[openib-general] RMPP implementation
Fab Tillier
ftillier at infiniconsys.com
Tue Nov 16 17:33:43 PST 2004
> From: Sean Hefty [mailto:mshefty at ichips.intel.com]
> Sent: Tuesday, November 16, 2004 5:07 PM
>
> I'm starting work on the RMPP implementation in the MAD code. If anyone
> has any ideas/preferences on the implementation, please let me know.
>
> For the send side, there are a couple of ways to perform the segmentation:
>
> 1. Issue one send at a time. Additional sends are not transfered until
> the first send completes.
> 2. Issue multiple sends using 2 data segments per request. This
> requires allocating and mapping space (36 bytes) for copying the MAD
> common and RMPP headers.
> 3. Issue multiple sends using 3 data segments per request. This is the
> same as #2, but only copes the RMPP header.
>
> I'm leaning towards #2 at this point.
>
Isn't #1 the simplest to implement? Turnaround on the send queue should be
pretty quick, so send performance should be fine. I say do whatever is
simplest, and then optimize from there, and to me that means #1 at the
moment. What are the reasons to *not* do #1?
- Fab
More information about the general
mailing list