Inverse Multiplexing

IT shops are scrambling to accommodate the convergence of data and the WAN. LANs pump more and more bandwidth to the WAN's edge, but few network engineers have optimized their systems to a WAN environment, where bandwidth is limited and you pay for every transmission.

The traditional WAN solution was leased from one of the phone companies. Leased services, although they provide dedicated bandwidth, are expensive. Depending on the distance between the two points, T1 lines can run you a five-figure bill per month.

In response to the constricted economy and the advent of LANs, WAN-service providers developed options more attractive to data users both in price and technology. The Regional Bell Operating Companies, interexchange carriers , and value-added network providers are heavily promoting these services, including frame relay, ISDN, and SMDS.

But you don't have to venture into a new and unproven technology to get better service; you can recycle what you already have. The public-switched network offers time-proven service, and the prices are low. A 56Kbit/sec data call now costs about the same as a voice call.

Bandwidth on demand dictates that users can access dial-up lines to call up additional WAN bandwidth. So when a LAN application bursts and more data needs to be sent than there is pipe to send it through, another switched-service line can automatically be called into action.

From bandwidth on demand comes inverse multiplexing.

More Elegant Than Its Name

What inverse multiplexing accomplishes is more elegant than its moniker suggests. Ordinary multiplexing takes multiple low-speed lines and puts them onto one high-speed line. Inverse multiplexing spreads a high-speed output over multiple low-speed (and presumably lower-cost) lines, while maintaining the appearance of a high-speed transmission. Hence the name "inverse."

Inverse multiplexing works as follows . A T1 line is made up of 24 channels, each running at 64Kbits/sec, for a total throughput of 1.544Mbits/sec. Each channel only has 56Kbits/sec of throughput, because 8Kbits/sec is necessary for in- band signaling. The key is that a T1 line consists of 24 separate channels.

For example, you can inverse multiplex two 56Kbit/sec lines into what appears to the application to be one 112Kbit/sec pipe. (The number and speed of lines is indicated using Nx, where the "N" denotes the number of lines and the speed follows the "x." So two 56Kbit/sec lines are referred to as 2x56.) Inverse multiplexing typically uses multiple 56Kbit/sec leased or switched lines, although 64Kbit/sec lines can be used internationally. You can also use ISDN Basic Rate Interface, ISDN Primary Rate Interface, ISDN HO Switched 384, or full T1 (see Figure 1).

Figure 1: An inverse multiplexer makes multiple lower-speed (and lower-cost) lines look like a single higher-speed (but still lower-cost) line to the applications at either end.

The benefit is that you are buying low-cost, low-speed lines and ratcheting up the speed yourself. With inverse multiplexing, you can get the bandwidth equivalent of a T1 but at a much lower price. You can buy fractional T1 today, but the applications view the channels as separate.

Inverse multiplexing will be most beneficial if you are trying to internetwork several LANs, but it is also applicable for videoconferences, traffic overflow, and disaster recovery.

Insert The Inverse Mux

Figure 2 shows how an inverse multiplexer works. Multiple high speed data streams are input into the inverse mux. These data streams can come from the different ports on the same device, perhaps a router, or they may come from different sources, perhaps a router, a terminal server, and a video codec. The inverse mux uses time-division multiplexing to split the data stream into multiple 56Kbit/sec channels.

Figure 2: An inverse multiplexer aggregates different data streams over higher-speed lines. The seperate channels take diverse paths through the network and arrive at their destination but not necessarily at the same time or in the right order. The inverse mux puts the packets back in the proper order and adjusts for any delay.

The separate 56Kbit/sec channels take diverse paths through the public switched networkunbeknown to the application. For example, Figure 2 shows that the "abcdefghijklmnop" message is broken into four streams: "aeim," "bfjn," "cgko," and "dhlp." The "aeim" stream goes from the switch in San Francisco to Chicago and New York before it reaches Washington, DC. The "cgko" packet goes to Los Angeles and Atlanta on its way to the District. The four data streams arrive at the same destination but not necessarily at the same time or in the right order. Then, like a router, which segments and reassembles packets, the inverse mux buffers the arriving packets and puts them in the proper order. It then passes the intact information to the router, then the user 's application.

A Standard Method

Inverse multiplexing improves on a feature some routers offerload balancing or load sharing. With load sharing, the router tries to balance the traffic across multiple outgoing ports; however, load sharing adds about 30 percent overhead to the transmission. Also, each vendor's implementation is unique, so one vendor's implementation will not work with another's. You're locked into a proprietary solution.

Inverse multiplexing accomplishes the same function but uses a standard called Bonding. This specification was written by the Bandwidth on Demand Interoperability Group, a consortium of 40 manufacturers, and has been passed to the American National Standards Institute TR41.4 group .

Bonding supports four modes. Mode 0 is a special mode for dual 56Kbit/sec calls, which offers special economies when both sides are using 56Kbit/sec lines. Mode 1 works with any number of lines (called Nx) and has no subchannel. The subchannel can be used for control signaling and other types of information. Mode 2 is also Nx but has an in-band subchannel overhead of 1.6 percent. Finally, Mode 3 is also Nx with an out-of-band subchannel.

Ascend, a leading manufacturer of inverse muxes, has its own protocol Ascend Inverse Multiplexing (AIM)which it recommends for higher performance. AIM supports four modes: Static, which has no subchannel; Manual, which has an in-band subchannel overhead of 0.2 percent; Delta, which is an out-of-band subchannel; and Dynamic, for dial-up bandwidth.

The ISDN H.221 specification can also be used for inverse multiplexing, but it is primarily used for video, not LAN, data.

Immediate Gratification

Inverse multiplexing offers a way to "roll your own" dial-up bandwidth service immediately. You don't have to wait for the phone companies to tariff new service offerings.

Later this year, the WAN-service providers will step up their offerings of commercial services that will deliver the equivalent of inverse multiplexing. For example, AT&T currently offers Switched 384, which provides the same bandwidth as six 64Kbit/sec lines and an inverse mux. Yet 384Kbits/sec is a rather unwieldy chunk of bandwidth and so less attractive to LAN users.

The carriers may make inverse multiplexing less attractive in another way. When you dial up additional lines using an inverse multiplexer, you are charged for each call setup. So if you dial up six 64Kbit/sec lines, you are charged for six separate calls. If you buy Switched 384 or similar service, you are charged for only one call.

This tutorial, number 55, by Patricia Schnaidt, was originally published in the April 1993 issue of LAN Magazine/Network Magazine.