why the performance of counter is so bad ?

["yue.zhang" <yu...@chinacache.com>]

env

------

3 node, every node has 8core CPU、32GB memory and 2*1TB disk

centos 5 64bit cassandra 1.2.5 opscenter3.1



my scenario

--------------

each node has a client program,  each client start 10 thread for increasing counter.

the replica=2, and other is defaut value.

my data model
------------------
rowkey: <date>_<domain>_<5min index(1-288)>, value have two cf :  IN:<value> and OUT:<value>
e.g. 20130503_www.facebook_24: IN:2342134, OUT:2423234

result
---------
everything is ok but the performance. from opscenter, the writers is only 6000/sec.

somebody tell me is normal situation. if not ,how to improve it.

P.S. CPU idle is greater than 60%, the load only 1.x(8core cpu) , and the memroy only use 50%.

thx
-heipark

Re: why the performance of counter is so bad ?

[Robert Coli <rc...@eventbrite.com>]

Briefly, counters read before write, unlike other writes in Cassandra.

http://www.datastax.com/dev/blog/whats-new-in-cassandra-0-8-part-2-counters
"
Counters have been designed to allow for very fast writes. However, increment
does involve a read on one of the replica as part of replication. As a
consequence,
counter increments are expected to be slightly slower than regular writes.
"

=Rob