Network Computingwww.networkcomputing.com

Along with physical layer options, you have choices to make at the data-link layer. HDLC and PPP are both appropriate for use in private networks. If you are using a T3 as a high-speed connection to the Internet, the ISP may choose to use frame relay encapsulation. This does not imply that your bandwidth will be under the constraints of regular frame relay connections that can drop packets during busy periods on the supplier's network. Frame relay encapsulated T3 lines are usually supplied as point-to-point connections; the ISP selects frame relay encapsulation just because their engineers are familiar with that option. Look at frame relay as another form of encapsulation, not a networked service. In practice, there is no significant difference in the overhead introduced by any of these protocols. The headers are roughly equivalent, and the higher- layer protocol headers have far more impact on bandwidth consumed than link layer protocols.

ATM-DXI is an option if you are taking a T3 to connect to a public ATM service. These connections use an ATM DSU to generate cells for transport over the ATM service, and communicate with your equipment using DXI (Data Exchange Interface specification) encapsulated frames. The T3 presents a single 43-Mbps channel to the router, which will supply variable length frames to the ATM DSU, which in turn will segment those frames into cells for transmission on the ATM network. The downside is that the router sending DXI format frames can't take advantage of any of the UNI features of the ATM network--such as specifying service levels. That is done by the ATM DSU. For point-to-point PVCs on an ATM network, this isn't much of a drawback, but it can be with SVC access to multiple remote sites (see "Physical Connections: ATM-DXI Specifications," to the right).

A newer alternative is frame-based UNI (FUNI). This is a curious option in that the router interface will generate variable length FUNI format cells, which the ATM network will convert into regular cells. This allows the router interface full access to the ATM UNI features (see "Physical Connections: FUNI Specifications," at left).

The benefit of variable-length frames or cells on the T3 link (which is slower than native ATM links that are typically 155 Mbps and up) is reduced overhead. With ATM cells, the header is 9.4 percent of the data transmitted, which is OK when you have lots of bandwidth, but on the slower links, like T3, bandwidth is more of a premium and the cell overhead may be too much. With variable-length frames, header overhead is generally far less on a percentage basis. So, consider DXI or FUNI when you need to access ATM networks via a T3 rather than by way of direct optical fibers.

In the simplest case, with a serial line such as HSSI, just assign an IP address to the interface and be sure it is not shut down in the configuration. That is the minimum required, and in that instance, a T3 will look like any other link on the network. The physical layer and data-link layer compatibility are taken care of automatically, with no special configurations necessary.

Equally simple is an ATM switch with a T3 PAM, which is largely plug-and-play when ATM ILMI (Integrated Local Management Interface) is enabled. Make sure the PAM is connected to the T3 and not administratively shut down. ATM switches only work with ATM addresses, so it isn't meaningful to assign an IP address to any interface on it that is used for ATM networking.

Configuring channelized services is a little more complex. With a channelized trunk card in a router, each subinterface on that trunk card will be configured for 24 individual DS-1 channels, a T1 by default. Most router vendors number these channels starting from 1, so that channel numbers agree with the telcos' numbering. Each channel equates to T1 and there is no grouping channels together as is possible on T1 channelized service.

To allocate bandwidth between T1 and T3 from your central location to a remote site requires an interface that can understand T3 signaling and decode individual channels. Given that you can identify individual channels, the best data-link option is to use either multilink PPP or rely on a routing protocol to balance traffic across multiple T1 channels. The telco will deliver multiple T1 lines to the remote site. These need to terminate into the same router, so that your central site equipment can see these multiple paths as equal cost routes to the same destination and split traffic between them. A word of caution: RIP version 1 will not support this type of operation; it will just choose one link and use it exclusively. IGRP, OSPF and other newer routing protocols all support this equal cost route load balancing.

For ATM, there are a couple of extra steps necessary. The ATM PVC number must be set and a destination protocol address must be mapped to that PVC. For ATM DXI implemented on Cisco equipment, a configuration would look like this:

ip address 10.1.1.1
encapsulation atm-dxi
dxi pvc 100 100 mux
dxi map ip 10.1.1.2 100 100 broadcast

This defines a PVC with VPI=100, VCI=100, using the mux encapsulation that means the PVC is only for use with one protocol and maps that VPI/ VCI pair to the destination IP address 10.1.1.2.

Send your comments on this article to Chris Lewis at clewis@cmp.com.