Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] Outstanding 1.3 RTE features
From: Ralph Castain (rhc_at_[hidden])
Date: 2008-05-07 13:15:32


On 5/7/08 11:08 AM, "George Bosilca" <bosilca_at_[hidden]> wrote:

> On May 7, 2008, at 10:24 AM, Ralph Castain wrote:
>
>> * static ports - basic infrastructure for procs and orteds to use
>> static OOB/TCP ports, but we don't currently take advantage of it.
>> This shouldn't require any API changes or major restructuring of
>> code as everything required is already there.
>
> This feature is already in Open MPI for the last 6 months. Port
> limitation was implemented for both the OOB TCP and the BTL TCP, with
> and without ipv6 support. Setting the min and max to the same value,
> force Open MPI (including ORTE) to use a static port.
>
> MCA oob: parameter "oob_tcp_port_min_v4" (current
> value: "0")
> Starting port allowed (IPv4)
> MCA oob: parameter "oob_tcp_port_range_v4" (current
> value: "65535")
> Range of allowed ports (IPv4)
> MCA oob: parameter "oob_tcp_port_min_v6" (current
> value: "0")
> Starting port allowed (IPv6)
> MCA oob: parameter "oob_tcp_port_range_v6" (current
> value: "65535")
> Range of allowed ports (IPv6)
>

Like I said in my note, the infrastructure is there. What is not currently
there is the logic that lets us take advantage of it to reduce socket count
and speed-up the launch. That will come, but not in time for 1.3.

>> * usage reporting - add appropriate mpirun cmd line option to
>> request the orteds to report proc resource usage upon proc
>> termination. Pretty trivial to do. Requested by a few users
>> and a couple of tool developers.
>
> I don't think this is the role of the runtime. Any user can write a 2
> lines sh script to do this, there is no reason to integrate it deeply
> inside ORTE.

Doesn't seem like a major issue to me - I've heard arguments both ways. It's
trivial to do, simplifies life for MTT and a few others, and probably not
worth a lengthy debate.

>
> george.
>
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel