Lundi, 03 janvier 2011
Deduplication in StorIQ 3.0 : setup and data
Some people wanted some more information on the test setup used in the deduplication benchmarks, so here are some more details :
- dual Opteron 2376 system
- 16 GB RAM
- 2 Adaptec 52445 RAID controllers
- 24 Hitachi Ultrastar 2 TB drives in 2 RAID-6 arrays, 64KB stripe.
- Kernel Linux 2.6.32.11 AMD64 (custom built from vanilla www.kernel.org source)
- RAID arrays agreggated with LVM and mounted as an XFS filesystem.
- Lessfs 1.1.9.10
Here are the bonnie++ results (meaningless and empty columns removed). You can clearly see the very low LessFS metadata performance (though Lessfs 1.2.x improves on this, more benchmarks to come).
| name | put_block | rewrite | get_block | seeks | ran_del |
|---|---|---|---|---|---|
| XFS | 864107 | 393836 | 1048367 | 555,5 | 28255 |
| XFS | 885654 | 392354 | 1076189 | 556,3 | 25665 |
| XFS | 812471 | 395010 | 1109478 | 563,5 | 28611 |
| XFS | 826676 | 389504 | 974248 | 573,7 | 28056 |
| lessfs | 114409 | 53563 | 218081 | 3090,8 | 24 |
| lessfs | 127666 | 71852 | 225472 | 3165,6 | 23 |
| lessfs | 124484 | 73638 | 216317 | 1720,4 | 23 |
| lessfs | 124739 | 73279 | 224726 | 2427,5 | 23 |
The block performance results are drawn below :
As you can see, deduplicated performance is more than enough to saturate a 1 Gb network. For more performance hungry users, it will be a great secondary storage. Notice that it’s possible to both have your cake and eat it as you might use both XFS — for maximum performance — and Lessfs — for maximum capacity — storage simultaneously.
Happy New Year!
posté à: 19:06 permalink