The MPLS WG Archive

Cell Relay Retreat>MPLS WG Archive>month:2002-Dec> msg00160



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

[MPLS-OPS]: Jitter and MPLS

  • From: Curtis Villamizar <curtis@fictitious.org>
  • Date: Mon, 09 Dec 2002 11:35:55 -0500
  • cc: "Naidu, Venkata" <Venkata.Naidu@Marconi.com>, Eric Osborne <eosborne@cisco.com>, "J.C.Ramesh Babu" <rameshbabu_j@infosys.com>, mpls-ops@mplsrc.com, mpls@UU.NET


In message <3DF48F65.840949E1@cad.zju.edu.cn>, Jing Shen writes:
> 
> OMP gives a good start on traffic engineering but its traffic division
> policy is too simple for self-similar internet traffic. 

I don't see any basis for this assertion.

The IP src/dst hash is used almost universally in ECMP for OSPF, ISIS,
LDP (after payload snooping consistently indicates that payload is
IP), and MPLS/TE implementations (when L3PID indicates IP traffic), as
well as in proprietary link bundling (for example: Avici Composite
Links (tm) and others).  It works very well with Internet traffic.

> I have another question, is OMP implemented in some commerical products?
> if not, why? 

There have been a few research implementations that have come to my
attention, mostly grad student projects.

Implementing in a commercial router is a much larger undertaking due
to added complexity (forwarding cards are separate from router
processors, more is done in hardware or microcode, 100s or 1000s of
interfaces pose greater scaling challenges, etc) and the software
quality requirements.  Without a customer expressing strong interest
this hasn't happenned.

Curtis