What is the best way to make use of all the cores of a multi-core machine running NDB?
My current set up is the following :
2 x Dual Quad Core for MySQL Heads raid 10, 15k RPM SAS drives 8G memory
6 x Dual Quad Core for Data Nodes raid 10, 15k RPM SAS drives 16G memory
The problem I see is that the ndbd and mysqld processes SEEM to only use one core on each of the machines. Am I correct?
Do I need to start up a new NDBD process for each core on each machine to take full advantage of the processing power of these machines?
Has anyone worked with this that might be able to advise?
Regarding the connections - most of the tests have been with a single connection. Load testing with multiple connections will be coming up in the next few weeks. I’ve been pretty impressed with the performance of MySQL Clustering, especially with large joins.
Besides making sure that similar nodegroups are not on the same machine, are there any other major gotchas that I should be aware of in running 2 ndbd threads on the same machine? It’ll double my data nodes from 6 to 12 - is that going to cause network problems on the data node dedicated high end, layer 2, gigabit switch or individual machines?
Looks like I finally got a reply about this!!! Here it is… Not a whole lot of info, but enough to keep me happy.
Posted by: Stewart Smith (IP Logged)
Date: March 22, 2007 02:32AM
It’s coming.
Jonas has been gradually working on it - both design and implementation. The design is pretty cool actually and should be incredibly efficient.
Currently it’s at a kind of pre-alpha stage - works for a little while with a bunch of limitations (e.g. create table doesn’t yet work ) - so progress is being made.
Can’t say which version it will appear in yet though.
Stewart Smith, Software Engineer - MySQL Cluster
MySQL AB, www.mysql.com