Re: gm_board_info only shows the node itself

Hi Patrick & everyone,

Thank you for many suggestions.
I updated the kernel to 2.4.20-20.7 and re-compiled GM. 
Currently it is working without problem !

So, two cases are probable.
1) kernel 2.4.18-3 was the cause. maybe too old ? (never know...)
2) I was the cause. I might have been using SMP or UP .config file instead 
of bigmem config.
(I thought I was using the right .config file - but... I cannot tell...)

Thank you very much for all the help !!! 

Atsuko

----------------------------------------------------
ATSUKO MIYASHITA
Linux Support Center, SWSC, Technical Support - IBM Japan
Tel : 03-3808-8993 ( ext. 1712-8993 )  /  Fax : 03-3664-4893
e-mail : ATSUKOM＠jp.ibm.com
** Moved to Hakozaki - mind that tel # has changed ! ** 




Patrick Geoffray <[email protected]>
2004/01/14 08:31

 
        To:     Atsuko Miyashita/Japan/IBM@IBMJP
        cc:     [email protected], "Dr. Markus Fischer" <[email protected]>, 
[email protected]
        Subject:        Re: [Myrinet] gm_board_info only shows the node itself

 

Hi Atsuko,

Atsuko Miyashita wrote:
> Yes, I aware this... and actually, if I do
> mpirun -np 1 ./cpi
> 
> then it works without problem.
> It means, when I am not using Myrinet, it works...

You are always opening a GM port, even with only one process.
The difference is that MPICH-GM does not try to allocate
DMAable memory if all of the processes are on the same machine
(and the shared memory support is not disabled).

As the GM driver fails to lock pages on your machine, it fails to
allocate DMA-able memory (which needs to be locked), that's why
you see these errors:

> >[root@hgt001 basic]# mpirun -np 4 ./cpi
> >Process 2 of 4 on hgt001.linux.coc
> >[3]: alloc failed, not enough memory (Fatal Error)
> >Context: <(gmpi_init) gmpi_dma_alloc: dma send buffers>
> >[1]: alloc failed, not enough memory (Fatal Error)
> >Context: <(gmpi_init) gmpi_dma_alloc: dma send buffers>

gmpi_dma_alloc fails because the gm_dma_alloc call fails, but it fails
because the driver cannot lock pages, not because lack of memory.

> I think I should try latest version, GM 2.0.8 tomorrow...

You can do that, but if the kernel mismatch is still there, I don't
think it
will make a difference.

Patrick
-- 

Patrick Geoffray
Myricom, Inc.
http://www.myri.com

_______________________________________________
Myrinet mailing list
[email protected]
http://email.osc.edu/mailman/listinfo/myrinet