Re: [myrinet] GM over MPI

On Fri, Jun 08, 2001 at 10:12:57AM +0930, Davis, Harvey wrote:

> Was wondering what the advantage was of using GM over MPICH over GM?

The bandwidth and latency curves of GM vs. MPICH-GM are roughly the
same. And you might find it fairly complicated to program raw GM, as
evidenced by this question:

> How much data will gm buffer if you only give it one receive buffer? ie is
> there a point at which it will drop data?

GM will drop 100% of data that doesn't have an appropriately-sized
buffer to fit in. MPICH-GM uses control messages to make sure that
large buffers are created before large messages are sent.

-- g