Re: Query faster 10 times when stop on data node

[Andrew Morgan]

When running with 2 data nodes, each data note processes all reads and writes for 1 of the 2 partitions; if the query needs data from both partitions then this means that both data nodes need to do part of the work and there needs to be passing of messages between them - this can mean extra latency.

When only 1 data node is running, it's the primary for both partitions (all of the data) and so no messaging is involved and so latency is reduced.

You can optimise for this by using a 'shard aware' design - this white paper: http://www.mysql.com/why-mysql/white-papers/guide-to-optimizing-performance-of-the-mysql-cluster/


Andrew.