[j-nsp] Re: MTU size of fxp0

[Richard A Steenbergen]

On Tue, Apr 26, 2005 at 11:18:08AM +0200, Daniel Roesen wrote:
> On Tue, Apr 26, 2005 at 10:58:38AM +0200, Jonas Frey wrote:
> > As far as i recall, it can be done using some (sysctl -w
> > net.pfe.transit_re=1 ?) sysctl setting. However this is not a good idea
> > since the routing will be done in software/go through the RE directly.
> 
> And clog the fxp1 link between RE and PFE.
> 
> All in all, it's a stupid idea. :-)
> 
> And personally, I've never seen transit_re working. There seems to be
> more to the story than just the sysctl.

Nah, I routinely use this, like when working remotely with newly installed 
routers which only have an fxp0 uplink to the outside world. Normally 
routes/nexthops from the fxp0 interface are not installed in the PFE, but 
when you set transit_re=1 those routes get installed.

Not enabling it by default is a perfectly reasonable decision by Juniper, 
it keeps stupid people from trying to route 100Mbps between the real 
hardware and the mgmt port. But being able to turn it on in a pinch if you 
know what you are doing is handy too.

> BTW: the FreeBSD ethernet driver for fxp0 doesn't support MTU >1500
> as far as I know, so you cannot raise MTU even with tricks.

As I learned today actually, tnp on the fxp1 link handles fragmentation 
and reassembly of packets larger than the 1500 octet MTU on the 
interface, which lets you take in larger packets off the PFE. TCP MSS is 
set off of the configured interface mtu, etc.

However, there is no way to do higher than 1514 on the fxp interfaces 
themselves. This means no jumbo frames on fxp0/mgmt, etc. You can 
configure vlans on fxp0, but it will be passing 1496 octet packets. :)

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)