[c-nsp] BGP Balanced

Andre Beck cisco-nsp at ibh.net
Mon Sep 6 10:01:40 EDT 2004


Hi,

On Wed, Aug 25, 2004 at 03:19:53PM -0400, Rodney Dunn wrote:
> 
> The best you'll get is with MLPPP because it solves the 
> out of order problem and the "only use one link" problem.

Nice to read about others who prefer MP in cases where you can stand
the extra CPU overhead. As for a real world example, consider a 2620
terminating one ISP upstream via three E1s. That's exactly the
constellation where I found MP to be working, but slightly too much
pressure on the CPU (it beamed up beyond the 60% load marker every
now and then), while load was staying below 50% with CEF per-packet.
So if you're mainly unicast, CEF per-packet is often an option where
MP tends to hurt the box.

Then again, as it wasn't yet mentioned in the thread, this is only
relevant for unicast. So far the only option I have found to load
balance *multicast* is MP.

BTW, as you mentioned other L2 based load distribution mechanisms
like EtherChannel in the same context as MP: EtherChannel will present
itself to L3 as one interface, but on L2 it will again introduce a
purely statistical load balancing, typically Dst MAC or Src MAC
based, seldom more complicated. Maybe that's different on the
flagship platforms like the 6509 with current SUPs, but it's often
surprising people on most "normal" platforms that a channel (link
aggregate) doesn't balance a single flow at all.

-- 
                  The _S_anta _C_laus _O_peration
  or "how to turn a complete illusion into a neverending money source"

-> Andre Beck    +++ ABP-RIPE +++    IBH Prof. Dr. Horn GmbH, Dresden <-


More information about the cisco-nsp mailing list