Hi, By avoiding to put the *transaction log dir* on a busy device, competing for disk I/O with the snapshot data and other systems (e.g, HDFS or HBase). The transaction log is critical to ZK performance because ZK syncs the transactions on disk before returning a response, according to http://zookeeper.apache.org/doc/r3.1.2/zookeeperAdmin.html. So, a fast disk is recommended, even if not that large. This principle of putting commit/transaction logs on dedicated devices is highly recommended by other big data systems like Cassandra, for example.
Regards, Ed On Mon, Oct 5, 2015 at 10:45 AM, Prabhjot Bharaj <[email protected]> wrote: > Hi, > > > I was going through the 'Other Optimizations' section in this link: > > https://zookeeper.apache.org/doc/r3.4.6/zookeeperStarted.html#Other+Optimizations > I would like to understand the relevance of 'dataLogDir' > > > How does having a different value of dataLogDir improve the performance ?? > > Regards, > Prabhjot >
