On Mar 28, 2014, at 1:17 PM, Joshua Ladd wrote:
> So after all that: I think you shouldn't need to specify the connection
> manager MCA parameter at all; the openib BTL should choose the Right one for
> you.
> [Josh] Nyet. See above.
Ah. Shows how much I know. :-)
Is there any way to make
I also believe that for iWARP and RoCE, the RDMA CM will be chosen
automatically, and UD CM will be automatically chosen for IB.
[Josh] If you want to run OMPI over RoCE on Mellanox hardware, you must
explicitly choose rdmacm with -mca btl openib,sm,self -mca
btl_openib_cpc_include rdmacm - th
On Mar 28, 2014, at 12:18 PM, "Shamis, Pavel" wrote:
>> Technically you may setup RoCE connection without RDMA CM.
>
> The version of the RoCE support that I implemented (in an alternative MPI
> implementation) did it through the regular OOB
> channel. As I remember the only difference is the
> On Mar 27, 2014, at 11:45 PM, "Wang,Yanfei(SYS)"
> wrote:
>
>> 1. In the RoCE, we cannot use OOB(via tcp socket) for RDMA connection.
>
> More specifically, RoCE QPs can only be made using the RDMA connection
> manager.
Technically you may setup RoCE connection without RDMA CM.
The vers
On Mar 27, 2014, at 11:15 AM, "Wang,Yanfei(SYS)" wrote:
> Normally we use rdma-cm to build rdma connection ,then create Qpairs to do
> rdma data transmit ion, so what is the consideration for separating rdma-cm
> connection built and data transmit ion at design stage?
There's some history her
Normally we use rdma-cm to build rdma connection ,then create Qpairs to do rdma
data transmit ion, so what is the consideration for separating rdma-cm
connection built and data transmit ion at design stage?
Maybe this question is not reasonable, and hope any response?
Thanks
Yanfei
Sent from m
Hi,
Thanks Ralph and Jeff.
Do we have full documentation on these parameters, further on open MPI
transport design architecture?
Pls recommend some website or paper.
Thanks
Yanfei
Sent from my iPad
On 2014年3月27日, at 下午10:10, "Ralph Castain"
mailto:r...@open-mpi.org>> wrote:
Just one other po
Just one other point to clarify - there is an apparent misunderstanding
regarding the following MCA param:
-mca btl_openib_cpc_include rdmacm
This param has nothing to do with telling openib to use RDMA for
communication. What it does is tell the openib BTL to use RDMA to establish
the point-to-p
it seem that all confusions have already been shot, thanks Jeff!
Thanks!
Yanfei
Sent from my iPad
> On 2014年3月27日, at 下午8:38, "Jeff Squyres (jsquyres)"
> wrote:
>
> Here's a few key facts that might help:
>
> 1. The hostfile has nothing to do with what network interfaces are used for
> MPI
Here's a few key facts that might help:
1. The hostfile has nothing to do with what network interfaces are used for MPI
traffic. It is only used to specify what servers you launch on, regardless of
what IP interface on that server you specify.
2. What network interfaces are used are a combinati
Hi,
Update:
If explicitly assign --mca btl tcp,sm,self and the traffic will go 10G TCP/IP
link instead of 40G RDMA link, and the tcp/ip latency is 22us at average, which
is reasonable.
[root@bb-nsi-ib04 pt2pt]# mpirun --hostfile hosts -np 2 --map-by node --mca btl
tcp,sm,self osu_latency
# OSU
11 matches
Mail list logo
