Shirley Ma wrote:
Hello Tziporet,
Eli has reviewed the IPoIB-UD 4K mtu patchset. He suggested an
alternative way (reserve one buffer for all header including GRH,
IPoIB-ethernet, IP header, TCP header, and leave the user data in
other buffer) for the implementation but This approach can't avoid if
condition check without changing IPoIB-2K MTU implementation. And I am
not sure whether or how much performance gain from this approach, it's
pretty risky to change IPoIB-2K MTU at this moment.
I think the way Eli proposed is better but the point you raised (do not
touch the 2K MTU) is also important for stability.
I would like to keep my approach, and will limit to one "if need-S/G
check" in fast path. and I have passed stress test for both mthca and
ehca in the past few days. I don't see any issues for OFED-1.3 RC3. Do
you agree to include this IPoIB-UD 4K MTU patch into OFED-1.3 RC4? I
need to regerate this patch today since Eli's small messages size
patch has been in Git tree.
OK - go ahead and regenerate patch and we will be able to include it in RC4
BTW - how did you test it with mthca? It does not support 4K MTU. You
can test it with ConnectX since it does supports 4K MTU (with a special
burning configuration). Please let me know if you have ConnectX and you
wish to test it with 4K MTU
Tziporet
_______________________________________________
ewg mailing list
ewg@lists.openfabrics.org
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/ewg