Re: Advice about how do distribute components across servers?

[Mikael Ronström]

The setup you have provided looks quite ok.
We have even provided the READ_BACKUP feature in 7.5 GA
that makes it possible to ensure that the mysqld uses
the local ndbmtd to get its data, this avoids the network
becoming a bottleneck.

In truly scalable scenarios with tens of MySQL Servers and
tens of data nodes then it doesn't pay off so much to
colocate mysqld and ndbmtd.

In this application I gather that the application is mostly
reads, so even the new FULLY_REPLICATED feature could be
useful. In this case one would colocate up to 48 mysqld and
ndbmtd's and scale 2 nodes at a time where all data is present
on all ndbmtd's.

Note that the ndb_mgmd doesn't require a lot of CPU resources.
But it should not be on the same server as the data nodes to
ensure that it can work as arbitrator to avoid network partitioning.