July 2002

Capacity Assessment Is The Key To A Successful Network

BY PHILIP JOUNG

With the rapid growth in data networking over the past decade, TCP/IP has emerged as the preeminent protocol running throughout much of our global network. Myriad applications and protocols run on TCP/IP-based networks, making its reliability and performance an important issue. Packet-based voice solutions have recently surfaced to take advantage of the efficiencies expected when using TCP/IP, the notable example being VoIP. VoIP aims to make voice communications not only less expensive, but to bring increased features to the end user, more manageability to the enterprises that use it, and high profitability for companies that succeed in delivering robust VoIP solutions.

Now, consider your current POTS phone line � it has become so reliable that barring extreme circumstances, everyone expects to get a successful dial tone and very high voice quality during a call. Current metrics put the uptime of POTS at 99.999 percent, making it one of the most reliable public utilities we use. The dedicated circuit used during a call also means a relatively high degree of security. Given this high reliability, quality, and security, VoIP � as a competing technology � has much to live up to.

Of all the protocols and applications that run on TCP/IP networks, VoIP is one of the most sensitive to network issues. Issues that are at best annoying with many other protocols and applications can often render VoIP unusable. E-mail and the protocols it uses, for example, can typically tolerate latencies in transmission, and are not significantly affected by jitter, IP fragmentation, or small amounts of packet loss. However, introduce small amounts of jitter, latency, or packet loss, and voice quality and usability with VoIP quickly degrades to an unacceptable level.

Given the complexity of most enterprise networks, adding VoIP infrastructure and functionality is a complex undertaking. Bandwidth must be allocated, quality of service rules devised and configured, and security features implemented. The only way to know for sure that VoIP service will be acceptable is through rigorous capacity assessment.

CAPACITY ASSESSMENT
What is capacity assessment? In general, it can be defined as testing the network under loads that replicate real-world conditions. We all understand that live networks (especially the Internet) will have a number of factors that complicate matters � large numbers of users performing large numbers of different tasks; different devices that react to and often alter the traffic; link latencies; packet loss; jitter; and the like. Capacity assessment incorporates these different factors during testing to ensure that the results of testing closely resemble what is or will be experienced in a live network. It combines the pertinent factors in user behavior and network behavior to assess the reliability, availability, scalability, and performance of the system under test. Capacity assessment with realism is the best way to achieve confidence in the final results.

Unless a network is dedicated to VoIP, many other things will be happening on the network. In most VoIP deployments, a shared network will be a given, as one of the major advantages for VoIP centers on saving money, and a dedicated network will typically not be cost effective. VoIP deployments should expect a network handling traffic with many different purposes, often with differing value and criticality to the enterprise.

Because of this diversity in network traffic, realism in the capacity assessment of critical networks will incorporate not only the application that is being tested (e.g., VoIP), but also all the other traffic that is running concurrently with that application (Web browsing, FTP, e-mail, streaming media, etc). These different protocols and applications will cause different stresses to the network, helping to locate possible issues and bottlenecks during testing.

USER BEHAVIOR
Most large networks will have users that fall into a certain pattern of usage. Some users will mostly browse the Web, others mainly download large files, others primarily talking on the phone via VoIP, and still others doing various other tasks. Quantifying these different user populations (and at times, predicting how their behaviors might evolve in the future) in order to incorporate their behaviors into testing will make capacity assessment results much more predictable, solid, and reliable.

Because of the variability of network traffic, I�ve included a list of user behaviors below that might be encountered in real networks. While they do not always directly apply to VoIP, they can cause effects that matter for other applications, indirectly affecting VoIP.

Downloads: Downloads of many small files simultaneously, large files, or streaming media can all have detrimental effects on network performance.

Bursty traffic: Users typically use networks in spurts, with periods in which the traffic jumps significantly from its normal amount. Simulating this user behavior can help ensure that the network handles these challenges effectively. These peaks often happen during critical periods, such as the crucial holiday buying season for an e-commerce Web site, the busy call volumes during the morning, or the streaming broadcast of a major event.

Frustration thresholds/user aborts: People browsing the Web often become frustrated when pages do not load after a few seconds and either try to reload or hit the stop key. This can cause further load on the system, since it continues to work on the original request even though the user already has moved on.

SSL usage: Use of secure sockets layer to encrypt data increases the security of network transactions, but carries a big price in terms of processing and network bandwidth.

Use of stateful applications: This can generally be defined as applications that keep track of the user throughout the session. Examples include Web sites that use cookies to maintain a user�s session, FTP logins, and SSL session IDs.

Browser versions: Different Web browsers can cause different types of load on a system. Some Web sites tailor content to specific browsers, which also uses processing cycles.

NETWORK BEHAVIOR
An important part of overall network performance will depend on what is happening to the traffic as it flows across the network. When certain conditions occur, they often create significant effects on system performance. Below are some common network issues:

Packet loss: Packet loss often happens from network/ system congestion, or transmission errors. Studies have shown that FTP performance drops by 50 percent with only three percent packet loss. Small amounts of packet loss cause clicks and pops in VoIP, but high levels can make the conversation unintelligible, or completely fail.

IP fragmentation: Devices often have a maximum amount of data they can accommodate at one time. When a device receives data that exceeds its maximum amount, it can fragment it into multiple chunks that need to be reassembled at the final destination. This reassembly takes processing, especially when the fragmented data arrives out of order.

Jitter: Applications that deal with real-time data, such as VoIP and streaming media can be especially sensitive to jitter. Jitter occurs when network data arrives at different rates. �Timing is everything!�

Latencies: High network latencies spell doom for real-time applications like VoIP. Indeed, a VoIP conversation becomes difficult to manage when network latency increases much above 250 milliseconds � people often end up speaking on top of one another. Network systems also exhibit increased loads with higher latencies because they must deal with increased network connections and buffering.

Data corruption: Corruption is usually a rare event in most modern networks. However, when it happens, the effects can be similar to packet loss.

QUALITY OF SERVICE
Because of the sensitivity of VoIP to network issues, having control of the user traffic being handled by the network is valuable. Mechanisms abound for helping to control this traffic, with the majority falling into the QoS (quality of service) umbrella.

QoS allows the network administrator to set priority levels for traffic. Therefore, VoIP traffic can be given a higher priority than say, instant messaging or Web surfing traffic. Capacity assessment will not only help ensure that the network properly applies and follows the rules set by the administrator, but also provide understanding into the effect of the rules on user traffic, especially for traffic that has been given lower priorities. Capacity assessment also provides information about how the QoS rules and the network will degrade once it approaches its capacity. This can be important in helping to predict failures before they occur, giving the administrator a chance to make changes to the network in advance of any detectable failures.

PAY A LITTLE NOW, OR PAY A LOT LATER
Given the complexity of the networks running at most enterprises and the business critical nature of the applications running on those networks, real-world capacity assessment should be part of any system prototyping or deployment. Although it involves a small amount of extra effort earlier on, the gains in system stability, performance, availability, and ultimately improved customer satisfaction and time-to-market more than pay for the up-front effort. Factor in the peace of mind gained from capacity assessment, and it becomes apparent that the justifications for not conducting capacity assessments are poor.

Capacity assessment can take place at all stages of a system project: Product selection: While scoping out different vendors� products, capacity assessment can prove, using your own parameters and network traffic mix, how the product functions and performs.

System design: Capacity assessment can help one choose between different design ideas, demonstrate scalability, compare performance, and can even scrutinize the tolerance of a design to certain user and network behaviors.

System deployment: Capacity assessment can help ensure that expected performance levels are being maintained and help adjust parameters.

Post-deployment: Realistic capacity assessment will validate the final performance of an entire system, providing a baseline from which future upgrades can be considered.

VoIP holds a lot of promise, and many investigations have commenced towards determining this technology�s eventual fit into the enterprise. Given the quality of the current POTS system in use, VoIP deployments must be sure to come close to matching that quality, or risk user complaints or even complete removal of the poorly functioning VoIP system. Real-world capacity assessment offers that extra insurance that the final VoIP deployment functions properly and scales effectively. The fact that the rest of the network will benefit from this assessment is simply an extra bonus that can help justify the effort.

Philip Joung is director of technical marketing at Caw Networks. Caw Networks� products are robust, customer-ready, highly scalable appliances that deliver unmatched performance and ease of deployment, use, maintenance, and administration. For more information visit the company online at www.caw.com.

[ Return To The July 2002 Table Of Contents ]