Janaury 2000

QoS Sets The Stage For IP Telephony Applications

BY BROUGH TURNER

We have seen IP telephony present an engaging first act — dramatic growth in VoIP arbitrage and rapid technology deployments to support VoIP in the telephony network's backbone. But before the applause fades, the curtain will rise on an even livelier scene — the proliferation of IP communications devices and IP telephony applications. Right now, behind the curtain, Quality of Service (QoS) architectures are rising, forming thefor the introduction of IP telephones in the enterprise and public Internet services supporting voice applications. QoS, it seems, has a key role in making IP telephony a box office hit.

INTERNATIONAL ACCLAIM/AVOID TYPECASTING
In the area of international arbitrage, VoIP has attracted quite a following over the past 12 months. According to IDC, VoIP has gone from negligible traffic to 4 percent of all international voice minutes. Meanwhile a major focus in the industry has been on the technology needed to put VoIP into the backbone of the telephone network — soft switches, decomposed gateways, and quality of service (QoS) in the backbone.

At this point, everything is focused on replacing the current telephony backbone with an IP telephony backbone, but for use with the existing telephone sets. The current focus is reasonable. After all, if you include mobile phones, there are more than 1.2 billion telephone sets in the world today. Clearly, a worthwhile market.

However, interesting new applications will appear only when there are new IP communications devices — IP phones or their equivalent — directly connected to, and usable over, the Internet. That’s a more challenging role for VoIP.

THE IP-PBX — A STAR TURN?
IP telephones will show up first in closed user groups, that is, in business telephone systems. Most traditional PBX vendors have IP-PBXs in development. Some are already for sale. During 2000 and 2001, the first major enterprises will convert partially or totally to IP phones.

With an IP-PBX, there is still a gateway function to connect back to legacy phones and the legacy telephone network, but new IP phones plug into Ethernet and connect phone-to-phone using IP packets over the LAN. There are steps you can take (at either the IP level or the Ethernet level) to guarantee adequate voice performance, but in the end voice-on-the-LAN works because LAN bandwidth is low cost — you can over-provision a LAN fairly inexpensively.

IP-PBXs, or at least IP extensions to existing PBXs, are about to be deployed en masse. All the major vendors have something in the works. Market estimates are for a greater than one billion dollar market within two to three years. So IP telephone sets are not far off, albeit only for use with IP-PBXs.

What does it take to have these new IP phones connect from one enterprise to IP phones in another enterprise? The answer is voice-quality IP bandwidth between the sites. We need some way to guarantee that packets get through between the sites despite congestion from Web browsers, file transfers, and so on. In other words, we need to be able to specify and receive an adequate QoS — from our Internet service provider(s).

QoS BEHIND THE SCENES
Several architectures have been developed or proposed to provide specific QoS for particular packet flows or for classes of packets. Pre-Internet solutions include virtual circuits as defined for ATM and Frame Relay and the centrally managed service classes of IBM’s System Network Architecture (SNA). For the Internet, the IETF has developed and proposed several approaches.

IntServ
Integrated Services (IntServ) defines a mechanism that provides specific service levels, such as a guaranteed maximum transmission delay for end-to-end sessions on the Internet. The IntServ architecture requires that every router along the path have both adequate processing power and adequate memory for tracking each session and reserving resources for it.

IntServ introduced the Resource Reservation Protocol (RSVP) as well as the idea of admission control (don’t let new traffic enter the system if adequate resources are not available) and traffic shaping (packets entering the network are spaced out according to the pre-arranged resource reservation).

Unfortunately, IntServ requires that per-session information be stored at every intermediate router, not to mention per-session, per-router negotiations to set things up. This approach doesn’t scale very well in an Internet that will have more than a billion users within the next decade.

DiffServ
A more scalable alternative is Differ-ential Services (DiffServ). With DiffServ, packets are marked to receive one of 64 possible treatments within the network,
so in the core of the network there is no specific knowledge of end-to-end sessions, only of a few specific, per-hop behaviors that are applied when forwarding appropriately marked packets. DiffServ still relies upon admission control and traffic shaping at the edges of the network to ensure that incoming traffic doesn’t exceed the network’s carrying capacity, but information and processing requirements within the Internet core are kept to a minimum.

HOW QoS MAY PLAY OUT
One view of how QoS will play out is IntServ in the LAN, DiffServ in the backbone. Alternately, IntServ in the LAN may not be important to all. It may be that for many, just plain over-provisioning, with or without simple differential service, is good enough.

QoS In The Backbone
Already, you can purchase backbone Internet services with certain guarantees. Today, these are provided either by over-provisioning (new players like Quest, Level 3, and Enron have substantial fiber capacity) or by running IP over ATM. But as WAN service is still more costly than LAN infrastructure, substantial investments are being made in backbone QoS architectures that use native IP technologies, such as DiffServ. Many of today’s backbone routers already contain the necessary queuing mechanisms to support either IntServ or DiffServ, and active interoperability experiments are being conducted.

With DiffServ in the backbone, three classes of services may be adequate:

• Low-latency, guaranteed bandwidth for live voice or video calls.
• Interactive response for client/server applications, Web browsing, or other applications where the bandwidth is not constant, but a rapid response is desirable when someone presses a key.
• Bulk transport to handle applications such as FTP and e-mail.

Services based on DiffServ should emerge within the next two years. Less clear are the business arrangements that would involve separate backbone ISPs so that a particular DiffServ class of service would be available across a significant portion of the Internet. But the IETF is defining per-hop behaviors, and it’s likely that business discussions are already in progress outside of the IETF.

QoS At The Edges
The real QoS issue arises in connecting the enterprise to the WAN backbone — through what is today a narrow and expensive pipe. This is the point where the 10 Mbps, 100 Mbps, or even 1 Gbps of the LAN is scrunched down to fit into what is typically a T1 or smaller pipe. This link is small because WAN access connections — the last mile — remain the most expensive part of any network. So the issue for WAN services is providing admission control and traffic shaping at the enterprise end of the WAN access link.

Figure 1 shows IADs at the enterprise end of the WAN access links. In the past, these devices used time-division-multiplexing (TDM) to allocate telephony bandwidth between telephony and various data services such as Frame Relay. But now, integrated IP products are emerging.

Figure 1. Combining QoS approacheds. BR: border router; IAD access device; SLA: service level agreement; TCA: traffic conditioning agreement

The new, integrated IP products accomplish packet classification and IP traffic shaping. They classify incoming packets from the LAN, maintain separate queues on the output (WAN) ports at the enterprise end of the bottleneck, and allocate which queues get to send specified quantities of bits into the shared pipe. Packet classification is necessary down to details such as the packet type (for example, a VoIP packet that is part of a new or ongoing conversation, an interactive packet, or a bulk e-mail packet).

Cisco, Xedia, and QWES.com are companies with a focus on admission control and traffic shaping of IP traffic on narrow links — for example, on the T1 link from an enterprise into the WAN. At the moment, Cisco has a variety of pre-standards capabilities built into their routers and a roadmap. Xedia (recently purchased by Lucent) has bandwidth management products for the WAN access link that scale from multiple T1s up to OC-3 WAN access. And QWES.com has both box-level and CompactPCI products to manage bandwidth at the single and dual T1/E1 level. Their CompactPCI product is directly useful to VoIP gateway vendors with CompactPCI-based gateways.

The more carefully you can allocate bandwidth on the bottleneck link the better. For example, if you merely give all voice calls higher priority than anything else, you could find users of interactive enterprise applications complaining of poor response. And, even if you can provide some guaranteed response for the interactive users, you could find the twelfth voice call suddenly degrades speech quality for all eleven callers who were previously talking. The ultimate system must support per-application bandwidth allocation and also provide admission control, so for example, new calls are prohibited when all available voice bandwidth is in use. Once the traffic reaches the WAN, DiffServ is likely to be adequate, but on the WAN access link, careful management is essential.

A SEASON OF NEW VoIP APPLICATIONS
So the stage is being set to move VoIP beyond “cheap” voice services and into new applications. The next 24 months will see the emergence of real IP telephones, at least within the enterprise. And the next 24 months will also see the emergence of public Internet services with QoS capable of supporting voice applications. Click-to-dial buttons will be widespread on e-commerce sites. With premium bits purchased for a set premium over bulk bits, per-minute charges for pure VoIP calls will disappear, although we’ll still see per-minute charges for access to legacy telephones via VoIP gateways.

At this point, the pressure will be on service providers to provide new services — simple connectivity will no longer be enough. The real impact of VoIP will emerge 24 to 36 months from now. I for one look forward to phones that actually sound good! Anyone for CD audio quality phone calls?

Brough Turner is senior vice president of technology at Natural MicroSystems, a leading provider of hardware and software technologies for developers of high-value telecommunications solutions. For more information, visit the company’s Web site at www.nmss.com. E-mail to the author is also welcome at rbt@nmss.com