February 2002

BY HENRY DEWING AND STAN POTTER

Lately, we�ve heard that voice mail systems from Atlanta to Anchorage are overloaded with messages from IT and voice vendors offering VoIP solutions. In today�s tightened economy, vendors are searching for additional sales to make their quotas and trying to convince enterprises that implementing a VoIP offering will help the enterprise�s bottom line. Some vendors concentrate on the financial return of implementing the system, while others emphasize the new features VoIP brings. In fact, new features may no longer be unique once a solution is carried from planning to implementation. IT managers must carefully evaluate the offerings, considering what developments may occur in the short term future, and select a solution that satisfies the enterprise�s business models and provides a return on investment (ROI.) A vendor�s claims will not materialize if the vendor cannot provide guidance toward implementation of a system that reliably delivers VoIP service over those spiffy new IP phones.

Network managers must understand that implementing VoIP will affect how business is done on both the voice and data sides. Legacy equipment, including routers and switches, must be properly configured to accommodate voice and QoS. Hubs, which cannot employ packet classification to discriminate between traffic, should be avoided in a QoS environment. QoS is as important to VoIP as 911 access is to traditional voice networks and security is to traditional data networks. For example, an end-to-end (E2E) QoS strategy sets up each component of the system to deliver the dial tone that users are accustomed to hearing even when the building loses power.

A company in the IT industry responded to the voice mail messages VoIP vendors had been leaving them and selected a reputable IP vendor for its solution. (This software company is located in California but prefers not to be named.) The organization listened to the vendor about QoS strategies and implemented them on their routers and policy servers. The company�s IT experience made it ideally suited to turn up a cutting edge converged communications solution. After the implementation, the enterprise network manager was surprised to find that the system could not achieve reliable results -- a percentage of the calls had �bad connections� with conversations getting clipped regularly. The first word of a sentence was occasionally lost, which resulted in halting, somewhat stilted conversation that adversely affected the end users. Management found this level of functionality unacceptable and decided to question the entire implementation. After several heated meetings and a second network study, the vendor realized that the company was not using a loop-free layer 2 topology, which caused temporary Spanning Tree Protocol-induced network outages. Despite the vendor�s earlier recommendation to the contrary, the company had not changed this network configuration because the IT staff had been wedded to it for its fast connections for data traffic and had mistakenly assumed it would be great for voice, too.

As this example demonstrates, the need to manage traffic intelligently across the entire infrastructure is especially important in the VoIP environment. To achieve efficient and effective traffic management, network managers need to educate themselves about QoS options. Enterprise network managers in particular need to be proactive in seeking answers to QoS questions in order to assess vendors� offerings. Many of them are highly complex so network managers may not have time to address all of the potential solutions, but it is important to understand the basics of the tools available and what might be appropriate for their network size and population.

Unfortunately, end users sometimes relegate QoS to the domain of the carrier. In fact, recent editions of Newton�s Telecom Dictionary begin the definition of QoS as �the measure of the telephone service quality provided to a subscriber,� with measurements such as volume and length of time until dial tone is achieved. However, QoS has evolved along with the convergence of networks and now encompasses additional considerations such as packet loss.

Intel commissioned research of more than 300 organizations and the results demonstrated that QoS is a most-recognized but least-understood aspect of network management. The research revealed that users plan to rely on vendors or consultants to develop a QoS strategy. These relationships will be helpful to IT managers, but they must secure an understanding of QoS to be able to select vendors or consultants.

QoS is a system-wide solution involving hardware, software, and network management that should be viewed from the PSTN to the desktop. It comprises the mechanisms that give network managers the ability to control the mix of bandwidth, delay, variances in delay (jitter), and packet loss in the network. Consistency in how the various switches and routers define parameters and service level agreements is required. Running QoS over only the WAN segment or only the LAN segment of a network will not ensure E2E coverage and will not eliminate all potential congestion.

Adopt A System-Wide, E2E Plan
Enterprises have evolved their network architectures from separate networks to consolidated and tiered network designs, thereby creating multiple points of convergence and speed mismatches. For example, a speed mismatch will occur in a LAN-to-WAN gateway where 100 Mbps Fast Ethernet feeds into a slower T1 or E1 trunk. QoS should entail a method of allocating resources in the network at multiple points of convergence so that data gets to its destination quickly and consistently.

A QoS strategy must be conceived before the first IP phone or gateway is installed. A rigorous planning and design process to ensure feasibility up front will avoid headaches and expenses down the road. Network managers should take into account the size and number of sites and the expected VoIP traffic (and other delay sensitive streams like video and audio). They should also consider whether IP phones are powered directly over the LAN from the Ethernet switch, and whether or not the gateway allows for automatic fallback to PSTN circuits in the event of quality degradation. This strategy must be collaboratively determined, followed during the implementation and tracked with on-going surveillance and fine tuning. Whether IT professionals maintain their own infrastructure or pay someone else to do it, they must ensure that on-going performance is monitored and that the network is tuned and optimized based on the current traffic profile.

Screen For Inherent Bias
Of course, implementations will vary in scope. The size of the implementation is important; however, research reveals that understanding whether staff is more proficient with voice or data applications and systems could become the key success factor when blending both types of traffic into one system. Staff background is important because their area of specialization will influence their implementation strategies. The Spanning Tree Protocol that nearly killed the VoIP implementation discussed earlier was the legacy of an IT professional more experienced in data networks.

Generally speaking, voice-centric managers will understand and design an infrastructure that is highly reliable and capable of handling VoIP without issue, avoiding topologies that drop packets or provide all packets equal access to network bandwidth. On the other hand, it is generally the data-centric managers who are comfortable controlling and tuning the parameters on the servers, routers, and other network elements that implement the QoS strategy. The goal for implementing QoS must be a single, ubiquitous strategy that considers the topology and each of the systems within the network E2E. Both a strategy designed for your particular environment and a highly available transport network are necessary to deliver reliable VoIP, but neither is sufficient by itself.

To aid network managers, we provide some observations below regarding QoS implementation. To ensure success, IT professionals will want to consider all aspects of the project -- planning, design, implementation, and operation (known as PDIO) -- and seek out expertise where needed.

IMPLEMENT QoS
Begin With Packet Classification
Even modestly sized networks can have multiple network devices, such as Ethernet switches, Frame Relay switches, routers, ATM switches, and Ethernet hubs (which should be avoided in VoIP implementations) between the client and the server. Network managers may be tempted to just �throw more bandwidth� at their problems, but increasing bandwidth won�t guarantee that traffic will always flow smoothly.

A more proactive tactic to implementing QoS is necessary to reliably deliver the level of service your VoIP clients will expect. A simple IP packet prioritization scheme is the place to start. This requires, of course, enterprise switches that include packet marking -- capabilities that may not be available on low end, less expensive switches. Frame headers can be marked with a Class of Service (CoS) at layer 2 using IEEE 802.1q and 802.1p specifications. At layer 3, packet headers can be marked with a Type of Service (ToS) using the IP Precedence or Differentiated Service Code Points (DSCPs).

In dealing with the vendors that will inevitably come knocking, network managers should bear in mind that each vendor adds enhancements and optimizations that may not be interoperable, and if the QoS strategy you implement is not interoperable across all network elements, then it may not deliver reliable, deterministic VoIP service.

IT managers also need to know that adding classification will introduce delays in routing traffic. This additional function imposes a penalty on the router�s performance in terms of latency and administrative overload. Wire-speed devices are required to deliver the performance characteristics required by voice and video traffic, in terms of delay, jitter, and packet loss.

Simplify With A Policy Server
Companies today are working hard to get more from existing resources. QoS can assist them in achieving the desired performance for business-critical applications while avoiding expensive link upgrades. QoS will certainly ease the burden of managing a network if the enterprise is adopting widespread VoIP or streaming video, and it will also assist with managing data security and ensuring employees� appropriate use of network resources.

There�s no such thing as infinite bandwidth; therefore, network managers must control which applications get bandwidth first. Unused bandwidth is a true shame. If one application is inactive, the device should reallocate the bandwidth for other applications. To achieve reallocation, there must be consistency in how various switches and routers define parameters. Ideally, 10-12 policies should manage the entire network. IT managers should take advantage of the trend toward simple implementation and �push-button QoS� and be certain to select an option that will scale with network needs.

As mentioned earlier, the QoS strategy should entail the PDIO methodology of planning, designing, implementing, and operating a ubiquitous scheme. To begin executing this type of strategy, network managers will want to examine several areas of importance.

Here are a few to consider:

Traffic Discovery: Analyze which applications are using bandwidth, considering both inbound and outbound flows. Keep in mind that the service quality is only as strong as the weakest link. Here is an example illustrating the impact on data traffic going over the Internet. One enterprise of about 250 users, many of them young and Web-savvy, undertook a one-week study of its traffic using an outside vendor. This organization found that non-critical applications, such as Napster, Real, and Winmedia, were using over 25 percent of the available bandwidth and crowding out mission-critical traffic such as hosted Web sites and e-mail. As a result, on the WAN, application response deteriorated at critical work times and network efficiency degraded during peak traffic times. The organization decided to move forward with a QoS strategy.

Prioritization: Determine which applications are business critical and require protection, considering all applications that are competing for network resources. Interdepartmental collaboration is required to determine configurations that provide the proper resources to the proper applications without harming others. The cooperation of both voice and data teams in concert with end user requirements will be required to achieve the ubiquitous strategy that is so essential to successful management of QoS. The organization mentioned above determined a plan to throttle back on Napster, Real and Winmedia to create more space for hosted Web sites and e-mail. It also reminded employees that the �entertainment� use of the network needed to be curtailed in favor of the business critical applications.

Analysis: Understand the characteristic of the applications to be protected with an understanding of whether they are sensitive to latency or packet loss or are �aggressive� bandwidth hogs.

Class Creation: Create a policy to mark each class of traffic using differentiated services control point of IP precedence as it comes into the router on the ingress interface.

Progression: Work from the network element closest to the traffic toward the core, applying markings on the ingress interface of the router.

Policy: Use the markings to build a policy and classify traffic on the rest of the network segments, keeping it simple by using no more than four classes.

Application: Decide if the policy needs to be applied in one or both directions -- applying in both directions is recommended.

Monitor Effects: Revisit the issues to ensure that the desired effects were achieved. The organization described earlier found that it improved performance of hosted Web sites (over three times faster), e-mail (not peaking as much, improved efficiency) and outbound FTP performance. Overall, the results showed more predictable response times for Web sites and improved overall network efficiency.

Include The Carrier For E2E
Network managers should keep in mind that controlling the outbound traffic from an enterprise solves half the problem. Prioritization back into the network is also needed. Traffic prioritization across public networks is also needed to ensure true E2E quality of service -- which should be every enterprise network manager�s goal.

Carriers and ISPs are evolving toward providing corporate customers with better service quality, at a premium. Although upgrade plans may have slowed after the disappearance of so many CLECs, some carriers and ISPs are investing to make QoS available as a service enterprise IT managers can buy. Given trends such as application service providers (ASPs) and e-commerce, enterprises are conducting more business over the public network, and thus becoming more closely tied to its performance. Enterprises are looking for high reliability, like the voice dial tone they are accustomed to with voice, from ISPs. Service providers offering quality assurances for E2E traffic will gain more business for the enterprise going forward.
With the advent of VoIP, many service providers that traditionally offer data-only services (such as frame relay) can now compete with local exchange carriers in offering VoIP-enabled enterprise customers access to the PSTN. Convergence in the service provider and enterprise spaces has created a market for integrated services that compete with traditional voice-only and data-only offerings.

New offerings based on the multiprotocol label switching standard (MPLS) are emerging that will recognize, label, and prioritize IP traffic. MPLS enables network managers to control the packets� journey to ensure they reach their destinations quickly and successfully. The MPLS Forum recently estimated that as many as 50 service providers either use MPLS in their core networks or are offering multiple classes of bandwidth service (e.g., silver, gold, platinum) to customers. Eventually, IT managers will purchase these services from their carriers or ISPs.

Choose A Solid Team
IT professionals should consider a solid planning and design partner during the QoS process. Experience shows that successful implementations begin there. In-house staff may need the support of consultants and partners who have expertise in VoIP and QoS and won�t be bogged down with technical details. Network managers should look to consultancies or certified resellers of VoIP whose staff has achieved various certifications. Regional certified resellers and national consultancies would be qualified partners. IT managers who understand QoS will be well positioned to manage an outsourced project. Competent people designing and tending to QoS will make the difference in the success and quality of the converged network.

Network managers who employ the PDIO strategy and understand the effort to deploy QoS will be well positioned to work with partners and vendors, helping contribute to a healthy bottom line for the enterprise.

Henry Dewing is director of enterprise market development at Intel Corporation. Stan Potter is business development manager with Cisco Systems, Inc. TIA is a leading trade association serving the communications and information technology industry, with proven strengths in market development, trade shows, domestic and international advocacy, standards development and enabling e-business. Through its worldwide activities, the association facilitates business development opportunities and a competitive market environment. Visit them online at www.tiaonline.org.

[ Return To The February 2002 Table Of Contents ]