Drew,
Can you remind us how many RX groups and RX rings per group, as well as
TX rings are you exposing out of the box? Creating a VNIC currently
disables polling on the primary client (this is something we are now
working on to address as part of dynamic grouping), so this issue could
be related to polling, and I would suggest looking in that area first.
Nicolas.
Andrew Gallatin wrote:
Hi,
I just updated an opensolaris machine (dual quad-core nehalem) from
OpenSolaris 2008.11 to the latest "dev" build (basically an update
from snv_101b_rc2 to snv_118). This is a very fast machine, and
(under different OSes / revisions) can easily keep 2 10GbE links
busy.
Before upgrading to 118, I saw performance from OpenSolaris like:
% netperf245 -H asus2-m -C -c -l 60 -- -S 512K -s 512K
524744 524288 524288 60.00 9868.72 9.29 8.91 0.617 0.592
After the upgrade:
524744 524288 524288 60.00 7061.58 14.63 5.67 1.357 0.526
So the transmit costs are more than doubled, and now even a
very fast machine is having trouble getting line rate, even
using jumbo frames.
I've tried enabling/disabling my driver's LRO, as well as
enabling/disabling the use of multiple hardware rings.
The only thing that seems to help is a vaugely remembered
hint to create an (unused) vnic. Once I did that, I got quite
a bit of performance back:
524744 524288 524288 59.99 9444.87 8.38 8.53 0.582 0.592
(though still not line rate).
Can somebody remind me what creating a vnic does? Is there
some setting which would do the same thing?
Thanks,
Drew
_______________________________________________
networking-discuss mailing list
[email protected]
_______________________________________________
networking-discuss mailing list
[email protected]
