Re: [OMPI devel] ORTE scalability issues

[Ralph H Castain]

Thanks Christian.

Actually, I was aware of that and should have clarified that these tests did
*not* involve the IPv6 code.

Ralph


On 4/17/07 1:31 AM, "Christian Kauhaus"  wrote:

> Ralph H Castain :
>> even though the HNP isn't actually part of the MPI job itself, or the
>> processes are opening duplicate OOB sockets back to the HNP. I am not
>> certain which (or either) of these is the root cause, however - it needs
>> further investigation to identify the source of the extra sockets.
> 
> If you are using the IPv6-ready code: in this case we need to create two
> sockets for each OOB/TCP. One uses AF_INET and one uses AF_INET6.
> IIRC, we close the superfluous socket once the connection attempt on
> either one succeeds. Adrian, correct me if I'm wrong. :-)
> Unfortunately, there's no easy way around this.
> 
> Christian

Re: [OMPI devel] ORTE scalability issues

[Christian Kauhaus]

Ralph H Castain :
>even though the HNP isn't actually part of the MPI job itself, or the
>processes are opening duplicate OOB sockets back to the HNP. I am not
>certain which (or either) of these is the root cause, however - it needs
>further investigation to identify the source of the extra sockets.

If you are using the IPv6-ready code: in this case we need to create two
sockets for each OOB/TCP. One uses AF_INET and one uses AF_INET6.
IIRC, we close the superfluous socket once the connection attempt on
either one succeeds. Adrian, correct me if I'm wrong. :-)
Unfortunately, there's no easy way around this.

Christian

-- 
Dipl.-Inf. Christian Kauhaus   <><
Lehrstuhl fuer Rechnerarchitektur und -kommunikation 
Institut fuer Informatik * Ernst-Abbe-Platz 1-2 * D-07743 Jena
Tel: +49 3641 9 46376  *  Fax: +49 3641 9 46372   *  Raum 3217