Re: SQL Cluster Performane worse than MyISAM.

[Peter Zaitsev]

This is well expected.

It is wrongly assumed what as NDB Cluster is In Memory database its performance should be close to HEAP (In memory) tables.

There is a giant difference between data being located in the same process address space (HEAP) or in different process or over network (NDB), So even if you run NDB on the same node as MySQL Server for test performance will be lower.

The access to different process on the same box is expensive (especially as NDB uses TCP/IP, as this is configuration only for testing any way), access over ethernet network
is even more expensive.

How large performance overhead is a lot depends on how much round trips will be involved. For single row lookups the difference is smaller as query parsing and reciving from client takes large part. For joins it is much larger as there are a lot of network round trips has to be made.

Also MySQL Optimizer was mainly done for On Disk and In Memory tables - working with
"remote" tables requires different join methods which are in development now.