Hi everybody,

I've a performance issue on my CouchDB servers.

My OS is a FreeBSD-8.3 with ZFS. My ZFS pool is a raid10 array (two 
concatenated mirror), using four 7.2K SATA Disk and a SSD for the ZIL (== write 
cache). I'm using CouchDB-1.2.0.

My database, before compaction, is 1TB with 91M documents. CouchDB says there 
is 90GB of data (~1KB per doc).


When running compaction, CouchDB takes very little CPU (~20% of one processor 
for beam.smp) and reads very slowly: 800KB/s (~200KB/s per disk, small 
transactions rate). Compaction takes days...

That is really strange because I'm able to read my full database, with dd, at a 
better speed: 80MB/s (~20MB/s per disk and bigger transactions rate) not 
perfect but quicker than CouchDB.

What can make CouchDB read so slowly ? Does anybody have an idea about this 
slowness ?

thank you for your help.

Best,
jym

PS: We plan to use CouchDB in a write-heavy application: 500 doc inserted by 
seconds, only one simple view. Servers will be bigger (12 or 24 disks, SSD for 
cache) and the bigcouch merge will be a great help but we need better 
performance on a single node while waiting the 2.0 release :-)

Reply via email to