This is a test environment for a future production environment.I've got two Physical systems running 1 vmware guest each.I'm using the HammerDB tpc-c tests for this.My process is to use hammerdb to build the tpcc database, resize all the files so there is no filegrowth during the test, no checkpoints until it's complete, then a timed 10 minute test, 8 warehouses, 8 users.This test results in roughly 140k tpm.Then I spin up the availability group between the two nodes and rerun the test with the AG in async mode. I get roughly the same 140k TPM results.Turn on the synchronous commit for both replicas and I get 1386 TPM. That's less than 1% of the performance in a normal or an async configuration.This can't be normal. Someone please tell me what I'm doing wrong.More information:When I look at wait types it looks like this after a typical run, in order of most to least.CXPACKET (CTFP=5, MAXDOP=0)LATCH_EXHADR_SYNC_COMMIT (this is the evidence of it waiting on something I guess)WRITELOG
↧