Re: performance of aggregate in Cluster

[Mikael Ronström]

Hi,
The aggregate query fetches all 10M rows from the data nodes, for each row
it gets the two columns you specified, so probably around 70-80 bytes sent
per row. So the MySQL Server have to process 10M messages of 70-80 bytes in
size, this will take a few seconds, so it is quite hard to improve it.

But one thing you can do is to ensure that you use the option to lock the
NDB receive thread to a specific CPU which nobody else is using. If you
can get this configuration working it is possible that you can improve
the results of this query.

Rgrds Mikael Ronström