G
George Potemkin
Guest
I run the test with –nosavepoint: no disk IO, particularly during MTX locks. The highest MTX locks per sec were at 20 sessions: 600,000-700,000 locks/sec vs. 200,000 locks/sec in the tests with BI writes. Naps start growing with 23 sessions and reached the max with 50 sessions: 1,600 naps/sec vs. 2,600 naps per sec at 100 users in previous test. Conclusion: disk IO (BI writes) does increase the duration of MTX latch locks. Only two latches were locked: MTX and TXT. Zero naps of TXT latch. But I don't see that both latches were locked by the same session.
Continue reading...
Continue reading...