1 IP Telephony Contact Centers Mobility Services WHITE PAPER Comparing MPLS and Internet Links for Delivering VoIP Services September 2005
2 avaya.com Table of Contents Overview... 1 Historical Perspective... 1 Approach... 2 Results... 3 Conclusion... 5 Learn More... 5
3 avaya.com 1 Overview Multi-protocol Label Switching 1 (MPLS) has often been presented as a technology that will help service providers deliver better performance and availability, including specific benefits for demanding applications such as voice and video. In recent years, many service providers have implemented MPLS technology in their core networks. Now, with MPLS backbones in place, these carriers are offering MPLS-based services to enterprise customers in support of voice over IP (VoIP) and video applications. The carriers are also aggressively advertising the benefits of MPLS, claiming large improvements in performance and availability when compared with using the public Internet. In view of all this activity, in the summer of 2004 Avaya undertook a project to measure the end-to-end application quality of these new MPLS services. With the help of one of its enterprise customers in the financial services market, Avaya measured the ability of MPLS to support VoIP. The results presented in this paper show that the availability and call quality of VoIP over MPLS is comparable to that of VoIP over the public Internet. Historical Perspective MPLS technology is the natural successor to IP over Asynchronous Transfer Mode (ATM) technology, whose success in the 1990s proved two points very clearly: 1. IP had become ubiquitous; the vast majority of customers demanded only IP services, and service providers were happy to oblige them. 2. Managing traffic across a carrier s network had become a challenge, especially considering the exponential growth in traffic experienced during the late 1990s. The flexibility of label switching provided by ATM helped significantly for provisioning and traffic engineering, while cell-based forwarding allowed for faster processing and larger bandwidth in the core. In addition to these benefits, however, IP over ATM presented numerous disadvantages. While small fixed cells helped speed up forwarding, they imposed at least 20% overhead. Also, as traffic load increased to the gigabit range, the overhead of fragmentation and re-assembly became a limiting factor, since it had become more efficient to forward larger IP packets. Another disadvantage of implementing IP over ATM was the burden of managing the complex interface between IP and ATM. Consequently, the Internet Engineering Task Force (IETF) took note of these challenges and formed the MPLS working group to standardize label switching for IP. Unlike ATM, MPLS was required to work independently of the underlying Layer 2 technology, hence its name. MPLS was also required to be compatible with the Integrated Services framework, and be able to utilize the RSVP signaling protocol. In short, MPLS was designed to give carriers the traffic engineering capabilities they expected along with the quality of service and fast reroute mechanisms intended to support more demanding applications like VoIP. Working with a number of its enterprise customers, Avaya had previously measured the availability and performance of Internet links with shared private WAN links, such as Frame Relay and ATM, and had found them to be comparable. In this white paper, Avaya presents similar data for MPLS services. The goal is to determine whether the adoption of MPLS has improved the performance and availability of VoIP over the shared private WAN infrastructure.
4 COMMUNICATIONS AT THE HEART OF BUSINESS 2 Approach Avaya worked with one of its enterprise customers to measure the availability of MPLS services in the support of voice communication. The customer uses Avaya Converged Network Analyzer (CNA) to optimize voice communication across the WAN that links its headquarters and numerous branch offices. CNA Server Internet Voice gateway Branch Site 2 CNA Server Voice gateway Branch Site 1 MPLS WAN Voice gateway CNA Server Branch Site 3 Figure 1: Test Environment As shown in Figure 1, Avaya CNA was deployed at three of the customer s branch offices. These three offices are linked to parallel network fabrics by MPLS and Internet links at T1 speeds. Through its monitoring and assessing mechanisms, Avaya CNA evaluated the quality of voice communication between Branch Site 1 and Branch Site 2 and between Branch Site 1 and Branch Site 3 for a period of three weeks in June, The assessment was done using Avaya CNA configured to use its Voice Application Model. Availability Application Outage Application Performance Application Delay/Loss Protocol Model Raw Latency/Loss/Jitter Figure 2: Application Model Structure Avaya CNA Application Models follow a five-stage methodology, shown in Figure 2. From the bottom, the five steps include: 1. The measurement of low-level network quantities such as latency, loss, and jitter for each available path between locations.
5 avaya.com 3 2. The computation of transport delay from the raw scores, or the impact on applications in general, not taking specific application sensitivities into consideration. 3. The computation of application delay, specifically the mouth-to-ear delay that affects VoIP. Note that high jitter would result in high application delay; similarly, sustained loss results in lost speech and unintelligibility, leading to increased application delay. 4. The determination of an application quality metric, using a ranking from zero to five stars, similar to movie ratings. For VoIP, this step effectively equates to a Mean Opinion Score (MOS) calculation for the network: the larger the jitter or the greater the sustained loss, the lower the MOS rating. 5. Finally, the time periods where the voice quality is determined to be unacceptable are logged as bad minutes for that path. Using this methodology, Avaya kept a cumulative count of all bad minutes observed on the various MPLS and Internet path choices between the sites in question. Avaya then computed for each choice the ratio of the cumulative bad minutes to the total minutes in the assessment period, resulting in availability values for the various choices. Results The table below shows the availability for all path choices from Branch Site 1 to Branch Site 2 and Branch Site 3. The table also shows the number of bad minutes for each measured path, normalized to 30 days. Number of Bad Minutes in 30 Days Bad Minutes Availability (%) Configuration Branch Site 2 Branch Site 3 Branch Site 2 Branch Site 3 MPLS Best Effort MPLS Premium Internet The number of bad minutes from Branch Site 1 to the other two locations is also plotted in Figure 3.
6 COMMUNICATIONS AT THE HEART OF BUSINESS 4 Figure 3: Bad Minute Comparison Examining these results, the following observations can be made: 1. MPLS Premium service is not always better than MPLS Best Effort service. For the locations Avaya instrumented, MPLS Premium and MPLS Best Effort services provided very similar availability, specifically 99.2% from Branch Site 1 to Branch Site 2, and 99.8% from Branch Site 1 to Branch Site 3. In terms of bad minutes per month, MPLS Premium service and Best Effort services to Branch Site 2 incurred a comparable number of bad minutes, at 321 and 320 bad minutes respectively. Similarly, the number of bad minutes for both services to Branch Site 3 was very close, with 69 and 65 bad minutes, respectively. 2. MPLS Premium service is not always better than default Internet. The availability of the MPLS Premium service from Branch Site 1 to Branch Site 3 is a respectable 99.8%, significantly better than the availability of Internet service between these two locations (measured at 98.6%). However, the results show that the availability of the MPLS Premium service from Branch Site 1 to Branch Site 2 is significantly worse than Internet service between these locations (99.2% versus 99.8%). In terms of bad minutes, the Internet path to Branch Site 3 incurred 596 bad minutes, significantly more than the 65 bad minutes incurred by the MPLS Premium service to the same location. However, the Internet path to Branch Site 2 incurred only 90 bad minutes, significantly less than the 321 bad minutes incurred by the MPLS service to the same location. It is clear that both MPLS and Internet paths can vary significantly in terms of availability; MPLS does not provide consistently better service levels for VoIP traffic. 3. SLA guarantees for MPLS service are of limited value. These results indicate that the availability the MPLS Premium service provides is in the two nines or 99% range. This level of availability is clearly inadequate for voice communications. Yet, the SLA for the provider of the MPLS Premium services being used claims a five nines or % availability. Despite the low availability scores achieved by the MPLS Premium service during the assessment period, the provider did not violate its SLA with its customer. This discrepancy is clearly explained in the details of the guarantees provided by the SLA.
7 avaya.com 5 Here is an excerpt: Daily Jitter is measured between PE s within the network. Jitter is calculated by sending a continuous packet stream of data between a sending server in each of the hubs to a receiving server in each hub. Any difference of time between consecutive packets is subtracted out and is used as the calculation for jitter between each of the consecutive packets of data. The mean average of jitter of all packets within a one second interval will be calculated and stored. The standard Deviation plus the average of a rolling five minute sample of the stored one second values will be used as the measure for jitter. In the event that the MPLS Premium service provider misses its jitter target in any given calendar month, Customer will be eligible to receive a credit equal to one day of its monthly recurring Video and Voice Grade IP service fee(s) for the month in which the target is not met. In order to be eligible for the packet delivery credit, Customer must notify the MPLS Premium service provider within five (5) business days. The following observations should be noted: This SLA is measured between hubs within the MPLS service provider s network and thus doesn t take into account the full end-to-end path. The enterprise, on the other hand, doesn t send traffic just within the provider s core network and cares instead about the quality of calls from end to end. The SLA methodology uses time frames of seconds and minutes, far too coarse for providing guarantees that are meaningful for voice communication. Conclusion Avaya measured the ability of MPLS services to support business quality VoIP communications. Avaya found that the performance and availability of both the MPLS Best Effort and MPLS Premium services are not better than that of default Internet connections. In addition, Avaya found that: MPLS Premium service is not always better than MPLS Best Effort service. The SLA guarantees for MPLS Premium service capture neither the stringent requirements of voice communication nor the business needs of the enterprise. These results suggest that an MPLS service is not the panacea for VoIP. MPLS service is in fact essentially comparable to Internet service. Both provide good base connectivity, but by themselves neither can deliver the quality and availability required for business-quality voice communication. To discover how Avaya can reduce the impact of bad minutes on VoIP communication over both MPLS and Internet links, please refer to the Avaya white paper titled Performance of Virtualized MPLS Internet Infrastructure in Delivering VoIP Services. Learn More For more information on how Avaya can take your enterprise from where it is to where it needs to be, contact your Avaya Client Executive or Authorized Avaya BusinessPartner, or visit us at 1 Avaya, a leading global provider of business communications software, systems and services, has acquired RouteScience, a developer of Multi-protocol Label Switching (MPLS) for enterprises and service providers.
8 About Avaya Avaya enables businesses to achieve superior results by designing, building and managing their communications infrastructure and solutions. For over one million businesses worldwide, including more than 90 percent of the FORTUNE 500, Avaya s embedded solutions help businesses enhance value, improve productivity and create competitive advantage by allowing people to be more productive and create more intelligent processes that satisfy customers. For businesses large and small, Avaya is a world leader in secure, reliable IP telephony systems, communications applications and full life-cycle services. Driving the convergence of embedded voice and data communications with business applications, Avaya is distinguished by its combination of comprehensive, world-class products and services. Avaya helps customers across the globe leverage existing and new networks to achieve superior business results. COMMUNICATIONS AT THE HEART OF BUSINESS avaya.com 2005 Avaya Inc. All Rights Reserved. Avaya and the Avaya Logo are trademarks of Avaya Inc. and may be registered in certain jurisdictions. All trademarks identified by the, SM or TM are registered trademarks, service marks or trademarks, respectively, of Avaya Inc., with the exception of FORTUNE 500 which is a registered trademark of Time Inc. All other trademarks are the property of their respective owners. Printed in the U.S.A. 10/05 LB2773