I have just setup the cluster and i started to test the performance and it's
very low in comparison to a simple mysql server.
Ever query that i tested takes 10 times more to execute.
Here is the configuration
the cluster is on 3 boxes each with 3.2 Dual Xeon with 4G RAM
each box runs 2 data nodes and 1 sql node.
the sql dump of the database has about 1G
The config has the following
NoOfReplicas= 2
MaxNoOfConcurrentOperations= 100000
DataMemory= 1500M
IndexMemory= 64M
TimeBetweenWatchDogCheck= 30000
DataDir= /var/lib/mysql-cluster
MaxNoOfAttributes = 2000
MaxNoOfOrderedIndexes = 5000
MaxNoOfUniqueHashIndexes = 5000
NoOfFragmentLogFiles = 32
StopOnError = N
RedoBuffer = 128M
each node uses:
Data usage is 74%(35683 32K pages of total 48000)
Index usage is 33%(2775 8K pages of total 8224)
my.conf has this:
sort_buffer_size=256M
using this formula: (SizeofDatabase * NumberOfReplicas * 1.1 ) /
NumberOfDataNodes " (1000 * 2 *1.1) / 6 "
each node should be using aprox 360M and yet it using 74% from 1500M =>1110M
Where is my calculation wrong?
free -m shows this (almost 1G of swap):
total used free shared buffers cached
Mem: 3933 3914 18 0 109 1000
-/+ buffers/cache: 2804 1129
Swap: 2000 938 1061
Even with the database being swapped is it normal for the cluster to be so
slow. I mean i compared the speed on a simple mysql ( 2.4 dual xeon with 4G
RAM ) and the cluster ( 3x 3.2 Dual xeon with 4G RAM )
Supose i upgrade the ram on the cluster to 8G each, will that speed up the
cluster 10 times to get it atleast to the performance of a single mysql
server???
very odd ...
