-
Type: Task
-
Resolution: Done
-
Affects Version/s: None
-
Component/s: None
-
None
Today I ran various tests with the voxer configuration. Specifically the one that uses value sizes of 130K. The workload is 20 threads doing a read:update ratio of 1:1.
The first config I ran had snappy off and it had one 30-second drop out with 0 throughput about 600 seconds into a 1 hour test (it really is supposed to run 5 hours). This one shows similarities to the test3 issues where we are not merging files and I ended up with 5000+ chunks, 3000+ bloom filters and 180000 open cursors. Surprisingly it did not get killed by the oom-killer.
The second config I ran had snappy on and initially I had high hopes. Throughput was 50% faster and I didn't see any drop offs by 1000 seconds. Well, at 1300 seconds into the test it dropped off to 0 for 3 minutes. Then again at 1750 for another 3 minutes. The one at 2250 lasted 5 minutes.
The good news is that I was running my pmp, perf record, etc script for the first 1500 seconds of both tests and stats are on. It will take a bit to wade through the information.