- When all CPUs were busy, the numad daemon was too reluctant to balance processes across nodes even though it would have resulted in significantly better application and system performance. With this update, numad is more aggressive about moving processes even when all CPUs are busy. As a result, overall system performance has improved significantly.
- Previously, it was not possible to set the system's hugepage "scan_sleep_millisecs" parameter. As a consequence, NUMA performance was damaged when process memory was migrated across nodes. The underlying code has been changed to accept the new "-H" option to specify "scan_sleep_millisecs", thus fixing the bug. The user can now set the value to fine-tune the numad's performance.
- A configuration file for the logrotate tool has been added to be fully supported by Red Hat Enterprise Linux 6.