D
dbeavon
Guest
On an evening when I'm the only one in the system, the *first* test does *not* hammer on the CPU at 100% and it *does* take 10 seconds longer (the duration is a full 15 seconds of time, and the CPU jumps around erratically between 50% and 100%.). After the *first* test, the subsequent tests use the CPU at 100% and take only five seconds. Not having a very good way to clean the database buffers, I cannot simulate a "first" test in a repeatable way. But I have done my best to rule out every other possibility, and the only thing left seems to be the disk I/O's to the underlying database files. I like the idea of running an online dbanalys against a large area with data that are not used in my tests. I'll report back if that gives me the consistent and reproducible 15 second results (ie a repeatable "first" test).
Continue reading...
Continue reading...