howto increase MTU size on 100Mbps FE
bogdan.costescu at iwr.uni-heidelberg.de
Fri Apr 26 14:08:32 PDT 2002
On Fri, 26 Apr 2002, Joseph Mack wrote:
> I know that jumbo frames increase throughput rate on GigE and was
> wondering if a similar thing is possible with regular FE.
I was just about to ask the same thing... but to be more precise I was
looking for first-hand experience with MPI/PVM over oversized Ethernet
frames. As you say, the current situation doesn't really encourages
experiments with Fast Ethernet, but I was thinking especially to people
using Gigabit Ethernet.
> I couldn't increase the MTU above 1500 with ifconfig or ip link.
There are also limitations in the drivers. The upper network layers will
refuse to set a value that is not supported by the driver.
> VLAN sends a packet larger than the standard MTU, having an
> extra 4 bytes of out of band data. The VLAN people have
> problems with larger MTUs.
I think that most of the problems come from the fact that most cards do
not have support for VLAN - allow the extra 4 bytes _only_ if the VLAN tag
is present. Most of the cards allow oversized frames but there is no
control over size and/or VLAN tag presence.
> which indicate that the MTU is set in the NIC driver and
> that in some cases the MTU=1500 is coded into the hardware
> or is at least hard to change.
There are actually some hardware limitations: cards have FIFO buffers
which are designed based on normal Ethernet frame size; while VLAN's 4
extra bytes usually fit, a 4-8 KiByte packet usually doesn't. Some drivers
also take active measures to prevent Tx underruns which are probably
disturbed by oversized frames.
> I don't know whether regular commodity switches (eg Netgear
> FS series) care about packet size, but I was going to
> try to send packets over a cross-over cable initially.
That's another question. Store-and-forward switches probably need to store
the whole packet before transmitting it further...
A week ago, I released a patch to add large MTU/VLAN support to the 3c59x
driver in 2.4.18. So far I haven't receive any feedback about it... It
still needs some work, but I first wanted to test functional support.
IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu at IWR.Uni-Heidelberg.De
More information about the Beowulf