Multilink PPP has been around a long time. It allows you to "bundle" two circuits together at layer 2 to get twice the bandwidth, increase the maximum receive unit and give yourself some circuit redundancy, so that if one fails, you still limp along at half the bandwidth. Until recently though, the router at either end still represented a point of failure. If you're not aware of Multichassis Multilink PPP, I suggest starting with this overview from Cisco.
Behind the scenes, MMP uses the Stack Group Bidding Protocol (SGBP) to figure out which of the boxes is going to own the "bundle", then it uses L2F to forward packets between the boxes. The configuration is done with virtual-template interfaces.
This technology is obviously very useful for organizations with lots of remote offices or users that need to dial in because it provides bandwidth on demand. That is, you may have an office with a variable population. Sometimes you may need 128k of bandwidth, and other times you may need 384k or more. You can configure the router at the site with multiple BRIs, which are relatively cheap when you don't use them, since you typically pay a small fixed charge plus a per-minute rate per channel in use, compared to a dedicated circuit which you pay a fixed rate for whether you're using it or not. If you configure the router so that it only brings up each additional circuit when your users need the bandwidth, and then drops the calls after an inactivity period, then your monthly bills could be quite low. Of course, this previously presented a problem in that you didn't want a single chassis fielding all the calls during your peak periods, and without multiple chassis, a given site would have to connect all its BRI channels to the same head-end chassis, which may not be possible if it's full. Multichassis Multilink PPP fixes this problem.
It also provides a more robust alternative for a WAN backup solution. For instance, if your primary circuits are T1s or even fractional T3s, you could set up multiple BRIs to connect only when the main pipe goes down.
Tom Lancaster, CCIE# 8829 CNX# 1105, is a consultant with 15 years experience in the networking industry, and co-author of several books on networking, most recently, CCSPTM: Secure PIX and Secure VPN Study Guide published by Sybex.