January 2000

Guaranteeing Application Performance Through IP Quality Of Service

BY DANA RASMUSSEN

Implementing quality of service (QoS) refers to the performance of network traffic as it flows through networks. The growing need for end-to-end QoS has been driven by the growth in the use of all networks (including the Internet) and by service providers’ desire to obtain high-margin business customers. Attracting business customers has become vitally important to service providers as they strive to improve their financial performance, which has been degraded by the decreasing margins available for consumer network services.

As corporations strive to maintain a competitive position, their need to focus spending on primary business objectives becomes essential. As such, corporations are looking to their networking partners to provide new services that ensure business critical applications obtain the best possible performance, at the best possible price, across their networks. In recent years, the Internet has had increased acceptance in the corporate network world, and its ability to offer more cost-effective network solutions is appealing to businesses. Today, businesses are using the Internet for remote access, information searches, e-mail, and other applications, but do not yet rely on the Internet for all of their networking needs. New services like those listed below, that exploit the potential cost savings of the Internet, are becoming quite attractive to business customers:

• Virtual Private Network (VPN). A private data network that makes use of a public network (such as the Internet), maintaining privacy with tunneling protocols and advanced security features. The basic premise behind a VPN is to give a company the features of a private network at the lower cost of the shared public infrastructure.
• Voice Over IP. Sending voice information in digital form in discrete rather than in the traditional circuit-committed protocols of the public switched telephone network (PSTN). The major advantage of Voice over IP (VoIP) is that it avoids the tolls charged by ordinary telephone service.
• Broadband Access Services. “Always on” high-speed access methods, such as xDSL and cable, that provide transfer speeds which are nearly 1,000 times faster than existing analog modems. These high-speed access products are able to take advantage of the increased bandwidth availability at the core and extend it to the local loop.

These services can represent great opportunities for business customers to reduce network costs and reduce the risks associated with deploying new technologies. From the service provider side, corporate networking services constitute a new and profitable revenue stream for providers who can solve the security and performance drawbacks of the current Internet. To capitalize on the need for these new services, it is imperative that service providers be able to implement end-to-end QoS in the network. In effect, the data network must become as reliable as the existing telephone network.

GUARANTEED SERVICE
To capitalize on corporate demand for these new services, service providers must provide additional levels of guarantee that include significant penalties if quality standards are not met. QoS agreements between the enterprise and service providers offer monetary penalties if agreed upon end-to-end performance metrics are not met. These QoS or Service Level Agreements (SLAs) allow enterprises to take advantage of new technologies with limited risks. SLAs are available to service provider customers and may range from a simple, standard contract for consumers to a customized agreement for business customers. An SLA typically defines end-to-end service specifications and includes guarantees for variables such as availability, latency, throughput, and packet loss rate.

To build a network with end-to-end QoS that has the ability to offer SLAs, the first necessary function is the ability to differentiate according to different flows of traffic. This can apply to different application flows within a single customer’s aggregate traffic, or it can be applied to differentiate one customer’s traffic from another customer’s traffic. Additionally, QoS offerings must include traffic conditioning: Policing, shaping, and marking.

Although the industry has yet to converge on a single network architecture for end-to-end QoS, there has been a great deal of work done within the Internet Engineering Task Force (IETF, www.ietf.org) regarding QoS. Resource reSerVation Protocol (RSVP) and Integrated Services (IntServ) specifications have gained some success in the enterprise, but little acceptance with service providers. IP Differentiated Services (Diffserv), Assured Forwarding (AF), and Expedited Forwarding (EF) Per-Hop Behaviors hold some measure of promise as building blocks for delivery of true end-to-end QoS services, but that promise is only now being realized.

RSVP. RSVP allows performance-sensitive applications to reserve Internet bandwidth, so that the given applications can operate at the proper performance level. The ability to reserve Internet bandwidth combats performance issues such as latency, jitter, and packet loss. With RSVP, users who wish to utilize a particular application can reserve bandwidth through the Internet and be able to receive it at a higher data rate and in a more dependable data flow than without RSVP.

Integrated Services (IntServ). IntServ is a “circuit-oriented” model, which abandons the fundamental strength of the “packet-oriented” Internet Protocol (IP). Because all routers are required to maintain “state” information (amount of bandwidth and buffer currently reserved), the ability to scale is severely limited. Also, router cycles are wasted on reservation capture and processing.

Differentiated Services (DiffServ). Defines ways of assigning specific service levels and priorities to IP traffic and supplies a very customizable QoS mechanism based on different classes of service. Traffic classification allows packets to be prioritized according to the needs of specific applications. Class of service networking ensures that traffic that is designated as high-priority always takes precedence over lower-priority traffic.

NEW SERVICES ARE THE BOTTOM LINE
Prior to these QoS advances, particularly those associated with DiffServ, businesses were reluctant to place their mission-critical voice and data applications onto public IP networks. Today’s service providers realize that to attract business customers today, they must provide new and customized services, with QoS guarantees.

By implementing these services, providers can look to improve their profitability by increasing revenue and attracting high-margin business customers. This will be done with new services that supply quality guarantees at a higher price point. Incremental savings will also be realized because of the more efficient use of existing network resources. Service providers can also expect to improve their competitive position, by providing customized services for business partners that are based on their specific network requirements. By moving away from the generic “one-size-fits-all” network solution, service providers can also expect to reduce their customer churn.

As of today, the QoS story is far from complete. However, the availability of end-to-end QoS will be a driving force behind the acceptance of carrier solutions for mission-critical business networks. c

Dana Rasmussen is vice president, product management for Unisphere Solutions, Inc. Unisphere, a Siemens company, develops platforms that enable service providers to create and deliver such innovative services as e-commerce, ERP, outsourced productivity applications, and VPNs. For more information, visit Unisphere’s Web site at www.unisphere.com 

BY LINDEN DECARMO

Although everyone assumes that cable telephony will succeed, there is intense debate about how it should be deployed. Several cable telephony architectures are vying for dominance, yet all have one common element: The Session Initiation Protocol (SIP).

SIP provides call setup and teardown functionality, and the cable industry is attracted to it because of its simplicity, services, and cross-network communication and expansion abilities. A significant advantage of SIP is its simplicity. You can create a functional SIP application by parsing only a few headers. Furthermore, since it is a text-based protocol, it is easy to debug. Ease of debugging is critical since most cable telephony devices are still under development.

Service creation is also facilitated by SIP. Because consumers crave services such as call waiting and voice mail, the service creation market can be extremely profitable. Although you can implement services with the Public Switched Telephone Network (PSTN), it is an arduous process. By contrast, SIP’s service creation dramatically reduces the work effort to create services. Consequently, SIP will enable cable companies to rapidly produce revenue-generating software.

SIP is useful for communication between disparate cable networks as well. For instance, if you want to call someone on another cable provider’s network, SIP is used to request resources and find the destination partyon the other network. In fact, because SIP is the common denominator for packet-based cable telephony, it is the only way to place a call between cable telephony architectures.

It also can be used as a gateway to PSTN resources. Most cable telephony protocols are tailored for packet-based communication and only have rudimentary PSTN support. Since SIP has richer PSTN capabilities, these protocols often rely on SIP gateways to access PSTN networks.

Finally, SIP is attractive because of its expandability. Since the cable telephony market is in its infancy, features are continually being refined. Therefore, it is imperative that SIP be able to adjust to these changes. Fortunately, SIP can not only incorporate new features, but it has specific provisions to preserve backward compatibility.

TROUBLE ALONG THE WAY
Although SIP is an integral part of packet-based cable telephony, the cable industry has had to overcome glaring weaknesses in the specification. For instance, the original SIP protocol had no mechanism to reserve network or device resources before a call was connected. If you don’t reserve these resources, it is possible that the call could be disconnected or experience audio breakups.

To avert such difficulties, the cable industry has proposed that SIP use a two-stage call setup process. The first stage reserves network resources and multimedia resources on endpoints (or multimedia terminals), and the second stage connects the parties. This eliminates the possibilities of disconnects or breakups.

Even more problematic are the security holes in SIP. As we previously discovered, SIP’s text-based nature makes it easy to debug. However, these plain text packets also make it vulnerable to hackers. For instance, it is trivial to determine the IP address of the originator of a SIP packet. Thus, even if you have caller-ID blocked, someone can still figure out the number where you initiated the call. Consequently, efforts are underway to fortify SIP’s caller privacy capabilities.
A related security problem when deploying SIP is the prevention of fraud and theft of service. To explain, SIP provides no features to detect the unauthorized usage of a network. However, it does provide hooks so that you can enable robust security features. For instance, some cable telephony architectures enable you to place trusted SIP guard servers on their networks to authorize network usage and prevent fraud.

The final issue with SIP in a cable environment is performance. SIP’s strength is its simplicity and the minimization of packets that must be exchanged between endpoints before a call can be connected. Unfortunately, the enhancements necessary to create a robust cable network (i.e., two-stage reservation and fraud prevention) can dramatically decrease performance. For instance, a poorly implemented two-stage reservation process will result in an excruciating wait before the destination phone rings. Therefore, the cable standards bodies are actively researching how to implement these enhancements while maintaining the performance of the original SIP protocol.

SIP is the common denominator for cable telephony protocols because it is simple, service-friendly, and very extensible. However, it does have noticeable deficiencies when it is used with cable. Fortunately, SIP’s flexibility will not only allow the cable industry to address these weaknesses, but also adapt to future challenges.

Linden deCarmo is a senior software engineer at NetSpeak Corporation, where he develops advanced call agent software. He is the author of technical articles and a book entitled Core Java Media Framework. You can reach him at [email protected]. NetSpeak (NASDAQ: NSPK) is a leading developer and marketer of advanced telephony solutions for IP networks. For more information, visit the company’s Web site at www.netspeak.com.

BY AMI AMIR

I was involved in VoIP before I even knew that it existed. So — I should know what I am talking about. Right? Wrong.

In spite of the fact that for two years our programmers were using Linux on their PCs, I never took the second most popular PC operating system seriously. We (RADVision) started offering a Linux version of our H.323 toolkit only this year. To paraphrase Julia Roberts in Pretty Woman: “Big Mistake.” So, probably a large part of what I am going to claim for the future is wrong.

THE VoIP STANDARDS WAR THAT WAS
A few months ago, there was a raging electronic war going on between two major camps. The Internet Engineering Task Force (IETF) camp was represented by Telcordia, Level 3, Cisco, and some teams within Nortel Networks. The International Telecommunication Union (ITU) camp was represented by Ascend (now part of Lucent), Lucent, Siemens, other Nortel Networks teams, Intel, and Microsoft. E-mails were rampant and emotions ran high.

Of late, there has been a very positive change, and the religious wars are giving way to reason:

First consensus — Transport. There is no argument that the IETF’s work on SIGTRAN, as a transport protocol of SS7 and ISDN signaling over IP, will be used by all. Likewise, the IETF’s Real-Time Protocol (RTP) is used by all as the media transport protocol.

Second consensus — Gateway Control. The control of decomposed gateways is covered by the MEGACO work of the IETF and the H.GCP ITU draft. Both works are headed for a merger, replacing Simple Gateway Control Protocol (SGCP), Media Gateway Control Protocol (MGCP), Internet Protocol Device Control (IPDC), and Media Device Control Protocol (MDCP) (did I forget anyone?).

Third consensus — Signaling Gateways. The ITU H.246 gateway standard will include issues formerly covered by IETF work on SS7. It will be the only standard for signaling gateways.

No Consensus (yet) – Endpoint Signaling. The only “big” open item is SIP vs. H.225.0 (the H.323 signaling standard) as the endpoint call signaling protocol. My guess is that H.225.0 shall win. I think that at the end of the day, it does not matter which will win. Even if two standards remain, the situation would be much better off than in many other industries.

THE REAL PROBLEMS?
Quality of Service (QoS). QoS does not exist. There is not a single QoS end-to-end solution that actually works and provides predictable, consistent performance. RSVP, DiffServ, and 802.1p will not solve the problem. We have a long way to go yet, and customers will be very reluctant to pay for inconsistent and unreliable service.

Scalability. To date, no one has deployed a sufficiently large network that supports millions of calls per minute. So how can we be sure that the solutions actually scale? We just hope it all works.

Security. Solving the security issues is tough. There needs to be network-level security to protect against hackers that can bring down an entire network. Witness the Melissa and Chernobyl viruses. There needs to be control and management level security that will prevent outside interference in network configurations, manipulation of user data, erasure of billing information, and toll fraud.

Privacy. VoIP privacy is a joke. The current implementations are wide open. Once you decide to use VoIP, you need to assume that everyone who has access to your network can participate in your call. And how do you actually know who has access? A single rogue PC with remote desktop sharing software could open your network to the whole world.

OSS. Operations, Services, and Support systems are not yet in place. Out of the total cost of operations of regular telephone networks, only a small percentage goes into switching and routing calls. A much larger part of the money is being spent on managing facilities, billing, and provisioning. An economically viable VoIP network needs to reduce the costs of OSS. There is no solution in sight.

THINGS TO CONSIDER
There are two big debates that really drive the work we all do now. The first involves the smart network/dumb terminal vs. the smart terminal/dumb network. The old phone system was based on a dumb terminal (your phone) and a network that tried, albeit not very successfully, to be smart. H.323 evolved around the concept that since telephone networks are dumb, let’s build a smart terminal. SIP evolved around the concept that since we can build smart IP networks, let the endpoint be dumb (and hence cheaper to build).

My guess is that we will end up with the best of both worlds, a smart terminal (“if my in-laws call, I am out”) that connects to a smart network (“I need to talk to Joe, please find him for me”).

The second debate is the large homogeneous networks versus the small service providers.

AT&T, Level 3, and Qwest adopt a model of a single seamless VoIP network (obviously owned by them) that is centrally controlled via hierarchical mechanisms. MGCP assumes a centralized control mechanism of each call via a call agent. That is why they promote MGCP. The concept is centralized. A single call agent “owns” the call.

Another model exists, of a flat, non-hierarchical network served by multiple providers and multiple gatekeepers. This is the underlying model of H.323, which assumed from Day 1 that the calling party and the called party are registered with different gatekeepers.

Which is right? The answer is obvious. Both solutions shall exist. The architecture that fits all models is a distributed network server architecture that allows one provider to provide the pipes, and another vendor to provide the services — an architecture where the transport is disassociated from the control and the services. We need to adopt a concept similar to SS7 in PSTN — an architecture that really allows each provider to customize the network and provide unique services.

SO WHAT’S GOING TO HAPPEN?
Toll bypass is dead. It is not cheaper to make calls on IP compared to regular phone networks. Now that regulatory barriers are down, and due to the glut in fiber-optic cable (thanks to those brilliant people who invented WDM), the cost of making long-distance phone calls has plummeted all over the industrial world. The IP-based toll bypass business, as a means to save on long-haul phone calls, is dead.

This does NOT mean that VoIP is dead. On the contrary, the VoIP industry is just starting to evolve. The opportunities are unbelievable. It is just a matter of time before VoIP shall reign. The MEGACO and H.GCP standards are merging. H.323 is definitely here to stay. The IETF carries the day as far as transport (RTP and SIGTRAN). The future of SIP is not yet really clear, but I expect to see SIP-based terminals deployed.

Initially, VoIP will be used mostly inside organizations, offering a way to create a “global PBX” for the enterprise. Internet call waiting and Internet commerce will become major forces driving deployment of VoIP gateways. Cable modems will support VoIP (using SIP), and will provide a way for CLECs to bypass the LECs without having to install copper wires.

Integration between the mobile/wireless network and the IP network SHOULD happen, but God knows when. Who stands to gain and who will lose? Everybody stands to gain! No one is the loser in this new game. There are huge opportunities for equipment vendors, systems integrators, and new and incumbent service providers.

Finally, all of us, as users of the new network, will be the biggest winners of all.

Ami Amir is CEO of RADVision, Inc. RADVision develops industry standard “building blocks” needed for supporting voice and video calls on packet networks like the Internet. RADVision core technology spans a broad range of product environments, from chipsets and hand-held devices through carrier class gateways and switches. For more information, visit the company’s Web site at www.radvision.com.