default option of 16 fields, and the-b parameter resizes the object being written, and the default object size is 4MB. The maximum object of security is 16MB.

Add the-- run-name option to control the name of the object written during the benchmark. Multiple rados bench commands can run out at the same time-- the run-name tag for each instance of the running command. This prevents potential I / O errors that may occur when multiple clients try to access the same object and allow different clients to access different objects. -- run-name is also useful when trying to simulate real-world workloads. For example:

[root@node141 mnt] # rados bench-p rbd 10 write-t 4-- run-name zabbix71hints = 1Maintaining 4 concurrent writes of 4194304 bytes to objects of size 4194304 for up to 10 seconds or 0 objectsObject prefix: benchmark_data_node141_28450 sec Cur ops started finished avg MB/s cur MB/s last lat (s) avg lat (s) 0 000-0 1 4 40 36 143.992 144 0.065704 0.103059 2 4 86 82 163.981 184 0.0763531 0.0955013 34 126 122 162.641 160 0.106746 0.0953542 44 173 169 168.972 188 0.0872597 0.0934176 5 4 218 214 171.171 180 0.170851 0.0927783 64 255 251 167.305 148 0.0943133 0.093864 7 4 294 290 165.686 156 0.0976048 0.0959294 84 340 336 167.971 184 0.0882485 0.0946923 9 4 386 382 169.748 0.0703278 0.0933611 10 4 430 426 170.37 0.0623777 0.0933391Total time run: 10.1109Total writes made: 431Write size: 4194304Object size: 4194304Bandwidth (MB/sec): 170.51Stddev Bandwidth: 16.6747Max bandwidth (MB/sec): 188Min Bandwidth (MB/sec): 144Average IOPS: 42Stddev IOPS: 4.16867Max IOPS: 47Min IOPS: 36Average Latency (s): 0.0935851Stddev Latency (s): 0.0433054Max latency (s): 0.579434Min latency (s): 0.0533891Cleaning up (deleting benchmark objects) Removed 431 objectsClean up completed and total clean up time: 1.62203

(7) Delete the data created by rados bench command

[root@node141 mnt] # rados-p rbd cleanup

Removed 965 objects

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.