Hi,

I am working on a benchmarking project involving OLTP transactions on DB2 9.7.We'll be testing performance of DB2 on 500 million customer accounts with a minimum target of 10000 transactions per second.We'll have 4 database servers (AIX),8 application servers(HP) and 2 servers to generate transactions.The size of database would be approx. 16 TB.
I have to implement something similar to RAC of oracle ensuring high performance (because this project is mainly DB2 vs oracle type of project and personally I want db2 to outstand in this testing ).Please suggest me which kind of database partitioning I should implement a) multi partition (shared nothing architecture) or b) clustered configuration (where storage where database is stored is shared amongst all partitions).Performance wise which one would be better?

You can also mail me to shuchi.mishra@tcs.com

Thanks in advance.

Study cases like this: Transaction Processing Performance Council
It shows the IBM approach.

Hi,

what do you mean by "b) cluster configuration)"?
A "big Iron" with logical partitions? Partitioning in DB2 always is shared nothing.

Lately DB2 PurScale (IBM - DB2 pureScale Feature) was announced. That's something like a Oracle RAC. PureScale is derived from the Mainframe (DataCouplingFacility and the other stuff).

Good Luck with the Benchmark! Sounds very intresting.