ssd noop anticipatory deadline cfq io scheduler performance test-阿里云开发者社区

ssd noop anticipatory deadline cfq io scheduler performance test

2016-04-05 3515

版权

本文内容由阿里云实名注册用户自发贡献，版权归原作者所有，阿里云开发者社区不拥有其著作权，亦不承担相应法律责任。具体规则请查看《阿里云开发者社区用户服务协议》和《阿里云开发者社区知识产权保护指引》。如果您发现本社区中有涉嫌抄袭的内容，填写侵权投诉表单进行举报，一经查实，本社区将立刻删除涉嫌侵权内容。

简介：

上一篇测试了SSD分区对齐和不对齐的IOPS性能.

本文将要介绍一下SSD分区对齐的情况下, 不同的IO调度算法的iops性能.

测试结果如下

1. noop调度

[root@db-xx ~]# cat /sys/block/sde/queue/scheduler 
[noop] anticipatory deadline cfq 

[postgres@db-xx pgdata]$ pg_test_fsync
5 seconds per test
O_DIRECT supported on this platform for open_datasync and open_sync.

Compare file sync methods using one 16kB write:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                       10574.965 ops/sec      95 usecs/op
        fsync                            9959.066 ops/sec     100 usecs/op
        fsync_writethrough                            n/a
        open_sync                       13894.208 ops/sec      72 usecs/op

Compare file sync methods using two 16kB writes:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                        7272.895 ops/sec     137 usecs/op
        fsync                            6892.519 ops/sec     145 usecs/op
        fsync_writethrough                            n/a
        open_sync                        6968.750 ops/sec     143 usecs/op

Compare open_sync with different write sizes:
(This is designed to compare the cost of writing 16kB
in different write open_sync sizes.)
         1 * 16kB open_sync write       13905.409 ops/sec      72 usecs/op
         2 *  8kB open_sync writes       8994.942 ops/sec     111 usecs/op
         4 *  4kB open_sync writes       5241.717 ops/sec     191 usecs/op
         8 *  2kB open_sync writes       2500.587 ops/sec     400 usecs/op
        16 *  1kB open_sync writes       1300.364 ops/sec     769 usecs/op

Test if fsync on non-write file descriptor is honored:
(If the times are similar, fsync() can sync data written
on a different descriptor.)
        write, fsync, close              9292.012 ops/sec     108 usecs/op
        write, close, fsync              9201.620 ops/sec     109 usecs/op

Non-Sync'ed 16kB writes:
        write                           124100.336 ops/sec       8 usecs/op

2. anticipatory调度

[root@db-xx ~]# echo anticipatory > /sys/block/sde/queue/scheduler 
[root@db-xx ~]# cat /sys/block/sde/queue/scheduler 
noop [anticipatory] deadline cfq 

[postgres@db-xx pgdata]$ pg_test_fsync
5 seconds per test
O_DIRECT supported on this platform for open_datasync and open_sync.

Compare file sync methods using one 16kB write:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                       10980.717 ops/sec      91 usecs/op
        fsync                            9989.117 ops/sec     100 usecs/op
        fsync_writethrough                            n/a
        open_sync                       13447.041 ops/sec      74 usecs/op

Compare file sync methods using two 16kB writes:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                        7001.547 ops/sec     143 usecs/op
        fsync                            6757.166 ops/sec     148 usecs/op
        fsync_writethrough                            n/a
        open_sync                        6852.453 ops/sec     146 usecs/op

Compare open_sync with different write sizes:
(This is designed to compare the cost of writing 16kB
in different write open_sync sizes.)
         1 * 16kB open_sync write       13445.661 ops/sec      74 usecs/op
         2 *  8kB open_sync writes       8902.478 ops/sec     112 usecs/op
         4 *  4kB open_sync writes       5137.514 ops/sec     195 usecs/op
         8 *  2kB open_sync writes       2473.485 ops/sec     404 usecs/op
        16 *  1kB open_sync writes       1248.299 ops/sec     801 usecs/op

Test if fsync on non-write file descriptor is honored:
(If the times are similar, fsync() can sync data written
on a different descriptor.)
        write, fsync, close              8992.447 ops/sec     111 usecs/op
        write, close, fsync              8999.892 ops/sec     111 usecs/op

Non-Sync'ed 16kB writes:
        write                           131947.640 ops/sec       8 usecs/op

3. deadline调度

[root@db-xx ~]# echo deadline > /sys/block/sde/queue/scheduler 
[root@db-xx ~]# cat /sys/block/sde/queue/scheduler 
noop anticipatory [deadline] cfq 

[postgres@db-xx pgdata]$ pg_test_fsync
5 seconds per test
O_DIRECT supported on this platform for open_datasync and open_sync.

Compare file sync methods using one 16kB write:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                       10475.719 ops/sec      95 usecs/op
        fsync                            9861.660 ops/sec     101 usecs/op
        fsync_writethrough                            n/a
        open_sync                       13804.860 ops/sec      72 usecs/op

Compare file sync methods using two 16kB writes:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                        6998.343 ops/sec     143 usecs/op
        fsync                            6704.945 ops/sec     149 usecs/op
        fsync_writethrough                            n/a
        open_sync                        6906.296 ops/sec     145 usecs/op

Compare open_sync with different write sizes:
(This is designed to compare the cost of writing 16kB
in different write open_sync sizes.)
         1 * 16kB open_sync write       13755.100 ops/sec      73 usecs/op
         2 *  8kB open_sync writes       8987.003 ops/sec     111 usecs/op
         4 *  4kB open_sync writes       5242.396 ops/sec     191 usecs/op
         8 *  2kB open_sync writes       2491.866 ops/sec     401 usecs/op
        16 *  1kB open_sync writes       1249.717 ops/sec     800 usecs/op

Test if fsync on non-write file descriptor is honored:
(If the times are similar, fsync() can sync data written
on a different descriptor.)
        write, fsync, close              8981.361 ops/sec     111 usecs/op
        write, close, fsync              9001.074 ops/sec     111 usecs/op

Non-Sync'ed 16kB writes:
        write                           121845.410 ops/sec       8 usecs/op

4. cfq调度

[root@db-xx ~]# echo cfq > /sys/block/sde/queue/scheduler 
[root@db-xx ~]# cat /sys/block/sde/queue/scheduler 
noop anticipatory deadline [cfq] 

[postgres@db-xx pgdata]$ pg_test_fsync
5 seconds per test
O_DIRECT supported on this platform for open_datasync and open_sync.

Compare file sync methods using one 16kB write:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                       10981.601 ops/sec      91 usecs/op
        fsync                            9312.220 ops/sec     107 usecs/op
        fsync_writethrough                            n/a
        open_sync                       13029.732 ops/sec      77 usecs/op

Compare file sync methods using two 16kB writes:
(in wal_sync_method preference order, except fdatasync
is Linux's default)
        open_datasync                                 n/a
        fdatasync                        6997.665 ops/sec     143 usecs/op
        fsync                            6082.214 ops/sec     164 usecs/op
        fsync_writethrough                            n/a
        open_sync                        6519.312 ops/sec     153 usecs/op

Compare open_sync with different write sizes:
(This is designed to compare the cost of writing 16kB
in different write open_sync sizes.)
         1 * 16kB open_sync write       12990.761 ops/sec      77 usecs/op
         2 *  8kB open_sync writes       8603.877 ops/sec     116 usecs/op
         4 *  4kB open_sync writes       4986.898 ops/sec     201 usecs/op
         8 *  2kB open_sync writes       2369.063 ops/sec     422 usecs/op
        16 *  1kB open_sync writes       1220.165 ops/sec     820 usecs/op

Test if fsync on non-write file descriptor is honored:
(If the times are similar, fsync() can sync data written
on a different descriptor.)
        write, fsync, close              8919.973 ops/sec     112 usecs/op
        write, close, fsync              8857.715 ops/sec     113 usecs/op

Non-Sync'ed 16kB writes:
        write                           124013.883 ops/sec       8 usecs/op

ssd noop anticipatory deadline cfq io scheduler performance test

热门文章

最新文章

相关电子书