In the demanding world of today’s IT environments, users expect rapid responses to requests and greater utilization of their computing platforms. In order to meet these requirements, Red Hat’s Performance Engineering Team devotes itself to analyzing these technology concerns, building automated optimization tools for Red Hat Enterprise Linux (RHEL) and creating useful tuning guides to help users customize configurations for their varied workloads.
In front of a packed room, John Shakshober and the Performance Engineering Team demonstrated RHEL’s performance analysis and tuning tools. With conversations covering cgroup controlled resource management, tuned performance profiles, realtime kernel principles, hugepage memory allocation, and even Non-Uniform Memory Access (NUMA) node balancing, there was something for everyone.
Of particular focus was NUMA node balancing and how it has evolved with RHEL over the years. NUMA itself refers to the assignment of blocks of memory to a specific microprocessor, allowing the processor fast, direct access to those local memory locations. This approach provides significant performance benefits, but obviously, these are best realized if effort is made to keep processes correctly aligned with their associated memory.
The Performance Engineering Team detailed how this is accomplished in both RHEL 6 and RHEL 7, explaining that in RHEL 6, this functionality is achieved with a daemon known as “numad”. The numad daemon automatically moves processes to CPU cores in the same NUMA node as their associated data, drastically improving overall system performance. In RHEL 7, this functionality has been built directly into the kernel and no longer requires an additional daemon to achieve the same performance benefits.
For more on NUMA node balancing, as well as the other topics covered in this session, please watch the video on our YouTube page!