Under certain load all available history syncers get stuck at 100% load and zabbix database start to lag behind. Changing number of history syncer processes make things only go worse, doesnt matter if we increase them to 8 or decrease to 2 from default 4
Current production server is running on AMD Opteron 8439 SE box (6x3 cores), and starts to lag at ~3000 NVPS. I temporarily moved zabbix_server to box with single Intel Xeon E5620 (8 cores) and everything worked like charm. I was able to bring NVPS up to 7500 without any issues. All other components - SQL, proxies stayed in place without any changes.
Is it possible that there are some AMD cpu specific issues in code which can cause such problem?
Or second possibility that history syncer process is able to use only single core for its computations, since it runs better on hardware with lesser but more powerful cores? But then adding more processes should help, which didnt happen.