[j-nsp] LDP Max PDU Length

[Hannes Gredler]

saku,

i do not think that tweaking ISO MTU is related to the
LDP problem that eric was observing.

@eric: could you pls send me a tcpdump trace file
showing the LDP conversation.

i.e.

monitor traffic interface <if-name> size 4474 write-file ldp-pdu-len-issue.pcap

and send me that file pls ?

/hannes


Saku Ytti wrote:
> On (2006-03-08 18:08 -0500), Eric Van Tol wrote:
> 
> 
>>I'm performing some MPLS testing between a J2300 and a Telco Systems
>>TM-100 switch.  I'm attempting to get an LDP session established between
>>the two devices and it's not working, failing on the Juniper with a "Bad
>>PDU Length" message.  The Max PDU Length from the Juniper shows it to be
>>a value of 4096.  The TM100 shows, in the LDP Initialization Message,
>>that the Max PDU is 1440.  
> 
> 
>  If your physical MTU's really are matching, you could try to decrease
> JNPR's ISO/CLNS MTU (set int <int> unit <unit> family iso mtu <mtu>) to
> match that of telcos MTU. It may be that telco isn't smart enough 
> to increase ISO/CLNS MTU from 1497.
> 
> 
>>It's my understanding from RFC3036 that the Max PDU is negotiated
>>between the two neighbors and the smaller of the two is used.  However,
>>it goes on to say that the receiving node can reject the PDU length if
>>it is too small or too large.
>>
>>My questions:
>>
>>1.  How exactly is the max PDU length determined - is it based of the
>>MTU of the physical medium?
>>2.  If it's an arbitrary value, why wouldn't the value then be
>>negotiated to 1440?
>>3.  What exactly determines that a received PDU length is
>>"unacceptable"?
> 
> 
>  debug, see what PDU length is reported in the TLV and see how long the
> received PDU was.
>  One thing I've noticed about telcos (not sure if it applies to TM-100) is
> that they put each network address to it's own fragment, consequently PDU's
> will be rather small, and would never go above 1500, so supporting higher
> MTU wouldn't make sense for them.
>  In typical CSCP/JNPR network you will not have any LSP's fragmented.
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp