R
Rob Fitzpatrick
Guest
It looks like life is better in 11.3.2. My preliminary test results: test: 11.3.2 sports DB, all application objects assigned to ABP at object level do more I/O on ABP than value of -B2 expected result: LRU2 policy will be enabled there will be LRU2 latch activity and more ABP OS reads than value of -B2 steps: convert sports db to v11 proserve sports -B2 10 run program to read all assigned tables with all assigned indexes result: 690 OS reads on ABP r&d 6 11 shows lots of LRU2 activity r&d 2 3 shows LRU2 policy enabled r&d 4 4 7 shows "disable" conclusion: expected result for 11.3.2 test: resize -B2 online to be large enough to hold all assign objects read all ABP-assigned objects expected result: LRU2 policy remains disabled no LRU2 activity steps: disable LRU2 policy proutil increaseto -B2 700 zero stats in promon run program to read all assigned tables with all assigned indexes result: r&d 2 3 shows 75 ABP OS reads, LRU2 policy enabled r&d 6 11 shows 7367 LRU2 latch locks r&d 4 4 7 shows "disable" conclusion: -B2 was not sized large enough? there were 75 ABP OS reads, so at most -B2 was 75 blocks too small test: resize -B2 online again, increasing it to 800 (100 larger than before) read all ABP-assigned objects expected result: LRU2 policy remains disabled no LRU2 latch activity steps: disable LRU2 policy proutil increaseto -B2 800 zero stats in promon run program to read all assigned tables with all assigned indexes result: r&d 6 11 shows 0 LRU2 latch locks r&d 4 4 7 shows "enable" r&d 2 3 shows 7367 logical reads, 0 OS reads in ABP, LRU2 policy disabled conclusion: expected result in 11.3.2
Continue reading...
Continue reading...