Thanks for reply. Now it looks much more clear. Top shows this: PID USER PR NI VIRT *RES* SHR 67 cassandr 18 0 6267m *1.6g * 805m S 0.3 79.0 24:35.80 java
It's Ok with huge VIRT memory since it's 64 bit architecture. But RES is keep growing. And I still have questions: 1) If I don't have swap on node, then every mmap will add usage to my real RES memory that is limited with 2GB? 2) Is there any way to limit mmap's RES memory usage? 3) Any common solution of problem when memory is out because of mmaps and Cassandra process killed by OS? I guess switching disk_access_mode to standard will solve this problem, but it decrease performance. Evgeny.
