[mvapich-discuss] performance impact of ch3:mrail and ch3:nemesis:ib, tcp

[Hari Subramoni]

Hello Jianyu,

Build #1 is the OpenFabrics (OFA) IB/iWARP/RoCE build for the CH3 channel
in MVAPICH2. This is the default interface on Linux and has all the
advanced performance and scalability features in MVAPICH2. Build #2 is the
Nemesis sub-channel in MVAPICH2. It only has a subset of features that the
CH3 channel has. We recommend that you use the CH3 channel for best
performance and scalability at present.

I do not understand your second question. Could you please re-state it?

Regards,
Hari.


On Thu, Apr 3, 2014 at 2:49 AM, Jianyu Liu <jerry_leo at msn.com> wrote:

> Hi,
>
> There are two sorts of configurations for building for OFA-IB.
>
> Just wondering
>
>    1.  Any performance impact for these two sorts of builds?
>
>              bld 1 ).   ./configure --with-device=ch3:mrail
> --with-rdma=gen2
>
>              bld 2 ).   ./configure --with-device=ch3:nemesis:ib,tcp
>
>   2. How to find the proper rdma type?
>
> Appreciating your kindly help
>
> Jianyu
>
>
>
> _______________________________________________
> mvapich-discuss mailing list
> mvapich-discuss at cse.ohio-state.edu
> http://mailman.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20140403/a46f2d26/attachment.html>