Hi folks,
I'm new to this list and I decided to subscribe due to a bug I found on
my 2.6.x kernel series and I don't know if it could be fixed upstream or
not.
I own a small 4 nodes cluster that contains 2 nodes with a dual e1000
onboard and 2 nodes with a dual tg3 onboard. All of them are connected
to the same switch.
For validating my network configuration I've a netpipe between all my
nodes : everything is fine I have the expected performances for some
gigabit network.
But (there's always a "but" ;)), when I start several connexions at a
time performances are not very good. Starting from one node 3 netpipe
shows me that 2 are handle at a time and one once this 2 streams are
ended. This is quite annoying...
I've reboot under my 2.4.28 kernel, and the 3 streams works at the same
time... so something had changed....
After a long while, my collegue discover that my e1000's didn't set the
flowcontrol... I've just set it up using "options e1000 FlowControl=3".
I restarted my network, and waow.. it works \o/
Performances are back, the 3 streams are ok.
I've been told by another friend that e1000 doesn't handle hardware
flowcontrol negociation so this is handle by some software. ok..
So my question is : "Why does on my 2.4 the flowcontrol was enabled and
not on my 2.6.x ?"
PS:I've validated this from the 2.6.11 to 2.6.14.2.
-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
