CNC Logo

Mixing Voice and Data

Peter J. Welcher


Introduction

This month you find me trying to decide: Is it time for a more basic router / switch article, perhaps focusing on configuration? Or should we look at something else? Help me avoid indecision in the future: if you have opinions on this or ideas for articles, feel free to send me email! (And thanks to those who have sent ideas, they're not forgotten, just queued).

This month I've been lucky enough, among other things, to get to look through a large number of Cisco presentations and other technical material. And thanks to the Mentor Technologies' folks who've also been working with this information. (Special thanks to John Stone, for his direct and indirect help on this article. Any errors that remain are mine!)

You may not have noticed (I certainly hadn't), but Cisco is starting to have quite a story concerning mixing voice and data traffic. This became more evident with heavy coverage in the last Packet magazine from Cisco. Consider that Cisco just announced a partnership with NEC America to integrate voice and data networks. There also partnerships with Symbol Technologies, Inc. concerning Voice Over IP, and with GRIC Communications, Inc. concerning Internet phone gateways and settlement services.

Let's a look at what's going on with Cisco and voice in this month's article. My intent in what follows is to condense the vast amount of information available into a high level overview.

We'll look at the options and technology first, then at specific products.

History

Part of the reason for buying Cisco StrataCom equipment has been what one might call "ATM as supermux", that is, ATM switches and trunks as an OSI Layer 2 network capable of transmitting data, voice, video, or whatever. If you're a service provider, the arguments for doing this are pretty clear.

But StrataCom has also focused on voice technology. They've got several pieces of technology that enhance their voice offerings: mature echo cancellation, various choices of voice compression, voice activity detection (VAD). The idea is to compress voice, suppress echo (in effect, reflected signal), and squeeze out periods of silence. The other end of the link needs to insert "pink noise", little active sounds that we all listen for to tell us the phone circuit is active. All of these affect our perception of the quality of the voice circuit.

When you compress voice, it has to be uncompressed so a typical PBX can deal with it. In so-called Tandem PBX voice networks, voice traffic is routed through one PBX to get to another. The ensuing double compression degrades the quality of the voice signal. Another related StrataCom offering is Voice Network Services (VNS), whereby you can use your StrataCom ATM network and PNNI routing to route voice traffic directly from one PBX to another. The single hop between PBX's improves compression quality. It eliminates redundant voice trunks needed for Tandem PBX switching. And ATM PNNI provides rapid dynamic re-routing when there is an outage.

Cisco / StrataCom has been doing voice for 11 years, this is mature technology. What's changing fast now is the cost of implementing it, and the platforms it is implemented on.

Technology Issues

Digital signal processors (DSP's) have gotten a lot cheaper (Moore's law). They're used to process voice and perform voice compression. Voice compression can give good quality at 8 Kbps, especially if multiple compression cycles are avoided.
 
Compression Algorithm Data Rate
G.711 PCM 64 Kbps
G.726 ADPCM 32 Kbps
G.728 LD CELP 16 Kbps
G.729 CS ACELP   8 Kbps

Technical issues in dealing with voice compression include delay and delay variation (jitter), echo cancellation, background noise, silence suppression, and language sensitivity. Other issues: avoiding double compression cycles (Tandem PBX's), and distinguishing modem and fax traffic (which can't be compressed, or can't be compressed in the same way).

Of these issues, delay and jitter, and double compression, are network design issues. They impose constraints on where voice networking is going to work as well as desired. We'll return to this topic later.

The remainder is technical issues, quality of the voice coding circuitry and software. Note that silence suppression saves at least 50% of the bandwidth, since there's only one speaker at a time (in polite conversations, anyway). So if you put all that together, using compression and so on offers the prospect of trading standard PCM at 64 Kbps for silence suppressed CS ACELP at perhaps 4 Kbps or less.

Let's convert that to the (over-simplified) sales argument for this. If your company has multiple T1's carrying voice circuits between sites, and you can reduce your leased line charges to 4/64 = 1/16 or 8/64 = 1/8 of what they are now, well, that could be a healthy amount of money. Realistically, the savings aren't that simple, because there are other benefits and costs (and because I don't want anyone getting mad at me if it works out differently for you).

There are alternatives: buying dedicated gear that does voice only over trunks, between PBX's. Doing it yourself offers the chance to combine voice and data trunks, reducing leased line costs and perhaps combining some of the operations and management overhead as well. You may be able to deliver a higher quality of service as well.

The Choices, Voice over WHAT?

Let's look at the choices Cisco offers. They now have offerings in all three of the relevant areas.
 
Voice over _______ Positioning Other Backbone Your Backbone
Frame Relay (VoFR) Enterprise branch access networks 
Frame Relay Service Providers
Frame Relay Service Provider's Private Frame Relay
ATM (VoATM) Multi-service Carrier ATM backbone 
Enterprise WAN ATM backbone 
Enterprise branch access networks
ATM Service Provider Private ATM or hybrid
IP (VoIP) Enterprise branch access networks 
Leased line router networks 
Voice over Business Class Internet 
PC desktop applications
IP Service Provider 
IP VPN Provider 
Internet (*)
Enterprise IP 

 From the corporate perspective, there's a basic choice if you're going to mix your voice and data traffic. Do you try to do your own trunking between sites, or does a service provider do it for you?

The argument for doing it yourself is cost, control, and perhaps quality. If you aren't getting the quality of service you want, persuading a service provider to give better service could be a bit tough. Perhaps in the future we'll be able to pay more and get lower latency; that market is still trying to emerge. The argument against is doing it yourself is cost (staff, staff skills, equipment and management, maintenance, etc.).  But let's not re-fight the Battle of Outsourcing right here on this spot!

Voice over Frame Relay / ATM

The new Cisco voice product of note for Frame Relay or ATM is the MC3810.

The MC3810 offers a choice of  either six analog voice ports, or a T1/E1 digital voice module (DVM). It also has T1/E1 multiflex trunk, ISDN backup, Ethernet LAN, and two standard Cisco router serial interface ports. It provides up to 24 channels of 8 Kbps compressed voice, and fax, over Frame Relay or ATM or HDLC. It can handle off-net dialing to the public phone network, on-net to off-net call routing, and fax or fax over IP. It's a Frame Relay or ATM mux, it's a router, and it does  voice too! It can use some dialed digits to route calls and then pass the remaining dialed digits to a central PBX.

The multiflex trunk divides the T1 by time slots, allowing N x DS0 for Frame Relay or HDLC, M x DS0 for PCM voice, and K x DS0 for TDM channels or video. The multiflex can alternatively be used to do T1 or E1 ATM instead. ATM use supports MPOA, CES voice, or compressed voice in AAL5 cells. Built in channel bank allows cross-connects from the serial interfaces and the voice modules to slots on the multiflex trunk (MFT). Signaling is similar to the 3600 voice card, see below.

What this all means is that you can put a 3810 at a remote site, and have the remote phones look like extensions on a central  site phone switch. You could also have some calls go off-net. You can also route data and provide CES video if needed: an all-in-one box!

MC3810 call connection possibilities:

Using the multiflex capability, you or your service provider could connect to you via T1. The T1 could be set up so that it provides 6 voice circuits to the public phone net (PSTN), and Internet link, 384 K for video, and 256 K Frame Relay carrying data and voice to a central site.

The StrataCom IGX voice modules will provide somewhat similar capabilities for voice in an ATM setting, at higher densities. Larger enterprises needing more than 24 compressed voice channels may want to investigate this further. The MC3810 capabilities are designed to interoperate with Universal Voice Module cards (UVM) cards for the IGX.

One has to be slightly careful to distinguish here between ATM circuit emulation (CES), and compressed voice. In the former, the ATM switch acts as super-mux, providing some fixed amount of bandwidth. This bandwidth is committed and cannot be used, whether or not a voice call is taking place. Voice compression and so on do not apply to CES. Compressed voice is treated as (special) data, potentially saving bandwidth. You can send compressed voice over ATM, but if you look to recover bandwidth for data, you'd better send the voice traffic as VBR and not CBR.

Relevant standards: FRF.12 (Frame Fragmentation) and FRF.11 (Voice Signaling Carried in Frame Relay).  The first of these is appropriate to fragment large Frame Relay frames into small 53 or 77 byte frames, so that the small voice frames can leave the router faster, without waiting for an entire 1500 bytes data frame to be transmitted. The second is so one compliant device can signal another across Frame Relay.

The Cisco devices are, I believe, not FRF.11-compliant, since Cisco needs an extra byte for Voice Network Signaling (VNS), to allow Tandem PBX bypass.

Currently, the MC3810 requires static mapping of E.164-like addresses to Frame Relay PVC's if Frame Relay is used. There is a design issue here: do you use a full mesh of Frame Relay PVC's, to carry voice traffic directly to its destination, or do you route via a central site, at the cost of two hops and poorer voice quality? Your choice! The practical issue: finding a Frame Relay service provider who can sell you low-delay PVC's.

Voice over IP

Cisco has announced a voice card for the 36xx routers (3620, 3640). It provides 2 or 4 analog voice circuits over IP (digital circuits are coming soon). This gives you up to 12 analog circuits in a 3640. If you need more, use the digital circuit capability. Also consider that at typical voice ratios of circuits to users, this might support many times more users, depending on your sites' calling patterns.

If your 2500 is running out of steam in some "power branch offices", this adds to the possible value of upgrading to a 3600 model.

Cards for other router models are in the works. Phones and faxes can plug directly into the 3600 card, or the key system or PBX can connect to it. A Java application is coming to configure voice ports, dial plan, and manage the system. Reporting is also in the works.

In terms of signaling, the voice card does E&M, FXO, FXS:

The Voice over IP stack (VoIP) uses the H.323 and H.235 standards for signaling and session protocols. RSVP is used to request guaranteed quality of service.

The point to the voice support in the 3600's is that the corporate IP network is made to appear as a trunk line to the PBX's.

When there's a call, a PBX can signal the connected router via the Q.931 message format (ISDN). This is passed to H.323 (which uses Q.931 for call signaling), and at the other end is passed as a line seizure to the PBX. The routers then forwards dial digits to the PBX. If the router can interact with the signaling, it opens the door for additional functionality. (Note that Cisco recently bought LightSpeed International, Inc., adding to their phone signaling capabilities).

The expected applications are:

All of these potentially reduce costs and/or enhance manageability.

One (somewhat future) attraction of this is that IP networks are connectionless, so if E.164-style telephony address are mapped to IP addresses, dynamic IP routing rather than static call tables can be used. This has the potential to simplify managing a voice network.

The 3600 cards are intended to have VoFR and VoATM capabilities in the future.

Design Issues

Just a couple of important things, since space is short and design for low delay is a complex subject.

The MC3810 requires one at each end. Interoperability with an IGX card is intended; the card may be available by the time this article appears. The IGX card would allow higher density at the central site, aggregating the voice connections to all the remote sites with 3810's.

I noted elsewhere that delay is the biggest factor. That's been repeated since it's important. This is the current real barrier to corporate voice over the Internet. Fax relay traffic is a fairly easy win, though: fax delivery doesn't have to be real-time. Be careful here, too, I gather that fax machines have tight end-to-end timing, so that you do need low delay for real-time fax delivery.

Some ideas on how to bring the end-to-end delay down:

Links

Voice/data articles


Dr. Peter J. Welcher (CCIE #1773, CCSI #94014) is a Senior Consultant with Chesapeake NetCraftsmen. NetCraftsmen is a high-end consulting firm and Cisco Premier Partner dedicated to quality consulting and knowledge transfer. NetCraftsmen has nine CCIE's, with expertise including large network high-availability routing/switching and design, VoIP, QoS, MPLS, network management, security, IP multicast, and other areas. See http://www.netcraftsmen.net for more information about NetCraftsmen. Pete's links start at http://www.netcraftsmen.net/welcher . New articles will be posted under the Articles link. Questions, suggestions for articles, etc. can be sent to pjw@netcraftsmen.net . 



3/98
Copyright 1998, Peter J. Welcher