Inverse multiplexing: Tailor-made for ATM

The story of asynchronous transfer mode wide area network services sounds a little like Goldilocks' lament: One service is too wide, one is not quite wide enough and one is just right.

The first of these, OC-3 and DS-3/E-3 ATM, provided more than enough bandwidth but at too high a cost for many applications. Then T-1 ATM provided less expensive ATM access but without enough bandwidth for many ATM applications. But some say that the bulk of pent-up ATM demand lies somewhere between T-1 and T-3. Deployment of wide area ATM has been disappointingly slow, and some believe it's the result of this bandwidth gap. New alternatives arriving soon will address this midrange bandwidth gap and may jump-start ATM services.

Many believe that we're now entering the "just right" phase of ATM services, and the "bear" making it happen is inverse multiplexing, a mature technology that currently delivers frame-based WAN services but is not yet being used much in ATM services. That will change soon, but the question is, what flavor of NxT-1 will suit your taste? We're going to sample the options, but first let's take a closer look at inverse multiplexing.

A Look Under the Hood As Figure 1 shows, inverse multiplexing is just what it sounds like-time division multiplexing in reverse. A single "pipe" or data stream is split apart into multiple T-1 or E-1 data streams and then recombined to form the original stream. The current T-1 inverse multiplexers work by splitting apart the incoming data stream bit by bit in a round-robin scheme (Figure 2). Every so often, a T-1 payload bit is "stolen" to provide a channel for alignment and overhead information.

Inverse multiplexing sounds simple, and the concept actually is. The difficult part is dealing with the different delays introduced by the DS-1 circuits. Even DS-1 circuits taking the same path through the network may differ slightly in delay. This differential delay must be canceled out by temporarily storing data from the faster circuits until the corresponding data from the slowest link arrive.

A variable delay buffer must be set at just the right size for each DS-1. The inverse multiplexer uses alignment information in the overhead channel on each DS-1 to precisely measure and compensate for the differential delay.

Clearly, inverse multiplexing is harder than it looks. If ATM is added to the mix, the reason why the ATM Forum has spent well over a year wrestling with the details of inverse multiplexing for ATM (IMA) becomes obvious. Originally called AIMUX, this work started with the standard T-1 and E-1 ATM physical layers so that the existing interface integrated circuits could be used. Figure 3 shows how the same round-robin distribution is performed but by one 53-byte ATM cell at a time. Instead of stealing a single DS-1 payload bit, a special IMA control protocol cell is regularly inserted on each DS-1 ATM circuit.

Frames or Cells? Most agree that ATM is clearly the best backbone networking technology for integrating all classes of services, including voice and video. However, some controversy still remains about the access side.

Much of this controversy stems from ATM's inherent inefficiencies. To achieve scalability and switching performance, ATM uses a fixed-length cell containing a header with virtual circuit addressing information. When frames are converted to cells, usually via ATM adaptation layer 5, efficiency is reduced by 10% to 20% as a result of cell headers and partially filled cells. For data-only applications, this overhead cost may be too high.

A compromise is to use high-level data link control (HDLC) frames to provide access to ATM services (Figure 4). The first frame-based access protocol, ATM data exchange interface (DXI), is based on the frame relay frame format. Frame relay's data link connection identifier field is used to extend ATM virtual circuits across the frame interface.

The frame-based user-to-network interface (FUNI) protocol extends DXI to include signaling and management over the frame interface, allow much greater control and flexibility. FUNI is specified in terms of the customer network interface to the WAN, ready-made for a tariffed service.

An existing frame relay network can also provide a front end to ATM services. Using frame relay to ATM service-interworking, frame relay permanent virtual circuits can be mapped to ATM virtual circuits at the demarcation point between the two networks. The interworking gateway also remaps frame protocol headers so that the user data packet appears in its native mapping on each network. This allows full interoperability between ATM and frame relay end points.

Frame-based solutions are attractive because the terminating equipment needs no new hardware interface to attach to the network. Any HDLC-capable device can talk DXI, FUNI or frame relay using a DSU or inverse multiplexer. However, frame-based ATM access can't provide the real-time guarantees. Most interactive or multimedia applications need the quality of service guaranteed by native ATM.

Enter the Inverse Multiplexer Today's inverse multiplexers are a mature, robust technology that's been delivering frame-based traffic for years. Typically equipped with standard V.35 or HSSI data interfaces, they are compatible with most customer premises equipment high-speed data interfaces. They are currently used to provide frame relay service and readily adapt to FUNI or DXI extension services.

Figure 5 shows a typical network based on inverse multiplexers. Stand-alone inverse multiplexers provide a low-cost CPE solution. These perform the equivalent functions of a DSU, providing a standard HSSI or V.35 interface to customer equipment. Remote management, extensive diagnostics including bit error rate testing and loopbacks, and full T1.403 termination can simplify customer support.

In the central office, an access multiplexer performs the necessary conversion and concentration. Multiple NxT-1 connections are brought in via standard T-1 or T-3 connections. Connections to service switches can be made directly via V.35 or HSSI. An ATM-capable access shelf may also provide the ATM segmentation and reassembly function and interworking or protocol support.

The ATM community responded to the demand for lower-speed ATM with a move to specify a new UNI: IMA. The physical interface committee of the ATM Forum usually defines standard mappings of ATM cells onto existing physical layer media. In this case, the UNI rides atop the existing T-1 or E-1 ATM physical interface, performing the inverse multiplexing via a cell-based control protocol-a major departure from the normal physical interface definition.

Opinions vary as to when the IMA specification will be completed by the ATM Forum. Revisions are still being made at a rapid pace and must be completed before the balloting process can begin. Once defined, equipment implementing IMA will also go through the normal maturation process for a new technology.

Early adopters can expect the usual problems and glitches as new hardware, software and protocols are debugged. Early implementations will need upgrades once the specification is final. When the dust settles, though, multivendor-interoperable NxT-1 ATM will be a reality, ready for widespread deployment. Indications are that this specification will be widely accepted in both the user and equipment vendor communities.

Until IMA is ready for prime time, service providers are faced with a dilemma. Should a frame-based ATM service be offered at NxT-1, which might quickly go out of style? Should a "bleeding-edge" service offering with early IMA implementations be brought on-line? If nothing is done, competitive access providers might just beat them to the punch. A little-known ATM Forum specification just might let them have their cake and eat it, too.

The ATM Forum's cell-based transmission convergence (TC) sublayer specification, also known as ATM-over-HSSI, specifies a standard format for transmitting cells over any clear-channel bit stream interface (not just HSSI). Actually, no electrical connector is specified (V.35 would work), so a better name would be "clear-channel ATM."

The cell-based TC specification defines ATM transport at the bit level rather than the electrical or optical level. Issues such as connectors, clocking, modem control and status are left to other standards. This approach is similar to defining a data link layer framing format such as ISO 3309 (HDLC). The bit order is defined, as well as how the start of cell is determined. Cells are simply placed bit by bit onto the transporting technology.

ATM Over Anything The clear-channel ATM bit stream may be carried over any WAN data circuit, including those provided by today's inverse multiplexers. There are currently few implementations of this interface, but more are coming. In fact, HSSI ATM interfaces have been announced by several equipment providers. Little is known yet about interoperability, but it shouldn't be difficult given the simplicity of the interface.

If frame services are already being provided using inverse multiplexers, clear channel ATM might be provided with the same equipment. Circuit provisioning is identical and requires no retraining, but ATM protocol or interfacing issues will still need to be dealt with. However, it may be some time before users can plug directly into a tariffed, clear-channel ATM interface.

And the ATM Forum is still working on new physical interfaces: three for video services and one for wireless ATM, not to mention ATM support for asymmetrical digital subscriber line services. Can't the ATM vendors make this process less painful?

ATM service providers may feel deja vu coming on when they hear about IMA. After deploying support for DS-3, OC-3 and perhaps T-1 UNI, they're now being asked to roll out yet another WAN interface. Given the slow availability of new interfaces for CPE, service providers are justified in proceeding with some caution.

ATM switches are usually the first to see a new UNI, then perhaps workstation NIC cards, if applicable. Access equipment vendors often take longer to equip routers and local area network switches with new technologies. In the meantime, extra equipment must be deployed to bridge the gap between old and new UNI technologies.

During the interim years of a new UNI's deployment, an ATM switch may be used to convert between old and new. Customers who already have a switch might be able to add the interface when available, but what about the others? Many NxT-1 ATM applications-router interconnection, for instance-do not require CPE ATM switches. This will add greatly to service cost in these applications.

Where Service Begins For service providers, this situation has added ramifications. If a customer must provide an ATM switch, some value-added services that are ATM-based-LAN emulation, for instance-are likely to be brought in-house by the customer instead.

Service providers can elect to provide the switch, but the cost of service delivery is greatly increased. Obviously, higher service costs will result in slower sales and make the service less attractive when compared with T-3 or OC-3 access, but the advantages are compelling. Providing a widely used interface such as OC-3 or DS-3 greatly broadens the universe of potential customers. It also simplifies upgrading from NxT-1 to another transport. The customer interface can stay the same. It should be possible to eliminate much of that cost by simply providing conversion only-with no switching-in a new class of equipment.

Sorting through the options for delivering mid-range bandwidth can be a challenge. Finding the answer to the question of which fits best in a company's ATM plans will take some time. But one part of the answer is already clear-inverse multiplexing will be at the heart of the third ATM wave.

Jim Frimmel is Systems Architect at Larscom Inc., Santa Clara, Calif.