Computer Science ›› 2014, Vol. 41 ›› Issue (6): 113-118.doi: 10.11896/j.issn.1002-137X.2014.06.023

Previous Articles     Next Articles

Optimizing Algebraic Multigrid on NUMA-based Cluster System

GU Jian and LIU Wei   

  • Online:2018-11-14 Published:2018-11-14

Abstract: Algebraic Multigrid (AMG) as a key method of many numerical simulation programs,a serious scalability problem is encountered when it scales on mulit-core NUMA based cluster.There is a problem that data tend to be allocated nearby the CPU core on which the main thread are running,and the data lack locality.By applying our NUMA-aware memory allocator,each partition of a data block can be binded to the corresponding NUMA memory nodes which the data owner thread is laid on,so as to successfully maintain more data locality when using OpenMP parallelizing AMG,and make BoomerAMG scaling up more easily and efficiently.In the experiment on a single node and a small 16nodes cluster,using NAAlloc allocator gets 16% and 60% peak performance improvement respectively.

Key words: Algebraic mulitgrid,NUMA,Multic-core,Locality

[1] 裴文兵,徐小文.国家863课题“激光聚变领域的高性能计算应用研究”进展报告[R].北京应用物理与计算数学研究所,2011
[2] Simon H.Progress in Supercomputing:The Top Three Breakthroughs of the Last 20Years and the Top Three Challenges of the next 20Years[M].Pennyhill Press,2014
[3] Office of Science,DOE.Top Breakthroughs in ComputationalScience.http://www.scidacreview.org/0901/html/bt.html
[4] Holter W.A vectorized multigrid solver for the three dimensional poisson equation[J].Appl.Math.and Comput.,1986,19(1-4):127-144
[5] Falgout R D.An Introduction to Algebraic Multigrid,Computing in Science and Engineering[J].Special Issue on Multigrid Computing,2006(8):24-33
[6] Falgout R,Brannick J,Brezina M,et al.New Multigrid Solver Advances in TOPS[C]∥Proceeding of SciDAC 2005,Journal of Physics:Conference Series,Institute of Physics.San Francisco,CA,2005
[7] McByan O,Frederickson P,Linden J,et al.Multigrid methods on parallel computers-A survey of recent developments[J].Impact of computing in science and engineering,1991,13:1-75
[8] Chan T,Saad Y.Multigrid algorithms on the hypercube multigrid processors[J].IEEE Trans.Comput.,1986,35(11):969-977
[9] McByan O.The SUPRENUM and GENESIS projects[J].Parallel Computing,1994,20:1389-1396
[10] Gahvari H,Baker A H,Schulz M,et al.Modeling the Performance of an Algebraic Multigrid Cycle on HPC Platforms[C]∥Proc.of the 25th International Conference on Supercomputing (ICS 2011).Tucson,AZ,2011
[11] Baker A H,Falgout R D,Gamblin T,et al.Scaling Algebraic Multigrid Solvers:On the Road to Exascale[C]∥Proc.of Competence in High Performance Computing,CiHPC 2010.Schwetzingen Germany,2010
[12] Baker A H,Falgout R D,Kolev T V,et al.Scaling hypre’s Multigrid Solvers to 100000Cores[C]∥High Performance Scientific Computing:Algorithms and Applications.2012:261-279
[13] Baker A H,Gamblin T,Schulz M,et al.Challenges of Scaling Algebraic Multigrid Across Modern Multicore Architectures[C]∥Proc.of the 25th IEEE International Parallel and Distributed Processing Symposium (IPDPS 2011).Anchorage,AK,2011

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!