No subject

copy the message too much (that is, never unless some headers are
modified).

It would also be nice if the message could end up materialized in one
contiguous sequence of bytes (like it does now).  However, it probably
doesn't matter if it ends up in a few separate, noncontiguous
regions-- for byte-based radio interfaces we can still avoid fully
materializing the message, while for DMA we can probably get away with
a small number of transfers.

I'm also interested in the interfaces used, although some of this is
protocol-dependent.  My feeling is that the prepare/send version is
going to be a lot nicer to use since then you know how much payload is
available.  Thoughts?

Steve

On Mon, May 18, 2009 at 6:03 PM, Omprakash Gnawali <gnawali at usc.edu> wrote:
> I hope the DT folks and others got a chance to look at the threads
> regarding mesasge buffer.
>
> From net2 perspective, I don't believe we care too much about
> representation of the packet unless it has serious performance
> implications or restricts some functionality that we use or need.
>
> In terms of representation, here are two things that might be relevant to us:
>
> - How do you initialize the message buffer? Do you need to
> reinitialize at every layer? Do you reinitialize when you use a
> different amsender?
>
> - Are pointers to different sections of the buffer preserved across calls?
>
> What else do we care about related to packet representation?
>
> - om_p
> _______________________________________________
> net2-wg mailing list
> net2-wg at millennium.berkeley.edu
> https://www.millennium.berkeley.edu/cgi-bin/mailman/listinfo/net2-wg
>