Skip to content

Latest commit

 

History

History
126 lines (76 loc) · 6.28 KB

CHANGELOG-2.3.md

File metadata and controls

126 lines (76 loc) · 6.28 KB
Raw

CHANGELOG of v2.3

Previous change logs can be found at CHANGELOG-2.2

Notable Changes

new features

optimization

Data Performance

Hardware: 3 nodes (3mds, 9metaserver), each with:

  • Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz

  • 256G RAM

  • disk cache: INTEL SSDSC2BB80 800G(iops is about 30000+,bw is about 300MB)

  • performance is as follows:

s3 backend is minio and the cto is disable(what is cto). and as you know, the performance of read may associated with read cache hit.

minio + diskcache(free) iops/bandwidth avg-latency(ms) clat 99.00th (ms) clat 99.99th (ms)
(numjobs 1) (50G filesize) 4k randwrite 3539 0.281 1.5 16
(numjobs 1) (50G filesize) 4k randread 2785 0.357 0.9 5.8
(numjobs 1) (50G filesize) 512k write 290 MB/s 1 600 ms 248
(numjobs 1) (50G filesize) 512k read 216 MB/s 4.3 275 ms 7.6
minio + diskcache(near full) iops/bandwidth avg-latency(ms) clat 99.00th (ms) clat 99.99th (ms)
(numjobs 1) (50G filesize) 4k randwrite 2988 0.3 1.2 18
(numjobs 1) (50G filesize) 4k randread 1559 0.6 1.9 346
(numjobs 1) (50G filesize) 512k write 266 MB/s 0.9 600 ms 396
(numjobs 1) (50G filesize) 512k read 82 MB/s 86 275 ms 901
minio + diskcache(full) iops/bandwidth avg-latency(ms) clat 99.00th (ms) clat 99.99th (ms)
(numjobs 1) (20G filesize * 5) 4k randwrite 2860 1.7 14 41
(numjobs 1) (20G filesize * 5) 4k randread 76 65 278 725
(numjobs 1) (20G filesize * 5) 512k write 240 MB/s 10 278 513
(numjobs 1) (20G filesize * 5) 512k read 192 MB/s 12 40 1955

Metadata Performance

Cluster Topology: 3 nodes (3mds, 6metaserver), each metaserver is deployed on a separate SATA SSD

Configuration: use default configuration

Test tool: mdtest v3.4.0

Test cases:

  • case 1: the directory structure is relatively flat and total 130,000 dirs and files. mdtest -z 2 -b 3 -I 10000 -d /mountpoint

  • case 2: the directory structure is relatively deep and total 204,700 dirs and files. mdtest -z 10 -b 2 -I 100 -d /mountpoint

Case Dir creation Dir stat Dir rename Dir removal File creation File stat File read File removal Tree creation Tree removal
case 1 1320 5853 149 670 1103 5858 1669 1419 851 64
case 2 1283 5205 147 924 1081 5316 1634 1260 1302 887

You can set configuration item ‘fuseClient.enableMultiMountPointRename’ to false in client.conf if you don't need concurrent renames on multiple mountpoints on the same filesystem. It will improve the performance of metadata.

fuseClient.enableMultiMountPointRename=false

Case Dir creation Dir stat Dir rename Dir removal File creation File stat File read File removal Tree creation Tree removal
case 1 1537 7885 530 611 1256 7998 1861 1614 1050 72
case 2 1471 6328 509 1055 1237 6874 1818 1454 1489 1034

bug fix