performance of ndbcluster (no replies)

hi all experts

I have a 4 Servers cluster-4 data nodes and 4 sql nodes.(There is a SQL node and a data node in a physical server)

The statement of [ndbd_defaults] in my config.ini follow as:

[NDBD DEFAULT]

NoOfReplicas=2
DataMemory=1024M
IndexMemory=512M
DataDir=/data/data
BackupDataDir=/data/backup
FilesystemPath=/data/fspath
LockPagesInMainMemory=1

FragmentLogFileSize=256M
NoOfFragmentLogFiles=16
RedoBuffer=64M

MaxNoOfUniqueHashIndexes=512

TransactionInactiveTimeout=50000

MaxNoOfExecutionThreads=8

BatchSizePerLocalScan=512

MaxNoOfConcurrentTransactions=10000
MaxNoOfConcurrentOperations=32768
MaxNoOfLocalOperations=40000


The statement of [mysqld] in my my.cnf follow as

[mysqld]
ndbcluster
basedir=/mysql
datadir=/data/sql
ndb-connectstring=10.45.26.102
ndb-connectstring=10.45.26.104
thread_concurrency = 8
max_connections=1000

When I startup the mysqld , I found the warning:

[Warning] Changed limits: max_open_files: 1024 max_connections: 214 table_cache: 400


I create two database, db1 and db2,they have the same data.

All the tables in db1 was defined as ndbcluster-table and all tables in db2 was defined as innodb-table.


I find the speed of Query (only select SQL) used in ndb-table was more slower than used in innodb.

And then I used the sysbench to test the performance of the ndb and innodb.

The result as this (r/w) :

ndb only used one SQL node transactions:210 per sec

ndb only used all SQL node together transactions:560 per sec(total) 140 sec(avg)

innodb transactions:590 per sec


The internal net used for cluster is GbitEthernet.



BTW，I used the ndbd for data node. When I try use the ndbmtd I find the process cost the more resource (cpu and memory) than ndbd.So I don't know why.


I don't know the problem of my cluster. I don't think the performance of ndbcluster is so terrible.