At LinkedIn, some of  the high volume topics are configured with more than
1 partition per broker. Having more partitions increases I/O parallelism
for writes and also increases the degree of parallelism for consumers
(since partition is the unit for distributing data to consumers). On the
other hand, more partitions adds some overhead: (a) there will be more
files and thus more open file handlers; (b) there are more offsets to be
checkpointed by consumers which can increase the load of ZK. So, you want
to balace these tradeoffs.



On Mon, Jan 14, 2013 at 11:55 PM, Andrew Psaltis <

NEW: Monitor These Apps!
elasticsearch, apache solr, apache hbase, hadoop, redis, casssandra, amazon cloudwatch, mysql, memcached, apache kafka, apache zookeeper, apache storm, ubuntu, centOS, red hat, debian, puppet labs, java, senseiDB