
Test Cycle: Testing for Performance
90
• Data and data distribution problems. Check data layout and be sure that
these match the productions system.
• Index definitions. Verify that your reindexing scripts worked. Using
sp_helpindex, compare the index definitions of the production and test
systems.
Resolve any problems before going on.
I/O
Verify that the test server is performing the same amount of I/O. Physical and
logical I/O should be in the same proportion as in the production system. Use
statistics io output to compare.
Pre-Upgrade Multi-User Tests
When the optimizer is behaving the same as it did in the production system,
you are ready to run the following benchmarks.
• Untimed Benchmarks
• Timed Benchmarks
Untimed Benchmarks
In multi-user mode, run untimed benchmarks as often as necessary, capturing
response time and throughput metrics. Resolve:
• Differences, such as cache hit rate, to the production system
• Bottlenecks caused by saturated devices
• Other problems or mistakes
Between runs, restore your databases from backup or perform a quick reset.
Timed Benchmarks
Always restore your databases from backup so they are in a known state when
running timed benchmarks. Gather response time and throughput metrics. Run
the tests as often as necessary to fix problems and be able to reproduce the final
results.
Commenti su questo manuale