cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Coli <rc...@digg.com>
Subject Re: Cassandra memory consumption
Date Wed, 16 Feb 2011 19:56:21 GMT
On Wed, Feb 16, 2011 at 11:22 AM, Victor Kabdebon
<victor.kabdebon@gmail.com> wrote:
> Thanks robert, and do you know if there is a way to control the maximum
> likely number of memtables ? (I'd like to cap it at 2)

That "likely number of memtables" is the number of memtables which :

a) have been created
b) may or may not have hit a flush threshold
c) have not been flushed
d) and therefore are resident in memory

The only way to keep the number of resident memtables at or around 2
is to never write to your node significantly faster than you can
flush. Given other consumers of system I/O like compaction and the
implications of multiple CFs which all contend to flush, this can be
non-trivial in practice. "3" in my calculation here is a "safety
ceiling" type value, in that you would really have to be pushing your
node or have a small number of high-write CFs to actually have 3
memtables per CF resident across all CFs on a node.

=Rob

Mime
View raw message