27 June, 2007

Some observations from the latest Oracle-HP Benchmark

A quick read of the latest Oracle-HP Benchmark (world record TPC-C, using Single Processor, Oracle Enterprise Linux, Oracle 10g SE-One) shows some curious things :
1. Oracle has priced for the Xeon QuadCore Processor at the rate of 1 Processor based on
the single socket justified as "When licensing Oracle programs with Standard Edition One or Standard Edition in the product name, a processor is counted equivalent to an occupied socket" for a 3-year licence. Thus, Oracle used the combination of "Processor, not Core" for SE/SE-One and 50% of List price for a 3-Year Licence.
2. This provides us a hint that we could go for a 2 QuadCore Processor server and pay for only 2 CPUs for the SE-One licence.
The licence costing for 1 Processor for 3 years at 50% did help Oracle achieve "a new world record for price/performance"

3. For the benchmark, the there were only 2 Online Redo Logs but of 12,817MB each ! Huge ! Obviously, log_checkpoint_interval and log_checkpoint_timeout were both 0.
Checkpointing only at the 12GB log switch.
4. DB_BLOCK_SIZE was 2048, not the "standard" 8192.
5. DB_CACHE_SIZE was 2000M but DB_KEEP_CACHE_SIZE was 16,500M.
6. The Benchmark used "ANALYZE TABLE/INDEX ESTIMATE STATISTICS" instead of the normally recommended "DBMS_STATS..."
7. Database Recovery was using the SQLPlus "RECOVER" command, _not_ with RMAN.
8. "Demonstration" of recovery from loss of a log file was actually demonstrating that "
system continued running because the logs are mirrored" . They didn't have to show Recovery in case of loss of online/active redo logs !

No comments: