Hello LAM,
since I have work around SUCCESSEFULLY by using
the `LAM_MPI_SESSION_PREFIX' mechanism:
before to launch `lamboot',
I set `LAM_MPI_SESSION_PREFIX' to `/mnt/ramfs',
i.e., under bash
export LAM_MPI_SESSION_PREFIX=/mnt/ramfs
where /mnt/ramfs is a mounted ramfs partition
with an appropriate write permission.
Then I can initiate `my' diskless LAM
without trouble:
I could lauch the test suite:
nevertheless,
the original node scratched at the `dynamic step',
but I guess this is a wider issue.
NOTE that rather a ramfs partition,
a `tmpfs' partition may be used:
for what one need
they play the same role,
but tmpfs partitions are more flexible
(more options and better documentation).
[NB: I use `ramfs' because the kernel of `my' local cluster
was not build with `tmpfs']
To finish,
if I do not use this mechanism,
I run the `hello' program
but after a while and sometime I get
an error message.
Further, the below issue reamins.
I hope that helps,
Jerome
Jerome BENOIT wrote:
> Not really,
> since I cannot run the testsuite:
> more precisely I can run one arbitrary test,
> but not a second one unless I `lamhalt and lamboot' between the tow tests.
>
>
>
> Jeff Squyres wrote:
>
>> On Sun, 4 May 2003, Jerome BENOIT wrote:
>>
>>
>>> I have just checked this issue with the NEW LAM (beta) realease 7.0:
>>> everything is fine now !
>>
>>
>>
>> Excellent!
>>
>> Given this, and given that our entire focus these days is on the new 7.x
>
>
> I f i can run the brand new 7.x, this is perfect !
>
>> series, it is probably unlikely that we will fix the problem in the 6.5.x
>> series.
>>
>
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
>
|
