One issue that I found in using load balancer in front of cassandra nodes is that a single node might become bogged down by compaction, or other actions unrelated to the client.  If the load balancer does not pick this up in time, it might route client requests to the node that is temporarily overloaded.  

In practice, I've found it better for the client to have a pool of connections, and then retry as needed to distinct nodes rather than use a load balancer.

Dave Viner

On Tue, Apr 5, 2011 at 9:51 AM, A J <s5alye@gmail.com> wrote:
Can someone comment on this ? Or is the question too vague ?


On Wed, Mar 30, 2011 at 3:58 PM, A J <s5alye@gmail.com> wrote:
> Does the following load balancing scenario look reasonable with cassandra ?
> I will not be having any app servers.
> http://dl.dropbox.com/u/7258508/2011-03-30_1542.png
> Thanks.