3 Mb question

Jay Hennigan jay at west.net
Thu Oct 14 00:59:24 UTC 2004

On Wed, 13 Oct 2004, Richard J. Sears wrote:

> If you search the list for ip load-sharing per-packet you will see how
> we manage all of our multi-customer T1s.
> Never had any long term luck with MLPPP.

We have used both, and have found that MLPPP gives better results for
real-time applications like voice at the cost of increased CPU.  For
generic data links, ip load-sharing per packet works fine.  If the source
and destination traffic is reasonably diverse, simple equal cost routes
without per-packet will work as well, but you won't get greater than
1.5mbps for a given flow.

> > I've got what seems to me like an innocuous question for this list...
> >
> > Someone is requesting access to about 3 mb of traffic up/dn. I figure 2
> > T1s will give them the 3 Mb I need, but I'm looking for suggestions on
> > either efficiently combining those 2 to get the most bandwidth for their
> > buck or else I have to look at getting them a ds3 and scaling back to
> > what they need.
> >
> > Is there an good low end suggestion for making effective use of 2 T1s to
> > give 3 Mb of bandwidth? In practice, I've seen 2 T1s load balanced with
> > CEF not do very well at giving a full 3 Mb. (This was without turning on
> > per-packet CEF)
> >
> > I'm not personally experienced with MLPPP or mux hardware if that helps,
> > but I could get it set up if that's the consensus as the best option.
> > The NRC of something that would effectively couple the 2 T1s would
> > easily beat the MRC of a DS3 which I think might be overkill for just 3
> > Mb.

Jay Hennigan - CCIE #7880 - Network Administration - jay at west.net
WestNet:  Connecting you to the planet.  805 884-6323      WB6RDV
NetLojix Communications, Inc.  -  http://www.netlojix.com/

More information about the NANOG mailing list