Hi Joe,
Check the source for your NIC driver (/usr/src/linux/drivers/net). You
may also want to check the VLAN mailing list
laphish:/usr/src/linux/drivers/net # grep -r jumbo *.c | cut -d: -f1 |
sort -u
acenic.c
dl2k.c
ns83820.c
tg3.c
laphish:/usr/src/linux/drivers/net #
They seem to be 1000Mbit/s NICs and I think the sk98lin driver support
jumbo frames too.
as to Michael Brown's posting...
SHould I expect a standard 100Mbps switch (eg Netgear) to drop
large MTU packets?
Depends how intelligent the switch tries to be. "An efficient
implementation of Gigabit Ethernet with MTU greater than 1500 bytes
would probably require switches to resegment Ethernet frames greater
than 1500 bytes and recompute the checksums. This would add to the cost
of switches." (from: Communication Performance over a Gigabit Ethernet
Network by Paul A. Farrell and Hong Ong)
AFAIK 100Mbit/s NICs don't support mtu>1500 per standard definition, but
I would need to check my copy of the IEEE-802.3 draft. There was a
meeting about things like that [1].
And Mr. Dijkstra is talking merely about Alteon. As everyone knows,
Alteon just tries to elude every possible standard below OSI 3 :)
I was going to try it locally at first, and then between two boxes
connected by a cross-over cable, to at least see if it was practical.
I doubt you get the NICs to send 1500 mtu without major patching but I
stand corrected. And I don't really think you gain that much in
performance. You might reduce CPU load but since there would not be a
early congestion notification or dynamic socket buffer sizes, you might
end up even decreasing performance.
After that I was going to deal with the switch problem.
Take a hub and get 2 Broadcom Tigon3 NICs, compile and 'ifconfig eth0
mtu 9000' :)
For some tests see [2].
[1] http://grouper.ieee.org/groups/802/17/documents/minutes/\
minutes110600.pdf
[2] http://www.scd.ucar.edu/nets/docs/reports/HighSpeed/
Cheers and see you at OLS,
Roberto Nibali, ratz
--
echo '[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq'|dc
|