PPOPP '91 Proceedings of the third ACM SIGPLAN symposium on Principles and practice of parallel programming
Experimental comparison of memory management policies for NUMA multiprocessors
ACM Transactions on Computer Systems (TOCS)
| Hi-index | 0.00 |
Non-uniformity of memory access is an almost inevitable feature of the memory architecture in shared memory multiprocessor designs that can scale to large numbers of processors. One implication of NUMA architectures is that the placement and movement of code and data become crucial to performance. The additional complexity of a main memory architecture with a distinction between local and remote memory demands that the operating system assume a larger role in providing memory management support. When one extends the investigation of NUMA memory management techniques found in the literature to consider limitations on the sizes of various main memory components and to include transparent management of the main/secondary memory hierarchy, the issues become more complex. This paper elaborates on some of these issues and proposes a dynamic page placement policy based on frame borrowing and page recall as a partial solution. The extension of our algorithm to attack the more general migration problem is also discussed. Results of experiments designed to evaluate the approach are presented.