Ruth Sivilotti | 4 Jul 09:35 2002

[Myrinet] gm-1.5.2.1 for Linux available

gm-1.5.2.1 for Linux is available.

This release fixes bugs related to the fork support
added in gm-1.5.2. The CHANGES file entries for 
gm-1.5.2 and gm-1.5.2.1 are included below for your convenience.

--Ruth

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

gm-1.5.2.1

BUG FIXES
1) Fix several bugs in the fork support for Linux, potentially causing kernel 
   panics.
2) Re-enabled the system() overload using vfork instead of fork for performance 
   purposes.
3) Initialization of the PCI Parity Error bit in the PCI config space (PERR)
   to 0 at load time of the driver. This bit may be set to 1 during POST, and
   stay high during operations, without real parity error detected on the PCI.

ENHANCEMENTS
1) New feature in the mapper to set the same MAC address to multiple NICs in 
   the same host, useful for channel bonding and high availability.
2) Small documentation change for ATI conflict on Linux.
3) gm_stress update about transient error management.

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

gm-1.5.2
(Continue reading)

Saurabh Agarwal | 10 Jul 16:59 2002
Picon

[Myrinet] NAS benchmarks on GM.

Hi, 

We have a 8 node linux 2.4 cluster (All P-II nodes, 128MB ram) connected 
through a 8 port myrinet switch. We are using Lanai-7 cards with 4 MB memory.
We are tyrin to run NAS benchmarks 2.3 using mpich-1.2.1..7b over GM 1.5.2 . 

The NAS benchmarks execute correctly for 8 nodes, if they are compiled for
the class W but if we increase the workload of the benchmark to class A and
higher, the benchmarks cease to work.

This happens when we run multiple (4, in our case) programs together. 
in short: We had 4 MG (class A)  programs running simultaneously, asking for 
all 8 nodes.

Note that if we run upto 2 MG (A) applications, it works fine. Problem starts
when we scale it up over 3 or more..

We are not sure what could be causing this. Is it possible that we get 
limited by the amount of memory on the LANAI card or could it be host 
memory limitation ? (128 MB !)    We are also not sure how can we find 
out what is the usage statistics of the lanai memory at any instant of
time...  

We will greatly appreciate your help on this.

Thanks a lot,
Regards,
--Saurabh

***************************************************************************
(Continue reading)

Patrick Geoffray | 15 Jul 07:05 2002

[Myrinet] Re: [Myricom help #12328] NAS benchmarks on GM.

Hi Saurabh,

> Note that if we run upto 2 MG (A) applications, it works fine. Problem starts
> when we scale it up over 3 or more..

What type of problems (benchmark crashes, segfaults, incorrect answer, 
error messages) ?

> We are not sure what could be causing this. Is it possible that we get 
> limited by the amount of memory on the LANAI card or could it be host 
> memory limitation ? (128 MB !)    We are also not sure how can we find 
> out what is the usage statistics of the lanai memory at any instant of
> time... 

The memory footprint on the Lanai is bounded. Anyway, it's almost 
constant and it does not depend on the host memory size or the size of 
the job (only number of GM ports, max number of nodes and there is a 
translation cache that will quickly grow to its maximum).

I don't know by heart the size of size of MG(A), but 128 MB may start to
be small (AFAIR, MG's memory footprint is 256^3.
Tell me how it dies, and I will try to reproduce it here if I have no clue.

Patrick

----------------------------------------------------------
|   Patrick Geoffray, Ph.D.      patrick <at> myri.com
|   Myricom, Inc.                http://www.myri.com
|   Cell:  865-389-8852          685 Emory Valley Rd (B)
|   Phone: 865-425-0978          Oak Ridge, TN 37830
(Continue reading)

Patrick Geoffray | 29 Jul 18:56 2002

[Myrinet] MPICH-GM 1.2.4..8 is now available

Dear Myrinet users,

It took much longer than expected to complete all tests on various
architectures/compilers, but finally here it is.

Please let us know as soon as possible if you experience problems.
There will be an update in a couple of weeks, including support
for Windows, Bproc (Scyld), for DDT debugger, port of the extensive
documentation from MPICH and eventually any corrections.

There are many things improved in this release, specially
configuration related. Hope you enjoy it.

Patrick

-------

We are happy to announce the release of MPICH-GM 1.2.4..8.

    http://www.myri.com/scs/

The new features of this MPICH-GM implementation include:

* MPICH 1.2.4 base, including ROMIO and MPE.

* New mpirun.ch_gm script:
    - Support for ProcGroup file, for spawning multiple executables
      (no support for heterogeneous configuration yet), using the flag
      "-pg".
    - Possibility to close stdin on the process MPI Id 0, to put jobs in
(Continue reading)

Yifeng Zhu | 30 Jul 07:48 2002

[Myrinet] How to monitor the network traffic?

Hi,

I need to write a program to monitor the throughput of 
the myrinet traffic on each node every three seconds.

Could anyone give me a hint? I am working on Red Hat
Linux. Does it just need to read some file in the
/proc/net direcoty to get the current througput?

Thanks.

Regards,
Yifeng

Oleg I. Vdovikin | 31 Jul 16:58 2002
Picon

Re: [Myrinet] MPICH-GM 1.2.4..8 is now available

Hi Patrick,

    I'm trying to compile mpich for our Alphas with Compaq compilers. And
the first warning messages appeared in the gmpi_priv.c file - there is a
number of fprintfs like this --

fprintf(stderr,
              "FATAL ERROR on MPI node %d (gm_id %d): Drop callback status "
              "is unexpected (%d)\n", MPID_MyWorldRank, status);

    i.e. having "gm_id %d" format specifiers, but no argument.

    Attached please find stderr log.

Regards,
    Oleg.

----- Original Message -----
From: "Patrick Geoffray" <patrick <at> myri.com>
To: "Myrinet Mailing List" <myrinet <at> osc.edu>
Sent: Monday, July 29, 2002 8:56 PM
Subject: [Myrinet] MPICH-GM 1.2.4..8 is now available

> Dear Myrinet users,
>
> It took much longer than expected to complete all tests on various
> architectures/compilers, but finally here it is.
>
> Please let us know as soon as possible if you experience problems.
> There will be an update in a couple of weeks, including support
(Continue reading)


Gmane