What frame relay switch is causing MCI/Worldcom such grief?

Vijay Gill wrath at cs.umbc.edu
Mon Aug 9 23:46:29 UTC 1999

On Mon, 9 Aug 1999 smd at clock.org wrote:

> Vijay -
> | Traffic flows change faster than the topology in most cases.
> So, in other words TE is a hack to make up for people's weaknesses
> in forecasting that leads one into a connectivity graph where the
> normally-selected logical path between two points is insufficient 
> for the amount of traffic one wants to shuffle between ingress
> and egress?

> "Our forecast horizon is shorter than our planning cycle, so therefore
> we need TE" is probably the most real argument I've run into so far
> for doing TE, but it's so terribly sad that this is the case.  
> Again, again, again, just like so many other times in the past.

The growth is bursty and that taken into conjunction with the fact that
sometimes telco's cannot deliver capacity on demand even when the circuits
have been ordered years in advance, definitely leads to a situation where
either you 1) have way too much capacity because the telco's suddenly
delivered and now you're paying for excess or 2) they don't deliver at
all, your traffic growth jumped and now you're forcing 4x through a 1x

It also helps the transient pipe cuts, where you have X capacity between
two points, and lose half your capacity to backhoe fade.  Allowing for the
95th percentile of traffic to pass, rerouting through the not-so-short
path will prevents a total meltdown. 

> The other "compelling reason" I've run into for TE/MPLS is that it allows
> some people to close the book on the ATM-switch-in-the-middle-of-a-
> cluster-of-7[05]xx-es hack that was done in response to another
> famous underprediction of growth about which the people who seem most
> in favour of MPLS seem the most unforgiving.
> Ironic, isn't it?

Life is full of these.

> | Traffic measurements from end to end (per city flows) are also
> | easier to measure with the city to city pipe as it were.
> Uhm, think about that carefully before you declare it a compelling
> reason - step 1: sort traffic into categories that group together
> combinations of the 5-tuple, step 2: count this traffic.

It is not a very compelling reason, but it is a reason.

> In other words, if extracting information about aggregate traffic
> flows is the means by which someone justifies deploying a pseudo-
> circuit-switched network like MPLS, I believe the problem lies in
> poor communications skills when communicating one's needs to one's
> vendor(s).

Maybe so.


More information about the NANOG mailing list