[mvapich-discuss] Installation Problem (MVAPICH2 0.9.8p2)

[Lundrigan, Adam]

Don,

Thanks for sharing your instructions.  I downloaded a fresh copy of MVAPICH2 v0.9.8, followed your instructions, and the installation went smoothly.  I must have made an error somewhere when modifying make.mvapich2.udapl previously, as it was not properly detecting that I was running on Solaris.  I have tested my installation on the four nodes that use ibd0 as their interface, and the OSU benchmarks return numbers that agree with the tables in your instructions (~5.5us minimum latency, max. bandwidth ~700MB/s).  

The head node (an x4200) has it's infiniband adapter currently configured as ibd2.  I need to reconfigure that node to also use ibd0 before I can do a full system benchmark, but things are looking good thus far.

Thanks again for your help, 
--
Adam Lundrigan
Computer Systems Programmer
Biological & Physical Oceanography Section
Science, Oceans & Environment Branch
Department of Fisheries and Oceans Canada
Northwest Atlantic Fisheries Centre
St. John's, Newfoundland & Labrador
CANADA   A1C 5X1

Tel: (709) 772-8136
Fax: (709) 772-8138
Cell: (709) 277-4575
Office:  G10-117J
Email: LundriganA at dfo-mpo.gc.ca



From: Don Kerr
Sent: Tue 7/17/2007 1:26 PM
To: Lundrigan, Adam
Subject: Re: [mvapich-discuss] Installation Problem (MVAPICH2 0.9.8p2)


Adam,

Caveat: I do not run mvapich regularly. I have not run on x4100/x4200 
but I have on other AMD boxes. And I have not tried v0.9.8p2, only  v0.9.8.

I have setup and run a couple of times in order to do some comparisons 
of mvapich2 to Sun's free version of MPI which is based on Open MPI 
(http://www.sun.com/download/products.xml?id=4612c490). Attached are my 
notes used to setup mvapich2 with udapl on a 2 node cluster. My hope is 
that these help.

Also, have you tried our implementation of MPI, link above?  We supply 
binaries so you don't have to build.  Anyway if you do have a chance to 
try it since you are already running Solaris 10 11/06 you would only 
need patch 125793-01 for the AMD box.

Goodluck.
-DON



Lundrigan, Adam wrote:

> We are trying to set up a Sun-based cluster using Solaris 10 11/06 
> running on Sun Fire x4100 & x4200 systems. When trying to compile 
> MVAPICH2 v0.9.8p2, there is a flood of errors and the installation 
> dies.  Here is a short sample of the error messages
>
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/channels/mrail/src/udapl/vbuf.h",
>     line 230: undefined symbol: u_int32_t
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/channels/mrail/src/udapl/vbuf.h",
>     line 232: syntax error before or at: u_int32_t
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/channels/mrail/src/udapl/mpidi_ch3_rdma_pre.h",
>     line 115: syntax error before or at: u_int8_t
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/channels/mrail/src/udapl/mpidi_ch3_rdma_pre.h",
>     line 214: syntax error before or at: u_int8_t
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/channels/mrail/src/udapl/mpidi_ch3_rdma_pre.h",
>     line 255: syntax error before or at: u_int16_t
>     "/software/mvapich2-0.9.8p2/src/mpid/osu_ch3/include/mpidpre.h",
>     line 163: syntax error before or at: MPID_Seqnum_t
>
> To the best of my knowledge, u_int*_t is a Linux kernel datatype, and 
> as such is not defined in Solaris (which uses uint*_t).  Is there a 
> flag or something that i'm missing, to turn off the use of u_int*_t 
> datatypes?  I have uncommented everything in make.mvapich2.udapl which 
> is flagged as Solaris-only. 
>  
> Has anyone else here successfully set up MVAPICH2 an environment 
> similar to ours?
>  
> Thanks in advance for your help,
> --
> *Adam Lundrigan*
> Computer Systems Programmer
> Biological & Physical Oceanography Section
> Science, Oceans & Environment Branch
> Department of Fisheries and Oceans Canada
> Northwest Atlantic Fisheries Centre
> St. John's, Newfoundland & Labrador
> CANADA   A1C 5X1
>
> Tel: (709) 772-8136
> Fax: (709) 772-8138
> Cell: (709) 277-4575
> Office:  G10-117J
> Email: LundriganA at dfo-mpo.gc.ca
>
>------------------------------------------------------------------------
>
>_______________________________________________
>mvapich-discuss mailing list
>mvapich-discuss at cse.ohio-state.edu
>http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>  
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20070717/4df0c0ed/attachment-0001.html