Re: Server load on data node

[David Rusenko]

To preface this, I'm not entirely sure these are all perfectly correct, but they all seem to be reasonable given the research I've done today. A short background: We are in a somewhat low-volume read/write scenario, but the importance is the redundancy, and scalability, versus the increased performance over a single database instance.

config.ini values:

[NDBD DEFAULT]
NoOfReplicas=2
DataMemory=300M
IndexMemory=100M
TimeBetweenLocalCheckpoints=20 #4MB writes
NoOfFragmentLogFiles=8 #512MB of REDO log -- reasonable, and we could most likely do with less
RedoBuffer=16M #We had some issues with RedoBuffer being overload, so we upped it
# MaxNoOfAttributes - Number of columns over all tables
# These are somewhat arbitrary, although make sense for the most part
MaxNoOfAttributes=20000
MaxNoOfConcurrentOperations=50000
MaxNoOfLocalOperations=50000
MaxNoOfOrderedIndexes = 5000
MaxNoOfUniqueHashIndexes = 5000

We also tuned the my.cnf file a bit, main changes were:
thread_cache_size = 8
ndb_force_send = 0

The biggest change I can tell that helped reduce the performance drag was reducing NoOfFragmentLogFiles to 8 from 200 (a rather ridiculous value).

HTH,

David