D.S. Nikolopoulos, T.S. Papatheodorou, C.D. Polychronopoulos, J. Labarta and Eduard Ayguad, "User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors", Proceedings of the 30th Anuual International Conference on Parallel Processing (ICPP 00), pp. 95-103, Vancouver (Canada), August 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the OS. Hot pages, i.e. pages to which a large number of misses are occurring, are migrated if referenced primarily by one process or replicated if referenced by many processes. Results of their experiments show a performance increase of up to 29 for some workloads. Nikolopoulos et.al. [5] present two algorithms for moving virtual memory pages to the nodes that reference them more frequently. The purpose of this page movement is the minimization of the worst case latency incurred in remote memory accesses. Their first algorithm works on iterative parallel programs and is based on ....

D. S. Nikolopoulos, T. S. Papatheodorou, C. D. Polychronopoulos, J. Labarta, and E. Ayguade. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. In Proceedings of the 29th International Conference on Parallel Processing, pages 95--103, Toronto, Canada, August 2000.

....the OS. Hot pages, i.e. pages to which a large number of misses are occurring, are migrated if referenced primarily by one process or replicated if referenced by many processes. Results of their experiments show a performance increase of up to 29 for some workloads. Nikolopoulos et.al. [5] present two algorithms for moving virtual memory pages to the nodes that reference them more frequently. The purpose of this page movement is the minimization of the worst case latency incurred in remote memory accesses. Their first algorithm works on iterative parallel programs and is based on ....

D. S. Nikolopoulos, T. S. Papatheodorou, C. D. Polychronopoulos, J. Labarta, and E. Ayguade. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. In Proceedings of the 29th International Conference on Parallel Processing, pages 95--103, Toronto, Canada, August 2000.

....if referenced primarily by one process or replicated if referenced by many processes. Results of their experiments show a performance increase of up to 29 for some workloads. This approach, however, relies on software instrumentation and hence introduces high overheads. Nikolopoulos et al. [13] present two algorithms for moving each virtual memory page to the node that performs the most references. The purpose of the page movement is to minimize the maximum latency due to remote memory accesses. One algorithm works with iterative parallel programs and is based on the assumption that the ....

D.S. Nikolopoulos, T.S. Papatheodorou, C.D. Polychronopoulos, J. Labarta, E. Ayguade, User-level dynamic page migration for multiprogrammed shared-memory multiprocessors, in: Proceedings of the 29th International Conference on Parallel Processing, Toronto, Canada, August 2000, pp. 95--103.

....does, but is intended only for the analysis of the interconnect trac with the goal to improve the modeling accuracy for network simulation systems. Data locality, on the other hand, has been addressed intensively since it has a severe in uence on performance of NUMA systems. Among these e orts [11, 4, 15, 2, 19, 16, 22], which are primarily based on compiler analysis and page migration. One of these e orts, especially closely related to the approach presented here, is the Dprof pro ling tool [2] developed by SGI. Dprof samples a program during its execution and records the program s memory access information as ....

D. Nikolopoulos, T. Papatheodorou, and etc. User-level dynamic page migration for multiprogrammed shared-memory multiprocessors. In Proceedings of the 29th International Conference on Parallel Processing, pages 95-103, Toronto, Canada, Aug. 2000.

No context found.

D.S. Nikolopoulos, T.S. Papatheodorou, C.D. Polychronopoulos, J. Labarta and Eduard Ayguad, "User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors", Proceedings of the 30th Anuual International Conference on Parallel Processing (ICPP 00), pp. 95-103, Vancouver (Canada), August 2000.

....the program running on these nodes. Aging is performed to avoid biasing future page migration decisions with obsolete reference history, thus increasing the chances of migrating pages that belong to the memory affinity sets of migrated threads. Details on the predictive algorithms are available in [9]. Table 1 summarizes the UPMLIB user level interface. Figure 2 gives an example of the use of UPMLIB in the NAS BT benchmark. 3 Implementation UPMLIB is implemented on the SGI Origin2000, using the IRIX 6.5.5 operating system interface. The runtime system is integrated with the NANOS OpenMP ....

D. Nikolopoulos et.al. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. Technical Report HPClab-TR-010100, High Performance Information Systems Lab., University of Patras, Submitted for Publication, January 2000.

....Our experimental results verify the effectiveness of the proposed framework and provide a proof of concept that it is not necessary to introduce data distribution directives in OpenMP and warrant the simplicity or the portability of the programming model. An earlier version of this paper [29] appeared in the IEEE ACM Supercompting 2000 Conference and won the Best Technical Paper Award. 1 1 Introduction Parallel processing is experiencing a convergence of both architectures and programming models into few well established paradigms [10] Parallel programming in particular seems to ....

....remote accesses to satisfy the criterion. The variable num migrations stores the number of migrations executed by the mechanism in the last invocation of upmlib migrate memory( and deactivates the mechanism when set to 0. 5 This case is not considered in this paper. The reader can refer to [29] for a treatment of the related issues. 16 . call upmlib init( call upmlib memrefcnt(u, size) call upmlib memrefcnt(rhs,size) call upmlib memrefcnt(forcing,size) do step=1,niter call compute rhs call x solve call y solve call z solve call add if ( step .eq. 1) or. ....

D. Nikolopoulos, T. Papatheodorou, C. Polychronopoulos, J. Labarta, and E. Ayguade. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. In Proc. of the

....enddo Figure 2: Usage of UPMLIB with NAS BT. avoid biasing future page migration decisions with obsolete reference history, thus increasing the chances of migrating pages that belong to the memory affinity sets of migrated threads. Details on the predictive algorithms are available in [9]. Table 1 summarizes the UPMLIB user level interface. Figure 2 gives an example of the use of UPMLIB in the NAS BT benchmark. 3 Implementation UPMLIB is implemented on the SGI Origin2000, using the IRIX 6.5.5 operating system interface. The runtime system is integrated with the NANOS OpenMP ....

....of the master thread which participates in the execution of parallel code. 4 Experimental Results In this section we provide a small set of experimental results that demonstrate the potential of UPMLIB. More results as well as thorough quantitative and qualitative analysis are available in [8, 9]. Figure 3 illustrates the performance of two application benchmarks from the NAS suite, BT and SP, both parallelized with OpenMP [5] The experiments were conducted on a 64 processor SGI Origin2000, with MIPS 4 Call Function upmlib init( upmlib end( UPMLIB initialization and termination. ....

D. Nikolopoulos et.al. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. Technical Report HPClab-TR-010100, High Performance Information Systems Lab., University of Patras, Submitted for Publication, January 2000.

....in the page migration engine. Competitive algorithms are used for standalone parallel programs, to tune the placement of pages across the nodes of the system according to the characteristics of the program. Aggressive page migration algorithms, such as predictive algorithms and page forwarding [8, 10], are used to cope with the effects of multiprogramming, which poses frequent migrations of threads between the nodes of the system. The aggressive page migration algorithms exploit scheduling information provided by the operating system with respect to the instantaneous mapping of threads to ....

D. Nikolopoulos, T. Papatheodorou, C. Polychronopoulos, J. Labarta and E. Ayguade. UserLevel Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. Submitted for Publication, January 2000.

....with a hysteresis of one iteration. The necessary migrations of pages are performed early and their cost is amortized well over the entire execution time. The fundamental difference with an explicit data 3 This case is not considered in this paper, but the reader can refer to a related paper [20] for details. 7 . call upmlib init( call upmlib memrefcnt(u, size) call upmlib memrefcnt(rhs,size) call upmlib memrefcnt(forcing,size) do step=1,niter call compute rhs call x solve call y solve call z solve call add if ( step .eq. 1) or. num migrations .gt. 0) then call ....

D. Nikolopoulos, T. Papatheodorou, C. Polychronopoulos, J. Labarta and E. Ayguade. User-Level Dynamic Page Migration for Multiprogrammed Shared-Memory Multiprocessors. Proc. of the 29th International Conference on Parallel Processing. Toronto, Canada, August 2000.

....implementation, the runtime system detects permanent thread migrations, that is, thread migrations that move a thread to a node for an amount of time sufficiently long to justify the activation of the page forwarding mechanism. The associated algorithms and implementation issues are available in [11]. Table 2. UPMLIB environment variables. Variable Functionality UMIGR POLICY Page migration criterion UMIGR THRESHOLD Competitive criterion threshold UMIGR PING PONG LIMIT Bouncing threshold for ping pong UMIGR SAMPLING PERIOD Period for the sampling based mechanism UMIGR PAGES PER SAMPLE ....

D. Nikolopoulos et.al. User-Level Dynamic Page Migration for Multiprogrammed SharedMemory Multiprocessors. To appear in the 29th International Conference on Parallel Processing. Toronto, Canada, August 2000.

No context found.

D. S. Nikolopoulos, T. S. Papatheodorou, C. D. Polychronopoulos, J. Labarta, and E. Ayguade. User-Level Dynamic Page Migration for Multiprogrammed SharedMemory Multiprocessors. In Proceedings of the 29th International Conference on Parallel Processing, pages 95--103, Toronto, Canada, August 2000.

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