[c-nsp] jumbo frames

[Alex Rubenstein]

On Fri, 3 Mar 2006, Tim Winders wrote:

> The discussion on MPLS got me thinking about jumbo frames.  Is there any
> reason NOT to set jumbo frames to 9216 if both ends of the link support
> it?

As a matter of pract, I set MTU to the maximum allowed (of course keeping 
the ends in common). In most cases, this is 9192, 9216, or rarely 4470.


> On switches such as the 3560G-GE where you have to set a global MTU, is
> there any reason not to set the global MTU to 9198?  In most of my cases,
> end workstations will not be configured for jumbo frames, but the trunk
> back to the core would support jumbo frames and there might be an
> occasional server that would support jumbo frames.

(IMHO): It is less important about the commonality of MTU on end points in 
the same L2 domain, so long as the L3 network does not exceed it.


> I have stayed away from jumbo frames as I had been told that all devices
> in the network must support jumbo frames.  Leaving the MTU at 1500 seemed
> the "safest" bet.  I'm not wondering what I am losing by doing this and
> what troubles I might find myself in if I suddenly start to enable jumbo
> frames on my network.

Simply turning it on is really not going to do anything. 99.9999999% (I 
ran out of patience typing 9's) of the hosts and endpoints have a MTU 
setting of 1500, so in most cases, mos traffic will not exceed 1500. 
However, setting larger MTU's on routers, etc., will have the possibility 
of (for instance) BGP TCP sessions have a MSS of larger than 1500 -- 
sometimes, this can help speed up convergence (depending on the platform) 
-- at least that is what I have noticed.

It will also aid in the 'down the road' applications (that you may not do 
today), such as EoMPLS, L2TPv3, etc. Also, if you have large file 
transfers happening between hosts (ie, backups, etc), setting a larger MTU 
on these hosts may speed up transfers considerably. Beware of how this may 
break other things.

I unfortunately know about this, as I've spent the last 2 to 3 months 
jumbo-enabling the entire NAC network. If you want to ping me offline for 
anything I might have missed, feel free.


-- 
Alex Rubenstein, AR97, K2AHR, alex at nac.net, latency, Al Reuben
Net Access Corporation, 800-NET-ME-36, http://www.nac.net