Reviewing PSARC/2004/174, it seems that some work has been
done on implementing zero-copy read for Solaris...

For the receive side, the design document mentions that
full classification is required in order to post buffers
correctly to ensure that only the right data ends up in
the buffers posted by the application. Being when it was,
i.e. pre-crossbow, it appears that only write side zero-
copy has been implemented.

But now we've got Crossbow to take care of that classification
problem for us! :)

So...

What barriers remain to implementing per-socket receive
side zero-copy?

Also, if I wasn't going to plumb an interface and give
it an IP address (i.e just use with snoop), what barriers
would there be to all the NIC's buffers being supplied
by the application?

Darren

_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to