Re: Test cluster slower than single machine

[Brad Crotchett]

Thanks Stewart. I still think I am missing something major here. In my latest test I decided to use a single SGI so I could eliminate the network as an issue. I ran 2 ndb nodes and 6 sql nodes on one of the 16 ia_64/16 GB ram machines. It taxed the machine a little more than my other tests (ran at a load of about 4.50) but the TPS were still low at 39.

Here is my config.ini:

[NDBD DEFAULT]
NoOfReplicas=2
DataMemory=2048MB
IndexMemory=2048MB
Diskless=1

[TCP DEFAULT]
PortNumber=2202

[NDB_MGMD]
id=1
hostname=xxx.xxx.xxx.19
DataDir=/var/lib/mysql-cluster

[NDBD]
hostname=xxx.xxx.xxx.19
id=11
DataDir=/var/lib/mysql-cluster/1

[NDBD]
hostname=xxx.xxx.xxx.19
id=12
DataDir=/var/lib/mysql-cluster/2

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=31

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=32

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=33

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=34

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=35

[MYSQLD]
hostname=xxx.xxx.xxx.19
id=36

And here is my my.cnf:

[MYSQLD]
ndbcluster
ndb-connectstring=xxx.xxx.xxx.19
max_connections=500
engine_condition_pushdown=1

[MYSQL_CLUSTER]
ndb-connectstring=xxx.xxx.xxx.19

The other 5 sql nodes are started with the same arguments as the main my.cnf.

I am sure that mysql cluster can perform better than what I am getting. Any help as to what I am missing would be greatly appreciated.

Thanks,

Brad