[petsc-users] MPI bindings for PETSc local and global numbering

[Justin Chang]

Hi all,

I am experimenting with mpiexec.hydra's binding options for the following
compute node:

Machine (64GB total)
  NUMANode L#0 (P#0 32GB)
    Socket L#0 + L3 L#0 (25MB)
      L2 L#0 (256KB) + L1d L#0 (32KB) + L1i L#0 (32KB) + Core L#0 + PU L#0
(P#0)
      L2 L#1 (256KB) + L1d L#1 (32KB) + L1i L#1 (32KB) + Core L#1 + PU L#1
(P#1)
      L2 L#2 (256KB) + L1d L#2 (32KB) + L1i L#2 (32KB) + Core L#2 + PU L#2
(P#2)
      L2 L#3 (256KB) + L1d L#3 (32KB) + L1i L#3 (32KB) + Core L#3 + PU L#3
(P#3)
      L2 L#4 (256KB) + L1d L#4 (32KB) + L1i L#4 (32KB) + Core L#4 + PU L#4
(P#4)
      L2 L#5 (256KB) + L1d L#5 (32KB) + L1i L#5 (32KB) + Core L#5 + PU L#5
(P#5)
      L2 L#6 (256KB) + L1d L#6 (32KB) + L1i L#6 (32KB) + Core L#6 + PU L#6
(P#6)
      L2 L#7 (256KB) + L1d L#7 (32KB) + L1i L#7 (32KB) + Core L#7 + PU L#7
(P#7)
      L2 L#8 (256KB) + L1d L#8 (32KB) + L1i L#8 (32KB) + Core L#8 + PU L#8
(P#8)
      L2 L#9 (256KB) + L1d L#9 (32KB) + L1i L#9 (32KB) + Core L#9 + PU L#9
(P#9)
    HostBridge L#0
      PCIBridge
        PCI 1000:0087
          Block L#0 "sda"
      PCIBridge
        PCI 15b3:1003
          Net L#1 "eth0"
          Net L#2 "ib0"
          OpenFabrics L#3 "mlx4_0"
      PCIBridge
        PCI 8086:1521
          Net L#4 "eth1"
        PCI 8086:1521
          Net L#5 "eth2"
      PCIBridge
        PCI 102b:0533
      PCI 8086:1d02
  NUMANode L#1 (P#1 32GB) + Socket L#1 + L3 L#1 (25MB)
    L2 L#10 (256KB) + L1d L#10 (32KB) + L1i L#10 (32KB) + Core L#10 + PU
L#10 (P#10)
    L2 L#11 (256KB) + L1d L#11 (32KB) + L1i L#11 (32KB) + Core L#11 + PU
L#11 (P#11)
    L2 L#12 (256KB) + L1d L#12 (32KB) + L1i L#12 (32KB) + Core L#12 + PU
L#12 (P#12)
    L2 L#13 (256KB) + L1d L#13 (32KB) + L1i L#13 (32KB) + Core L#13 + PU
L#13 (P#13)
    L2 L#14 (256KB) + L1d L#14 (32KB) + L1i L#14 (32KB) + Core L#14 + PU
L#14 (P#14)
    L2 L#15 (256KB) + L1d L#15 (32KB) + L1i L#15 (32KB) + Core L#15 + PU
L#15 (P#15)
    L2 L#16 (256KB) + L1d L#16 (32KB) + L1i L#16 (32KB) + Core L#16 + PU
L#16 (P#16)
    L2 L#17 (256KB) + L1d L#17 (32KB) + L1i L#17 (32KB) + Core L#17 + PU
L#17 (P#17)
    L2 L#18 (256KB) + L1d L#18 (32KB) + L1i L#18 (32KB) + Core L#18 + PU
L#18 (P#18)
    L2 L#19 (256KB) + L1d L#19 (32KB) + L1i L#19 (32KB) + Core L#19 + PU
L#19 (P#19)

I am currently using up to 16 processes and have the following:

${MPIEXEC}.hydra -n <procs> -bind-to hwthread -map-by socket ./myprogram
<args>

For 16 processes, the binding looks like this:

process 0 binding: 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 1 binding: 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
process 2 binding: 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 3 binding: 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
process 4 binding: 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 5 binding: 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
process 6 binding: 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 7 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
process 8 binding: 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 9 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
process 10 binding: 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 11 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
process 12 binding: 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
process 13 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
process 14 binding: 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
process 15 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0

where the ranks are staggered (but evenly distributed) among the two
sockets. However, if I grouped the ranks sequentially like this:

process 0 binding: 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 1 binding: 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 2 binding: 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 3 binding: 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 4 binding: 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 5 binding: 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
process 6 binding: 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0
process 7 binding: 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0
process 8 binding: 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
process 9 binding: 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0
process 10 binding: 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
process 11 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0
process 12 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0
process 13 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0
process 14 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0
process 15 binding: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0

Would this be more efficient? I was wondering if, in the context of
local/global numbering for PETSc's matrices and vectors, this would make
any difference. I have used ParMETIS to distribute/organize my DMPlex.

Thanks,
Justin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150511/105926a6/attachment.html>