Re: [nsp] RES: [nsp] BGP maximum datagram size

From: Scott Whyte (swhyte@cisco.com)
Date: Tue Oct 30 2001 - 13:57:25 EST


On Tue, 30 Oct 2001, Loureiro, Rodrigo - (Bra) wrote:

> Hi Gert,
>
> Thanks in advance for response.
>
> >So your application should be able to handle everything from 1 byte to
> >4096 bytes - more than that is against the RFC, but your neighbors are
> >free to send up to that amount.
>
> This is linked to a fragmentation problem that i´m investigating with Cisco
> in PE routers (VPN/MPLS architecture). Actually, if i could guarantee that
> Cisco BGP´s implementation will not send a packet greater than 536 bytes, i
> could isolate the problem scope, once it will not affect my routing
> environment. However, if it follows RFC1771 in such a way that it is
> possible for Cisco´s peer to send a packet till 4096 bytes, than i should
> consider BGP inside the problem scope.

Currently cisco BGP will send updates with a max of either 512 or 256
bytes depending on IOS version, and will accept updates up to 4096 bytes.

This static max is subject to change, however, as Gert points out the RFC
allows any size up to 4096 bytes in an update.

-Scott

>
> I made a lab generating 2000 routes and than debugged the update packets
> between two MP-iBGP neighbors. I could see that BGP sends a lot of small
> packets, instead of aggregate a lot of NLRI´s in a single big packet.
> Actually, none of the packets observed reached 536 bytes.
>
> Based on the lab i could assume that BGP will not send datagrams greater
> than 536 bytes, but i could not get this information in any Cisco
> documentation, which means that i cannot guarantee this behaviour will be
> uniform.
>
> Do you have a pointer to any documentation that could solve my doubt ?
>
>
> Regards,
> --
>
> Rodrigo Loureiro
>



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:52 EDT