For example, most ATM switches move voice traffic over ATM Adaptation Layer 1 (AAL1)-based Circuit Emulation Services, statically reserving mostly unused bandwidth and adding the extra overhead (approximately 12 percent) that AAL1 introduces. Thus, you'll need 1.74 Mbps just to support the traffic from a T1 (1.5 Mbps) interface, and you won't gain efficiency relative to leased lines and time-division multiplexing.

A Work in Progress Fact is, proprietary ATM implementations outperform the standards. Voice su pport is the most prominent example of how vendor-specific WAN optimizations outdistance the standards. Voice equipment vendors, such as PBX manufacturers, have paid little attention to ATM, so it's up to the network to provide efficiency without changing the interfaces this equipment expects. At the time of our testing, the ATM Forum was wrapping up the specification for Voice and Telephony over ATM (VTOA), which specifies interoperable methods of efficient voice trunking, among other things. None of the switches tested has interoperable voice optimizations. Based on the current progress in the ATM Forum, it will be mid-1998 before interoperability will occur, and even then single vendor solutions will likely be more efficient.

Each vendor offers a grab bag of voice tricks--the most obvious first steps being compression, silence removal, idle detection and modem/fax detecti on (fax and modem calls should remain uncompressed). All implementations inject ambient noise during silences to avoid the "drop-out " effect that complete silence on the line can create. GDC and Nortel both offer 32-Kbps ADPCM compression (ITU-T Standard G.726), and Nortel also offers 16-Kbps ADPCM. IBM offers a proprietary compression scheme that gets traffic down to about 13 Kbps (not counting ATM overhead). If you're lucky enough to have common compression algorithms between dissimilar vendor switches, you're stuck with inefficient AAL1 virtual circuits between end points. So, once again, only single-vendor solutions come close to delivering ATM's promised efficiency.

Nortel offers excellent voice support through its Variable Bit Rate-Real Time (VBR-rt) over AAL 5 implementation. In an all-Magellan network, a typical voice conversation consumes ridiculously small amounts of bandwidth (three concurrent, but silent, calls consumed 1 Kbps of bandwidth in our tests). GDC has introduced an "early standard" AAL2-VTOA implementation in its Voice Service Module, which was not available in time for testing. GDC's focus on standards in voice networking gives its products especially good long-term value.