incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Edward Capriolo <edlinuxg...@gmail.com>
Subject Re: Does servers with different capacities in a cluster affect the overall performance?
Date Tue, 22 Feb 2011 16:57:11 GMT
On Tue, Feb 22, 2011 at 5:13 AM, XiaoboGu <guxiaobo1982@gmail.com> wrote:
> I mean servers with different CPU cores ,memory, or disk space, does
> Cassandra allow this kind of configuration?

This is allowed but managing this may be more difficult in production.
Most settings are applied globally at the column family level, such as
memtable_flush_mb for example. This means that the you will you will
never be able to get tuning settings "perfect" because you will always
have to take a middle ground approach.

Moreover the Random partitioner works best when each node has an equal
share of data. Unbalanced ring is the enemy because nodes with more
data see more requests, and each request has to work through more
data. Thus unbalanced nodes typically become the ones that start
showing performance issues first.

It also becomes really difficult to diagnose performance issues with
an increasing number of variables (this node has 2x data but 4x the
ram of node X, and 30% the processing power of node Y.)

Short of suggesting hardware I "hint" that 1U's and Blades are good
platforms over "big iron" because scale out is less difficult then
scale up. Drastically mismatched hardware is something i would avoid.

Mime
View raw message