Since the dawn of VoIP, the chief concern of users has been Quality of Service (QoS), or the quality of a voice or video transmission. But what is this “quality”?
From the perspective of a typical user, a phone call (or video conference) either sounds good or it doesn’t.
Network engineers, however, have a different definition of QoS. To them, technically, QoS is an analysis of delays in IP packet arrivals (anything above 150 milliseconds is considered unacceptable), packet loss (you could consider these packets as having an infinite delay in arrival), and jitter (variations in packet delay).
Still another view of QoS comes from IT department managers who hold a more business-like perspective: QoS tends to be linked more with the telecom provider’s SLA (service level agreement) and so the network must meet a certain level or standard of availability, such as no more than an hour or two of downtime per year, that users and applications will be allocated the proper bandwidth, and that the network’s performance will meet some specific requirements favored by the business.
While many technically-minded people think of QoS as dealing with impediments in the flow of packets through a network, QoS in fact encompasses the workings of all seven layers of the OSI (Open Systems Interconnection) Model. Operating systems (most operating systems don’t handle realtime processes very well), competing data streams, communications protocols, scheduling and traffic management issues, all come into play. Indeed, QoS can involve any process or network elements stretching from one endpoint to the other, since the quality of a conversation is only as good as the weakest link in the transmission “chain” between endpoints. Everything from a burned-out router to a buggy softphone application running on a laptop with a defective network interface card can affect QoS, regardless of the pristine nature of the rest of the network. Thus, the ultimate judgment of QoS is the subjective rating of the actual user of IP Communications.
The most popular way to measure telephony voice quality is the MOS or Mean Opinion Score, a number between 1 and 5 used in an attempt to quantitatively express the subjective quality of speech in communications systems, particularly digital networks carrying VoIP traffic. Anything above a 4.0 is considered toll grade.
It’s quite easy to achieve a MOS of 4.4 on a LAN, even using big (1,500-byte) clunky Ethernet packets, since the bandwidth is high and the enterprise owns the whole LAN, and can guarantee the quality of network equipment and cabling end-to-end. Once voice or video packets are sent across the public WAN, however, the MOS can easily drop to 3.0, since no individual corporation owns network paths from end to end.
Bandwidth vs. Prioritization
The two principal competing methodologies for keeping QoS at acceptable levels has been bandwidth overprovisioning versus traffic engineering or prioritization. Provided that all of your network elements are in tip-top shape, “limitless” (i.e. optical) bandwidth yields superb QoS for all traffic flows. These utopian conditions are almost always impossible to achieve, which means that network operators must resort to the second option, traffic engineering.
Traffic engineering itself falls into two categories:
Integrated Services (resource reservation), where network resources are apportioned according to an application’s QoS request, and subject to bandwidth management policy. The ReSerVation Protocol (RSVP) provides the signaling to enable network resource reservation.
Differentiated Services or “DiffServ” (prioritization) where network traffic is classified and network resources are apportioned network resources according to bandwidth management policy criteria. The delivery of packets is “prioritized” in the network, depending on the application that originated the packet, thus yielding the concept of Class of Service (CoS). As is the case with many systems (Cisco, for example), applications are associated with three grades of service. Voice, video and other real-time applications get the highest prioritization and thus the most preferential treatment by the network elements. This is the most popular way of maintaining high QoS, if only because some 800- pound gorillas (Cisco, Juniper and Avici) favor it.
The principal technique that tends to be used in a differential services environment is MPLS (Multi Protocol Labeling Switching) which takes aggregates of packet streams and determines the priority of packet delivery based on labels inside (encapsulated by) the packet headers. MPLS does its job inside of routers and can handle multiple, non-IP protocols, so it works with ATM, IPX, PPP or Frame-Relay. It can even run directly over the OSI data-link layer.
Note that a kind of resource reservation occurs with Diffserv and MPLS, but its “guarantee” of quality is more statistical in nature. Instead of monitoring the call status and updating the reservation of resources (as in the case of RSVP), in DiffServ the call status is monitored and the necessary bandwidth is calculated. RSVP comes closer to continually achieving toll-grade quality than DiffServ techniques, but the cost of resource reservation with RSVP, which involves many communications with UDP messages to and from routers, is too high.
Similar QoS challenges occur in wireless networks, perhaps more so, since even 3G wireless connections are of lower bandwidth and are prone to interference but nevertheless must support multimedia applications with pretty high QoS requirements. The IMS (IP Multimedia Subsystem) framework specifies that end-to-end QoS support requires signaling, traffic regulation and resource allocation capabilities. QoS signaling can provision and enforce QoS parameters between endpoints and operates in the OSI application layer, network layer and link layer. Session-specific QoS parameters can be exchanged via SDP or SIP header fields.
Quiz That QoS
Since the whole future of IP Communications hinges on voice and video quality, a vast sub-industry of companies that deal with testing and monitoring networks for QoS has arisen. Psytechnics (http://www.psytechnics.com) for example, became a major force in this business after it was spun-off from British Telecom with backing from 3i.
Psytechnics’ (news - alert) products furnish performance information regarding the design, optimization, and monitoring of PSTN, mobile and VoIP telecommunications networks. The company prefers to use the term QoE, or “Quality of Experience” to describe their bailiwick, rather than QoS. They feel that QoS is merely a network-centric, technical analysis of bits and bytes on a per application or link basis. The user experience is really of primary concern, so they’ve replaced QoS with a more user-centric paradigm, QoE, that scrutinizes individual user experiences, with in-depth application intelligence that captures experience information on a per-user and per-session basis, said to be invisible to more rudimentary QoS tools.
Recently, Psytechnics performed a voice quality performance evaluation study of a pre-release (beta) version of Microsoft Office Communications Server 2007 and the Microsoft Office Communicator 2007 desktop VoIP solution. The study also included a comparison of the Microsoft solution to some prototype USB handsets as well as Cisco’s CallManager 5.0 and 7961 IP phones. Both the PC VoIP solution and the IP phone were evaluated by both real endusers (the subjective tests) and by Psytechnics’ QoE software (the objective tests). Psytechnics’ evaluation reveals that “Overall, the one-way listening speech quality provided by the combination of Microsoft’s client and a USB handset was consistently better than that provided by Cisco’s IP phones and CallManager, whether using G.711 or G.729.” Psytechnics has demonstrated that the quality of Microsoft’s offering is high enough so that companies should feel free to integrate voice communications with PCs, which in turn suggests that they could eliminate the purchase of expensive IP phones almost entirely if they so desire.
As Mike Hollier, CTO at Psytechnics, says, “This evaluation emphasizes the positive transformation that software-based VoIP solutions will have on unified communications and telephony in the future. The familiar PC can now outperform the IP phone.”
Ironically, for its new Response Point small business solution, Microsoft has begun partnering with handset vendors to make compatible endpoints; e.g., the D-Link DVX- 2000, Quanta Syspine and Uniden Evolo.
Another company in this space, Telchemy (news - alert) (http://www.telchemy.com) is known for its VQmon and SQmon families of service quality monitoring and analysis software products, technology that enables both service providers and enterprises to monitor and manage the performance of VoIP, IPTV, IP videoconferencing, high definition Telepresence, 3G / 4G mobile and other converged real-time services. Telchemy’s products provide real-time visibility of service quality, accurate estimates of user experience, QoS, IPTV QoE, VoIP QoE (MOS scores and R factors), and detailed analysis of the root cause of quality degradation.
VQmon integrates into both network infrastructure and test equipment. Once there it can provide perceptual quality scores for every call, reporting metrics using RTCP XR (RFC3611), SIP RTCP Summary Reports and other key protocols. Telchemy’s also relies on OEM probes and distributed active monitoring applications.
Telchemy’s latest technical achievement is DVQattest/EN, an active test tool for VoIP, IP videoconferencing and highdefinition telepresence service assurance. Available as software for license to OEMs and network equipment vendors, DVQattest/EN uses VQmon technology to provide network assessment, pre-deployment testing, SLA monitoring and advanced network troubleshooting for enterprise networks. DVQattest/EN sessions can generate 200 concurrent VoIP streams, 20 concurrent high definition 1080p simulated IP video streams and a range of network diagnostic tests, with an interactive application for configuration and reporting. Despite its high level of sophistication, the DVQattest agent is small enough for direct integration into network equipment.
The Quality of Packets is Not Strained
All in all, quality of service is really an umbrella term for a collection of defined user experiences and technologies which allow network-aware applications to request and receive predictable service levels in terms of data throughput capacity (bandwidth), latency variations (jitter) and propagation latency from QoS-enabled IP networks which can respond to requests from critical applications for either resource allocations or differentiated levels of service among shared resources.
Ultimately, however, the only thing that matters is whether or not you like the way a voice call sounds or a video call looks.
Richard Grigonis is Executive Editor of TMC's IP Communications Group.
VoIP Quality of Service Myths
The folks at Psytechnics (http://www.psytechnics.com), the masters of IP Communications testing, are fond of pointing out and clarifying some of the cherished myths about IP-based voice quality still believed by may IT departments.
Myth: IP QoS will solve all of my VoIP issues.
Reality: Increasingly the challenges in major deployments are not at the network level in the IP infrastructure, but are at the application (voice level), in interconnecting with the PSTN or legacy voice networks, which can introduce noise and distortion or intermittent faults at gateways.
Myth: A good IP SLA tool will measure QoS for voice.
Reality: IP SLA tools measure network level QoS performance very effectively, but they do not tell you anything about what an individual session (a call) on the network experiences. Additionally, a user’s experience of the call is affected by much more than the IP performance. This is why Quality of Experience tools are increasingly being used in VoIP roll outs. Especially tools that work across both the traditional voice world and the VoIP world.
Myth: VoIP quality is lower that the PSTN or traditional voice networks
Reality:New forms of encoding voice (CODECS) are starting to be used which actually produce voice calls that give a higher perceived quality than traditional phone calls. This can be verified with perceptual measurement tools, such as the ITU approved PESQ, used by the telephone companies to measure their voice quality.
Myth: This CODEC is better than that CODEC.
Reality: Different CODECs (compression/decompression methods) do result in different quality of voice, but also different IP handsets and gateways respond in very different ways to packet loss and jitter in the network. This is why handsets must be chosen carefully, and any IP measurement tools must take account of the brand of handset or they will be inaccurate.
Myth: Voice quality in an IP network is based on Jitter and latency in the network.
Reality: Real network deployments show that there is often little correlation between the MOS (Mean Opinion Score) that IP systems and IP management tools report and the actual user experience. This is why standardsbased quality assessment tools are useful when a deployment is underway. You can achieve consistent and accurate measures that will help you pin-point faults, rather than over-optimizing the network or wasting money on excess bandwidth.
|
SUBSCRIPTIONS
TMCnet LOGIN
WEBINARS
