질문하기

포럼 검색:

• Microsoft 포럼 모두 검색

 

A problem about Linpack test on WCCS2003

• 2008년 5월 5일 월요일 오후 3:08songtao_eric

   
  

  0

  로그인하여 투표

  Hello, Everybody, I built a cluster with WCCS2003 , on which I am doing Linpack test. But there is a problem haunted me for a long time. My cluster has 3 nodes,  each node owns 4 Intel Xeon Dual Core processors and 16G RAM. Using the HPL.DAT below, I  did  Linpack tests successfully on every single node. However, after I only modified the Ps*Qs (e.g. 1*24 or 4*6) and assigned this task to 3 nodes,  the computing time was much longer than that on a single node.
  What's more, when I executed "mpipingpong.exe" on 3 nodes, I found that the cpu usage of all nodes kept 100% for several hours but I still did not get the result of pingpong test.  I do not know to solve this problem and wish your help.
  Does anybody meet the problem as mine ?

   

   

  HPLinpack benchmark input file
  Innovative Computing Laboratory, University of Tennessee
  HPL.out      output file name (if any)
  6            device out (6=stdout,7=stderr,file)
  1            # of problems sizes (N)
  10000        Ns
  1            # of NBs
  200          NBs
  0            PMAP process mapping (0=Row-,1=Column-major)
  1            # of process grids (P x Q)
  1             Ps
  4             Qs
  16.0         threshold
  1            # of panel fact
  1            PFACTs (0=left, 1=Crout, 2=Right)
  1            # of recursive stopping criterium
  4            NBMINs (>= 1)
  1            # of panels in recursion
  2            NDIVs
  1            # of recursive panel fact.
  2            RFACTs (0=left, 1=Crout, 2=Right)
  1            # of broadcast
  1            BCASTs (0=1rg,1=1rM,2=2rg,3=2rM,4=Lng,5=LnM)
  1            # of lookahead depth
  0            DEPTHs (>=0)
  2            SWAP (0=bin-exch,1=long,2=mix)
  80          swapping threshold
  0            L1 in (0=transposed,1=no-transposed) form
  0            U  in (0=transposed,1=no-transposed) form
  1            Equilibration (0=no,1=yes)
  8            memory alignment in double (> 0)

  • 응답
  • 인용