|
|
NUMA Versus SMP: A Case Study on AMD Opteron
AMD Opteron(tm) can be configured to place DRAM in a Shared-Memory or a Cache-Coherent Non-Uniform Memory organization. This allows us to experiment with various Linux Kernel and Memory allocation optimizations for both and contrast their benefits with various workloads. Rich Brunner's presentation will begin with a brief overview of the memory-hierarchy in AMD Opteron. We will then review the current state of SMP and NUMA support in the current Linux Kernel. This will be followed by a case-by-case review of each experiment, its code patches, and its results. We will conclude with some recommendations for NUMA and SMP directions in the future.
|
|
