Moosefs tuning notes

The report says moosefs The write performance of the group starts from 10MB/sec Down to 3MB/sec.（ Read performance 10MB/sec unchanged ）

Zha Zhi .

hold MFSMASTER.CFG In the default ： CHUNKS_WRITE_REP_LIMIT = 1 CHUNKS_READ_REP_LIMIT = 5

Change to ： CHUNKS_WRITE_REP_LIMIT = 0 CHUNKS_READ_REP_LIMIT = 0

Write speed restored to 10MB/sec.

Then change to ： CHUNKS_WRITE_REP_LIMIT = 10 CHUNKS_READ_REP_LIMIT = 50

Write speed drops to 300KB/sec.

These parameters are adjusted moosefs Balance to the new machine chunk The strength of , Test validation is not a control client Read and write . It seems moosefs stay ChunkSever There's a balance between chunk The performance of the whole cluster will be degraded , I think I went there a few days ago moosefs New in the group 4 Caused by a machine .

Official FAQ Also say this replication You bet “it can substantially slow down the whole system.”.

The official link ——http://www.moosefs.org/moosefs-faq.html#rebalancing-speed

moosefs Few parameters can be adjusted in terms of performance , This is the only place in the configuration file where performance can be adjusted , Remember .

p.s. XFS be used for moosefs Always crash , I'm still honest and practical ext3 Well ..（2011.11.10）