------------------ 原始邮件 ------------------

发件人: "Balaji, Pavan";<balaji@anl.gov>;

发送时间: 2014年2月19日(星期三) 中午12:38

收件人: "discuss@mpich.org"<discuss@mpich.org>;

主题: Re: [mpich-discuss] Communication Error when installing MPICH onmulti HOSTS.

It’s hard to tell, but this does indicate some problem with your communication setup. Did you verify your /etc/hosts like described on the FAQ page?

— Pavan

From: 维洛逐风 <wu_0317@qq.com>
Reply-To: "discuss@mpich.org" <discuss@mpich.org>
Date: Tuesday, February 18, 2014 at 5:21 AM
To: discuss <discuss@mpich.org>
Subject: [mpich-discuss] Communication Error when installing MPICH on multi HOSTS.

HI.

My environment:

Two Vmware VMs with ubuntu-server12.04 OS, called mpimaster,mpislaver1

they both linked to a virtual network 10.0.0.1;

they can ssh to each other without password;

I have disabled the fire walls with "sudo ufw disable"

I install mpich3.0.4 on a NFS servered by mpimaster.

I installed mpich3.0.4 follow the "readme.txt", it has Communication problem when progresses from different host comunicate with each other.

From picture above we can see it's ok to run "cpi" on both hosts separately.

If you can't see the picture,plz see the shell's below.

ailab@mpimaster:~/Downloads/mpich-3.0.4$ mpiexec -n 4 ./examples/cpi

Process 0 of 4 is on mpimaster

Process 1 of 4 is on mpimaster

Process 2 of 4 is on mpimaster

Process 3 of 4 is on mpimaster

pi is approximately 3.1415926544231239, Error is 0.0000000008333307

wall clock time = 0.028108

ailab@mpimaster:~/Downloads/mpich-3.0.4$ mpiexec -hosts mpimaster -n 4 ./examples/cpi

Process 2 of 4 is on mpimaster

Process 0 of 4 is on mpimaster

Process 1 of 4 is on mpimaster

Process 3 of 4 is on mpimaster

pi is approximately 3.1415926544231239, Error is 0.0000000008333307

wall clock time = 0.027234

ailab@mpimaster:~/Downloads/mpich-3.0.4$ mpiexec -hosts mpislaver1 -n 4 ./examples/cpi

Process 0 of 4 is on mpislaver1

pi is approximately 3.1415926544231239, Error is 0.0000000008333307

wall clock time = 0.000093

Process 1 of 4 is on mpislaver1

Process 2 of 4 is on mpislaver1

Process 3 of 4 is on mpislaver1

ailab@mpimaster:~/Downloads/mpich-3.0.4$ mpiexec -hosts mpimaster,mpislaver1 -n 4 ./examples/cpi

Process 0 of 4 is on mpimaster

Process 2 of 4 is on mpimaster

Fatal error in PMPI_Reduce: A process has failed, error stack:

PMPI_Reduce(1217)...............: MPI_Reduce(sbuf=0x7fff73a51ce8, rbuf=0x7fff73a51cf0, count=1, MPI_DOUBLE, MPI_SUM, root=0, MPI_COMM_WORLD) failed

MPIR_Reduce_impl(1029)..........:

MPIR_Reduce_intra(779)..........:

MPIR_Reduce_impl(1029)..........:

MPIR_Reduce_intra(835)..........:

MPIR_Reduce_binomial(144).......:

MPIDI_CH3U_Recvq_FDU_or_AEP(667): Communication error with rank 1

MPIR_Reduce_intra(799)..........:

MPIR_Reduce_impl(1029)..........:

MPIR_Reduce_intra(835)..........:

MPIR_Reduce_binomial(206).......: Failure during collective

================================================================================

= BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES

= EXIT CODE: 1

= CLEANING UP REMAINING PROCESSES

= YOU CAN IGNORE THE BELOW CLEANUP MESSAGES

================================================================================

[proxy:0:1@mpislaver1] HYD_pmcd_pmip_control_cmd_cb (./pm/pmiserv/pmip_cb.c:886)

[proxy:0:1@mpislaver1] HYDT_dmxu_poll_wait_for_event (./tools/demux/demux_poll.c

[proxy:0:1@mpislaver1] main (./pm/pmiserv/pmip.c:206): demux engine error waitin

[mpiexec@mpimaster] HYDT_bscu_wait_for_completion (./tools/bootstrap/utils/bscu_

[mpiexec@mpimaster] HYDT_bsci_wait_for_completion (./tools/bootstrap/src/bsci_wa

[mpiexec@mpimaster] HYD_pmci_wait_for_completion (./pm/pmiserv/pmiserv_pmci.c:21

[mpiexec@mpimaster] main (./ui/mpich/mpiexec.c:331): process manager error waiti

ailab@mpimaster:~/Downloads/mpich-3.0.4$

plz help,THX!

------------------

Jie-Jun Wu
Department of Computer Science,
Sun Yat-sen University,
Guangzhou,
P.R. China