[c-nsp] Packet order and MP

[Jessup, Toby]

Does MP (MLPPP) protect packet order?

I just read RFC 1990 (again) and it seems to indicate that the sequence
numbers in a MP packet header are only used to reassemble fragments in
the receiver. The RFC does not appear (to me) to ensure FIFO
higher-layer packet order from the sender MP process ingress to the
receiver MP process egress. I was thinking it did ....

It appears it would be possible for OOO packets to occur in a case where
MP uses a fragment length of one MP frame (a fragment has both the
beginning and ending fragment bits set to 1 in the frame) and the
differential delay between fast MP bundle elements is high (such as for
a bundle of T1/E1 spans).

So, is it true that if MP is used and "fragmentation is disabled" (a
fragment size of 1, really), then OOO packets are a possibility for
higher layer applications? In regards to ensuring payload packet
ordering across a high speed bundle, MP is no better than per-packet
load sharing at the IP layer (CEF ECMP) when the MP fragment size is set
to 1. Agree?

Toby Jessup 
Qwest Communications
National Technical Services
1600 Seventh Avenue, Room 1911
Seattle, WA 98191
(206) 224-5565