On Fri, Apr 03, 2009 at 10:45:06AM -0700, Ted H. Kim wrote: > Was there ever an idea that we could rely on > the IB checksum to protect the > payload instead? I vaguely remember some > old presentation talking about checksum > bypass or something similar.
Some patches were made but I don't think they were ever completed. The difficulty was arranging things so that an absent checksum could be negotiated and then ensuring that as the rx'd packet passed through the Linux stack it was properly flagged as valid data, no checksum so that the checksum is recomputed if the packet happens to be forwared out another interface. Essentially within the Linux architecture it would become similar to how something like Xen works to provide checksum/segmentation offload to VMs (and probably using the code they wrote..) IIRC the interest was actually beyond just checksum but the whole bag of tricks including segmentation offload. Essentially just connect two Linux stacks together over RC with all the offload features turned on. Jason _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
