RAC Issue with high Transaction OLTP + GC's

Question

In the Lab i am testing in-house application with Oracle 10g.
Setup
- 2 Database servers (RHEL OS 3, IBM H/W, 4GB RAM)
- 2 in house Application servers (RHEL OS 3, IBM H/W, 4GB RAM)
- IBM Disk Array with 2Gbps fiber Cable Speed
- Cisco LAN Switch with 100mbps/sec Port
Application Detail-
-- Its a OLTP application.
-- Doing lots of insertion,update,delete,commit for each transaction.
-- With high application load,its doing 5000 DB tranactions/sec with database.
The following test scenario need to test
1. Start one application server and conenct with DB instance1 and push high load..
2. Start both application server and connect with DB Instance1b and push high load from both applications.
3. Start Both application server and connect with both DB servers ( For Ex.
(A1-DB1 & A2-DB2) and check the application compatibility with RAC environment ( Note- Here need to push very less db transaction data to DB servers).
4. Similar like 3 but with very high db transaction rate
Now these are the results
1. Full Success
2. Full Success
3. Full Success
4. Fail
Detail description of Case4 Failure.
When ever i am starting applications with high load, within few minutes application quesues are getting full and db instances are not able to handle load.
When i checked the Oracle ASH Report and then found following major events
1. gc buffer busy
2. gc current block busy
3. gc current multi block request
4. gc cr block 2-way
i checked the oracle documents and found three are major root cause for this issue.
-- Interconnect
-- Load issues
-- SQL Execution against a Large shared working set
But i am not sure, how i can come out from this issue.
If any one faced this issue with Oracle RAC. Please share your experience or give some suggestion to come out from this issue.
Rgds
Sumit
Bangalore,India

Answer 1

Hi,
you can use fiber for interconnect, Once i have looked for this option but haven't implemented.......
in this case your global cache requests are too much. one thing more if is it possible then divide the load according to domains. like same queries on on node and other domain queries on second nodes...... e.g. reporting on node2 and oltp transations on node1.
and kindly look into it...
http://www.ardentperf.com/2007/09/12/gc-buffer-busy-waits-in-rac-finding-hot-blocks/