If you have different work loads for your data the first approach is to use different CF's. You can make changes to the row / key cache to allocate more cache to the important things. The OS will take care of using the rest of the memory to cache the disk access.

In this case there is not much difference between multiple CF's and multiple keyspaces. 

If after tuning CF's you still see an issue (or you have really different workloads) the next step is to use different clusters. See this recent post from edward http://www.mail-archive.com/user@cassandra.apache.org/msg17576.html

Aaron Morton
Freelance Cassandra Developer

On 1/10/2011, at 5:18 AM, Ramesh S wrote:

Thank you Jonathan for your quick reply.
The reason I asked the question is because I have noticed some companies have used Cassandra to store data (which needs frequent read writes ) in-memory and have used other databases for data which is less frequently read/updated.
So this created my doubt in my design decision.
 Also we have a major portion of data which need not be in-memory . Right now we have a design where everything is in the same keyspace. 
So is there a way where we can may be swap that data, which is less demanding , to disk , by keeping in a different keyspace.


On Fri, Sep 30, 2011 at 10:30 AM, Jonathan Ellis <jbellis@gmail.com> wrote:
Short answer: no.

Longer: What problems do you hope to avoid by adding this complexity?

On Fri, Sep 30, 2011 at 9:45 AM, Ramesh S <investtrco@gmail.com> wrote:
> I have some data that would be read/write might be once during the entire
> session, like authentication info.
> When designing database with Cassandra is it advisable to keep the data that
> is not frequently read/written to the database else where ?
> Thank you
> regards,
> Ramesh

Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of DataStax, the source for professional Cassandra support