Roger L. Adema and Carla Schlatter Ellis. Memory allocation constructs to complement numa memory management. In Symposium on Parallel and Distributed Processing, Dallas, TX, December 1991. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Region-Oriented Main Memory Management in Shared-Memory NUMA.. - Gamsa (1992) (6 citations) (Correct)
....[JWE91] This can be used to reevaluate past decisions, and removes the need for the expensive defrost daemon. A similar technique was also discussed by Unrau, Stumm, and Krieger [USK92] A method of segregating data to avoid false sharing among different objects is discussed by Adema and Ellis [AE91] Type information similar to that in the Munin system is used to ensure that different options with different access patterns are not co located on the same page. The intention is to have the user do the partitioning, and let the system manage locality. Malkawi and Patel discuss techniques for ....
Roger L. Adema and Carla Schlatter Ellis. Memory allocation constructs to complement numa memory management. In Symposium on Parallel and Distributed Processing, Dallas, TX, December 1991.
Hot Spot Analysis in Large Scale Shared Memory Multiprocessors - Karim Harzallah (1993) (4 citations) (Correct)
.... continue to outstrip improvements in memory performance, forcing future processors to spend a larger fraction of their time waiting for memory [NSTN91] BSF 91] Previous studies have shown that ignoring the location of data in NUMA systems leads to a significant performance penalty [SZ91] AE91] In this paper, we investigate the effect of a nonuniform traffic pattern consisting of a single hot spot. We analyze the performance impact of such contention, and consider, as one example architecture, the Hector system. Our analysis attempts to answer the following questions within the ....
....from which the request is made. We denote by C i the memory access time from level i. In our study, i = 1; 2; 3 correspond to local, on station, and off station access costs respectively. The model hardware parameters are summarized in Table 1. 2. 1 The Software Model Previous work [SZ91] AE91] has indicated that code and data placement is critical to good performance in NUMA systems. It is desirable that the data reside in the memory of the processor that is accessing it most frequently. This results in fewer remote accesses, and hence lower latency and less contention. Also, in ....
R. L. Adema and C. S. Ellis. Duke university, computer science department. Technical report, Memory Allocation Constructs to Complement NUMA Memory Management, 1991.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC