The MPLS-OPS Archive

["Amos Rosenboim" <r.amos@zahav.net.il>]

Hi,
 
Of course that the best solution is to increase mtu, but alot of the times this cannot be done due to various limitations. in cases like this were you must have a working solution i suggest the following (both solutions are "ugly" but both work and served me well until we solved all the mtu bottlenecks in our network):
 
1. Ip tcp mss adjust mss <mss value> - mss is maximum segment size. it's a tcp paramater negotitated during tcp connection establishment and it's used to negotiate the maximum length (size) of a tcp segment that will be transmitted between the parties. it's usually derived from the mtu (altough no necessary, cisco for example use a default mss of 576 for it's tcp connections and this is why enabling ip tcp path mtu discovery can do charms to your bgp convergence time). The above command let's you control that value. if you set this value for something like 1370-1400 bytes then you should be ok. not every ios version support this feature but T train or B train images should be ok. 
Note - this feature is performed in software - from my experience a 7200 npe G1 with alot of traffic experienced between 10 - 15% cpu time increase but i enabled this feature for all traffic. 
If you decide to apply this feature i recommend that you do it in the CE router or in the PE router under the interface facing the CE. this command is applied in the interface level (not subinterfaces though).
 
 
2. if the above feature doesn't help for some reason  u can alwayes create a route map that sets the don't frag bit in any ip packet to 0. this will enable routers to frag large packets. of course fragmantation is not recommended neither for the routers nor for the appliaction but sometimes you have no choice.
Such route map will look like this:
 
route-map frog permit 10
 set ip df 0
 
As for the previous feature it's important to apply this route map at the closest possible place to the customer network.
 
I hope this helps.
 
Amos
 
 

	-----Original Message----- 
	From: Garry Glendown [mailto:garry@regio.net] 
	Sent: ש 18/09/2004 20:22 
	To: mpls-ops@mplsrc.com 
	Cc: Pelle 
	Subject: Re: [MPLS-OPS]: IP Routing problem after enabling MPLS?
	
	

	OK, I'm officially out of ideas ... I was at our customer all day today
	... setting up the MPLS VPN wasn't a problem, worked like a charm ...
	the problem is the Ethernet bridge our customer is operating on ...
	While I was able to configure the switch involved to run with an MTU of
	1512, as well as our core routers which run on GigE ... but I have not
	been able to configure the FE's to run with anything larger than the
	1500 MTU, keep getting the "doesn't support user settable MTU" ...
	Now, for the MPLS VPN we're implementing for our customers, this
	wouldn't be that much of a problem, as most of the communication will be
	to DSL dialups, which are running with way below the 1500 MTU ... but
	the MPLS network is also on the backbone (of course), and the ethernet
	bridge (50mbit limited 100MBit link) to the customer ... and that's
	where it is all messed up at ... while many web pages e.g. will continue
	to work, some (like e.g. heise.de or ebay.com/de) are hardly usable if
	at all ... I've tried messing around with the "ip mtu" which of course
	didn't work either ... also neither google nor cisco.com were much of
	help ...
	
	Is it just the IOS versions I'm running? How do other people do their
	MPLS? I refuse to believe that everybody running MPLS does so only on
	either GigE, E3, POS or others, but not FE or Eth ... somewhere I read
	that Cisco FEs do not support larger MTUs, but "show cont" will display
	a max MTU of 1524 (while GigE has a Max MTU of 1536, but allows for
	higher sizes in "conf t" ...)
	
	Help appreciated,
	
	-gg
	
	
	-------
	The MPLS-OPS Mailing List
	Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
	Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
	


-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml