Francesco DiMambro wrote:
> Hi Andrew
>    So you can get to line rate on those machines? Thanks for
> letting me know, so far I've been unable  but I've also spent
> more time on x86 as well. I can now justify spending more
> time on it.

Well, line rate - epsilon, where epsilon tends to be
a bit bigger on sparc than on x86 (9.1x vs 9.4x for
1500b).

I was really much more interested in why you were
seeing 1.2 for MDT vs the 2.3 I was seeing for
LSO.  I think we can factor out hardware differences
since I assume both our NICs are capable of line
rate for this MTU, so our NICs are mostly idle.
So that leaves the big difference being my GLDv3/LSO
vs your streams/MDT.  And it seems like GLDv3/LSO wins
by a factor of 2.   If you have a streams/LSO driver,
then maybe we can try to figure out exactly how
much of that difference is streams vs GLDv3 and how
much is MDT vs LSO.

Drew
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to