TMCnet News
Making VoIP Perform as AdvertisedMaking VoIP Perform as Advertised By Manickam Sridhar CTO Converged Access, Inc. The promise of major cost savings has led many enterprises to consider moving voice traffic to the enterprise IP wide area network (WAN). However, a successful Voice over IP (VoIP) implementation must deliver on both cost savings and user satisfaction with the quality of VoIP call service. Achieving well-defined toll quality metrics requires the application of a number of QoS techniques acting in concert to manage delay, jitter and packet loss across the IP network. This paper describes: • The integrated QoS techniques required to achieve toll quality VoIP The Need for QoS Enterprises of all types are excited about the potential benefits of converged IP voice, data and video networks. New business applications such as distributed call centers promise to increase customer satisfaction, while remote teleconferencing will maximize employee productivity, save money and simplify management. However, the nature of an IP network, which is characterized by bursty traffic and "best-effort" delivery, presents challenges for a real-time application like voice. Best-effort delivery may be acceptable for Web traffic, but voice requires guaranteed delivery in order to achieve acceptable standards for business communications. To date, some companies are limiting their VoIP implementation to intra-company calls, using gateways that convert traditional voice into IP packets and then route over the WAN infrastructure (see Figure 1). Intra-company calls are identified by a prefix or code that the PBX uses to forward calls to a VolP gateway. The calling gateway makes a VoIP call to the receiving gateway over the WAN. The call is then forwarded to the receiving PBX. Cost savings are accrued through toll-bypass, and are applicable in instances where the toll calls are expensive, or when there are many voice lines e.g. in a branch office network. This approach also allows the enterprise to preserve investments in legacy PBXs and handsets. Other companies are taking convergence to the next level, replacing traditional hand-sets and analog PBXs with IP phones and IP-PBXs. The migration is moving slowly, however, in large part due to issues with VoIP QoS. Why is guaranteeing QoS for VolP an issue in converged IP networks? First, when voice and data are mixed on an uncontrolled network, the result is increased jitter, packet loss, and excessive delay. Callers experience voice distortion, loss of portions of words or sentences, echoes, talker overlap, and dropped calls - all unacceptable for a business-critical application in an enterprise environment. Second, there are different encoding algorithms used to translate circuit-switched voice to packetized voice. WAN bit rate requirements vary with codec type, making it difficult to pinpoint specific voice call bandwidth requirements. Finally, calls are made without knowledge of conditions on the WAN and with no topological reference. The result is calls with no dial tone, no response to a dialed number, or a poor connection where quality degrades over time. These conditions are due to calls being placed on an overloaded network. Without QoS mechanisms to mitigate these problems VoIP deployments are limited or avoided altogether. Companies that have not yet implemented VolP are reluctant to adopt the technology, thus missing the potential to reduce wide area costs. Callers shy away from an existing VolP network due to poor quality. They revert instead to the public-switched telephone network (PSTN), undermining investments already made in VoIP. In addition, companies miss the opportunity to implement advanced applications based on packetized voice if they can't get basic VoIP to work. This further calls the initial decision to invest in VoIP into question. Thus, QoS is the key to making converged IP voice and data networks a practical reality. This white paper first identifies the specific network problems that adversely affect packetized voice. It then reviews the QoS mechanisms, or rather the combination of mechanisms, required to deal with these problems. As this paper shows, it is the completeness of the QoS solution (e.g., the integration of multiple QoS mechanisms working together) that determines whether or not the network delivers the QoS guarantees required to make VolP perform as advertised. What Are the Key Challenges for VolP? Jitter, packet loss, excessive delay, and poor call provisioning can wreak havoc on call quality. To understand how these threats can be handled effectively with QoS, it is first necessary to understand their impact on voice. Jitter Voice packets are generated at periodic time intervals by codecs, which incorporate the encoding algorithms used to compress and packetize voice traffic. The number of bytes in a packet and the time interval between packets are determined by the particular codec that is used. On a converged network, small voice packets will be interleaved with data packets of varying sizes, causing what should be normally orderly packetized voice packets to arrive at disorderly intervals. This variability in packet arrival time, or jitter, will compromise voice quality and can make voice unintelligible. To compensate for this condition, VolP equipment manufacturers provide jitter buffers in gateways or handsets. However, if the variability between packets is more than the jitter buffer can handle, the packet is thrown away and treated as lost. Throwing away packets means throwing away voice content. The result sounds like a bad cell phone call. Managing jitter requires a VoIP QoS mechanism capable of automatically detecting the required interval between packets, and adjusting traffic management parameters in real time to ensure that this interval is maintained. Packet Loss The next condition that degrades voice is packet loss. When voice packets are dropped, there is a disconcerting "clipped" quality to the human voice. Experts agree that packet loss in excess of 2.5-5% is unacceptable for voice traffic. The design goal of enterprise IT organizations deploying VoIP is to keep loss to less than 1% during network congestion. Packet loss occurs during periods of network congestion when queues in the routers begin to over-flow. This forces the routers to drop packets. Because routers are packet-agnostic, they have no way of deciding which packets to drop when the input queue fills up. Voice packets will be dropped as randomly as any data packet. In reality, the router can't be told to queue all voice packets and never drop any of them. First, there is only a finite amount of queuing available. Second, even if there were an infinite queue the voice delay budget would be exceeded. [NOTE: The delay budget is the amount of time between the speaker saying a word to the receiver hearing it. For voice to be perceived as toll-quality delay should be kept to less than 100 milliseconds.] To deal with packet loss, some VolP equipment includes a repairing algorithm which makes up for packet loss by inserting silence packets meant to emulate pauses in human speech. However, silence insertion and other such repairing algorithms do not prevent packet loss. They simply minimize the effect after the fact. Excessive Delay Keeping voice within the delay budget is a difficult task. There is a certain amount of delay inherent in every VolP implementation. The core network itself adds roughly 40 - 50 ms of delay to each voice call. Codecs add a minimum of 20 ms on the transmission side. On the receiving end, the jitter buffer also adds approximately 20 - 30 ms of delay to the VolP call. When you add these delays, you're left with approximately 10 - 20 ms to play with before voice quality falls below toll quality. Now, consider this scenario: if a voice packet gets stuck behind a single large data packet, the delay budget is obliterated. For example, a single 1500-byte e-mail packet takes 23 ms to be transmitted on a 512 Kbps link. If one voice packet gets stuck behind that e-mail packet, the delay budget has been exceeded. More commonly, there would be several big packets queued as part of an e-mail transmission, not just one. It is easy to see how this problem is compounded when there are multiple voice calls plus data packets from other applications contending for the same line. Repairing algorithms constitute damage control, not proactive VoIP QoS. Therefore, the QoS mechanism must prevent the conditions that lead to packet loss in the first place. Another factor that affects overall transit delay is transient congestion. Occasionally, more packets enter a LAN/WAN interface in a given period of time than can be sent out over the WAN. This causes congestion and excessive queuing delay. Without proper control mechanisms, all the queued packets will be transmitted based on respective bandwidth allocations and the excessive delay will continue for the duration of the call. Different Codecs, Different Bandwidth Requirements No amount of control over jitter, packet loss, and delay will matter if there is insufficient bandwidth for each voice call. However, pinpointing voice bandwidth requirements is not straightforward. The amount of bandwidth required per call varies depending on the codec and the overhead associated with it. Therefore, it is essential to deploy a QoS solution capable of monitoring voice bandwidth consumption to provide accurate voice bandwidth allocation for each and every call. The VoIP QoS solution must: • Understand the bandwidth required on a call-by-call basis based on the codec selected. • Understand which other applications are competing for bandwidth. • Be able to set and guarantee bandwidth and priority on a per call basis. • Monitor and report on the mix of traffic flowing over the network to identify the traffic types impacting voice performance and to also take steps to control them. The Challenge of Call Provisioning In VolP deployments, the call controller (including the gatekeeper in case of H.323, or the SIP server in case of SIP) is tasked with functions like call admission control. Because these devices usually sit on the LAN they lack information about real-time conditions on the WAN. Therefore the call controllers do not know whether or not there is capacity on the WAN line for one or more voice calls. If the call controllers allow more calls on the line than the prevailing network conditions can support, all of the calls in progress can be severely affected. Setting hard limits on the number of calls allowed results in inefficient and uneconomical use of the WAN because calls will end up getting denied even when there is more than enough capacity on the WAN. Yet another problem is the lack of topological reference. In an enterprise with several branch offices, where the call controller is deployed at headquarters, the call controller only knows about the total number of calls provisioned in the network. It does not know how many calls are provisioned for each branch office, the link speeds to the different branches, or the number of users in each branch. Blindly admitting calls leads to unpredictable performance since calls to and from one branch may be totally fine at one point in time, while a call to the second branch is unacceptable. This situation may be completely reversed at another instance in time. To prevent excessive delay, the VoIP QoS mechanism must be able to cap queuing delay. It must also be able to discard excess packets - rather than exceed the delay budget - and control data packet sizes to optimize inter-packet delay. Putting it All Together: Converged Access Delivers Toll-Quality VolP The Converged Access solution integrates a comprehensive suite of QoS techniques to enable the delivery of toll-quality VOIP. Based on the company’s advanced Application Control System, the Converged Access family of products deliver toll-quality VoIP for each call admitted to the network. The Converged Traffic Manager (CTM) can be deployed behind an existing router, bringing toll-quality performance to an existing VoIP deployment. Voice packets pass through the CTM before entering the WAN. The Application Control System applies policies that protect VoIP performance while other applications share the network. The Converged Access Point (CAP) provides the same level of service for SMB/SME locations that require an all-in-one package that meets strict ease of use and TCO requirements. Important capabilities included with the Converted Access solution include the ability to: • Monitor the network to determine voice bandwidth requirements • Accurately provision bandwidth on a per call basis • Manage voice and data traffic to maintain the required interval between voice packets • Intelligently manage traffic to keep voice packets within the delay budget • Prevent the conditions that lead to packet loss for voice traffic • Provide the centralized intelligence to provision the necessary per call bandwidth Monitoring One of the most difficult challenges in protecting voice traffic is obtaining precise information on bandwidth consumption for both voice and data traffic. Using wire-speed application- layer classification, the Application Control System provides a real-time window into all traffic traversing the network. User-friendly charts and reports make it easy to see how much bandwidth each user/application is consuming, and how well these users/applications are being served by network resources. Network managers can use this information to set and adjust policies to protect voice traffic while maintaining the proper quality of service for other applications sharing the pipe. The effectiveness of policies in force can be monitored in real time using "at a glance" status screens. The network manager can adjust the policy in seconds and then evaluate the results to see if the new policy is more effective. Class-Based Queuing (CBQ) CBQ is the most sophisticated form of queuing available, providing the flexibility to protect voice from all other types of data traffic. CBQ assigns both priority and bandwidth to ensure the quality of a voice call. Simpler queuing techniques such as priority queuing (PQ) or weighted fair queuing (WFQ) offer either priority or bandwidth control, but not both. CBQ calculates and supplies the different bandwidth and delay budgets for all traffic types carried over both TCP and UDP, making it the ideal tool for ensuring high-quality voice. The VoIP QoS solution must provide intelligence to assess network resource availability to ensure optimal quality for each accepted voice call. CBQ also provides the ability to: • Prioritize and Allocate Bandwidth among Different Types of UDP Traffic Voice traffic is carried using UDP, the same protocol that is used for applications like audio and video streaming. However, where voice consumes relatively little bandwidth and is well behaved, audio and video streaming consume a great deal of bandwidth and can adversely impact data traffic. Therefore, network managers usually seek to protect data traffic from audio and video streaming, and VolP from data traffic. In a network running multiple types of UDP traffic, CBQ allows the network manager to prioritize voice running over UDP while controlling other UDP-based traffic. • Protect Voice and Signaling Traffic It is critical to protect not only the quality of the call itself, but also the call setup (signaling) traffic which is carried by the TCP protocol. TCP and UDP have different characteristics. Signaling and voice traffic have very different bandwidth requirements as well. In addition, signaling traffic must be prioritized over the voice call and all other traffic, as signaling between the call controller and the endpoints is critical to ensure the timely setup and tear down of VolP connections. Voice traffic should be prioritized just below the signaling traffic, but higher than all other traffic types. CBQ provides the ability to accomplish both of these tasks. • Dynamic Optimization of Different Voice Codecs CBQ differentiates between the codecs associated with each VolP conversation, i.e., higher-bit-rate codecs used for toll-quality (high-quality) voice and lower-bit-rate codecs used for sub-toll-quality voice. CBQ can calculate and apply the different bandwidth and delay budgets required for different codecs. TCP Rate Shaping By controlling the size of the TCP Acknowledgement Window, TCP rate shaping can control the rate at which hosts transmit data on to the network. This process smoothes out the traffic flow, and, in conjunction with CBQ, provides precise rate control and fair allocation of bandwidth for all flows. Limiting the arrival rate of the packets at the input queues of the routers minimizes queuing delay, keeps the queues from overflowing thereby avoiding packet loss. The CBQ and TCP Rate Shaping combination is critical to maintaining toll quality voice. Per Session Bandwidth Per session bandwidth is important because voice calls cannot be managed in aggregate, as other data traffic such as Web browsing can. Rather, each voice call must be handled granularly as a single flow, with its own unique characteristics. Otherwise, either bandwidth will be wasted or the voice calls will interfere with each other and get starved. Session bandwidth control guarantees that each voice call will get the rate that is expected. Packet Size Optimization Controlling the packet size of lower-priority traffic that could be queued in front of voice packets is a crucial mechanism for minimizing jitter. There are actually two different ways to control packet size: fragmentation and controlling packet size at the source. The Application Control System packet-size optimization feature controls packet size at the source, a method that has numerous advantages over fragmentation. Fragmentation involves chopping up big IP packets into smaller IP packets. The disadvantage with fragmentation is that the routers downstream have to route dozens of smaller packets instead of one packet, and the receiver has to reassemble all the little packets. In addition, fragmentation needs additional buffering at the routers and hence has problems when a large number of flows are involved. In this case as the number of fragmented packets increase the packets invariably are dropped, resulting in poor voice quality. Controlling packet size at the source (e.g., the e-mail server, the HTTP server, the FTP server) is far more efficient because the source machine never sends packets larger than a specified size. With packet size optimization, the network manager is essentially optimizing the network for voice by specifying that the source devices never put more than a certain amount of information in an IP packet. In this way, the network is always more balanced between voice and data. Maximum Queuing Delay Maximum queuing delay allows network managers to limit the amount of wait time for a voice packet in any queue, regardless of how much traffic is queued up. By providing this additional QoS mechanism, network managers prevent voice packets from queuing up in a long line (due to some transitory condition where the packets are queued up elsewhere for a short time and arrive at the QoS device all at once) and exceeding the delay budget. Because it is preferable, the maximum queue delay will allow the QoS device to drop voice packets at the end of the queue in order to preserve the integrity and continuity of the voice calls. This feature allows for graceful recovery of the call after a transient impairment such as the one mentioned above. This also maintains jitter buffer tolerances on the receiver side. WAN-Aware Call Provisioning The Converged Access solution also complements existing call controllers by bringing an awareness of WAN conditions to the call provisioning process. The Application Control System includes the ability to signal the call controller when set call thresholds are exceeded. The call controller can intelligently block or allow calls to proceed. This ensures that when there is a new call requesting to be cut-through there is always adequate bandwidth on the WAN to guarantee the quality for that call. Admission Control Admission control traditionally is an admit/don't admit capability. With VolP admission control, signaling intelligence must check the number of voice calls being initiated against the current capacity of the network and then signal the call controller to admit the call or not. Without admission control it is possible that when the new call is admitted, existing calls are adversely affected because the current capacity of the network is insufficient to sustain all the calls. This sort of uncontrolled admission has very severe implications since conversations that were perfectly normal before this new call are uniformly wiped out. With admission control the new call will be denied admission on to the WAN thereby protecting the calls already in progress. Centralized management via the Converged Policy Manager™ (CPM) It can be complicated and time-consuming to determine the amount of bandwidth to configure between branches and headquarters. For example in a multi-branch retail chain, there may be an average of two simultaneous calls going to one branch, four to another branch, and six to a third branch. The network manager has to figure out the bandwidth requirements and policies for each branch and also the bandwidth required at headquarters to handle the total volume of call. CPM has the tools and templates to reduce the administrative burden associated with complex configuration. CPM also makes it more practical to distribute VolP policies across an enterprise network by allowing the network manager to create template policies for different branches, and then simply apply the same template to all branches that have the same VolP requirements In addition, CPM provides the ability to change policies on the fly. For example, if the user needs to add more bandwidth to support a conference call between 4PM and 5PM on Tuesday, the network manager can set the policy in CPM and it will automatically implement that policy on the responsible CPM managed devices. Guaranteed VoIP Quality Means Guaranteed Returns The benefits of VoIP have been discussed for a long time. It is tempting to think that the time has arrived. However, for VoIP to work as advertised over an enterprise WAN, IT managers require a combination of critical QoS technologies. Converged Access provides a family of solutions, based on the Application Control System, that deliver toll-quality VoIP performance for businesses, large and small. With this solution, enterprise IT managers can deliver the cost savings and productivity promised with IP convergence, while maintaining the toll-quality performance assurances that will translate into guaranteed returns for the enterprise. CORPORATE HEADQUARTERS 305 Foster Street Littleton, MA 01460 Phone: +1 978-742-1400 Fax: +1 978-742-1493 GENERAL INFORMATION www.convergedaccess.com / [email protected] SALES Toll Free: +1 866-474-8272 (US and Canada) [email protected] Converged Access, Converged Access Point, Converged Traffic Manager, Converged Policy Manager, and the Converged Access logo are trademarks of Converged Access Inc. All other trademarks are property of their respective holders. Printed in the USA. Copyright 2006 Converged Access. All rights reserved. Specifications are subject to change without notice. CA v2.0506 |