Quality of Service Network Design Considerations for Cisco TelePresence Systems
By Tim Szigeti
For Cisco TelePresence conferences to be natural, realistic, and effective, the network infrastructure must service the high-definition audio and video packets sent between TelePresence endpoints with high availability and quality. A major benefit of Cisco TelePresence Systems over competitive offerings is that real-time, high-definition video and audio are transported over a converged IP network rather than a dedicated network (although dedicated networks are also supported). The key enabling technology to achieve such convergence is quality of service (QoS).
From a QoS viewpoint, Cisco TelePresence requires stricter service levels for loss, latency, and jitter than virtually any other application on the network. Additionally, Cisco TelePresence requires detailed bandwidth and burst provisioning. Furthermore, priority servicing of TelePresence must not interfere or degrade VoIP services: both applications require real-time services, yet are radically different at the packet level, as illustrated in Figure 1.
Therefore, to meet these design challenges, extensive attention needs to be given to the QoS designs of the network.
The first step in QoS design is to fully understand the service-level requirements of the applications being provisioned. The service-level requirements for Cisco TelePresence are as follows:
Bandwidth: The bandwidth required by Cisco TelePresence systems will vary according to the number of screen/camera pairs (called “segments”), the video resolution levels (1080p vs. 720p), the motion handling levels (Good, Better, or Best) and the optional components (such as the Interoperability feature and/or auxiliary video, etc.) Typically, TelePresence systems require 5 Mbps per segment when operating at 1080p-Best, which would translate to 15 Mbps for a 3-segment system (like a Cisco TelePresence System 3000 or 3200). Detailed bandwidth breakdowns can be referenced at: http://www.cisco.com/en/US/docs/solutions/Enterprise/Video/tpqos.html#wp1045102
Burst: Burst is defined as the amount of traffic (generally measured in bytes) transmitted over a (variable) sub-second interval. In the context of Cisco TelePresence provisioning, the interval of interest is 33 ms; this is because TelePresence codecs operate at 30 fps, and therefore transmit a video frame every 33 ms. This 33 ms period is also referred to as a frame interval. The maximum data that Cisco TelePresence systems send within a 33 ms frame interval is about 64 KB per segment. Therefore, a 3-segment system (like a Cisco TelePresence System 3000 or 3200) should be provisioned with at least 192 KB of burst at each node. More burst is required if optional components (like auxiliary video) are also configured.
Latency: The latency target for Cisco TelePresence is 150 ms, which is in line with the G.114 specification for real-time transmissions. However, there may be instances when this target cannot be met—due to extreme geographical distances involved and/or satellite circuits. In such cases, user interaction-lag doesn’t become unnaturally excessive until latency exceeds 200 ms.
Jitter: Jitter is defined as the variance in network latency. Measurements within the Cisco TelePresence codecs use peak-to-peak jitter, but at the video frame level (and not the packet level). Cisco TelePresence has a (video frame) jitter target of 10 ms, peak-to-peak. However, Cisco TelePresence System de-jitter buffers are dynamically adjustable to adapt to higher levels of jitter.
Loss: Cisco TelePresence is highly sensitive to packet loss, and as such, has an end-to-end packet loss target of 0.05%. Calls will higher levels of loss can still proceed, but with noticeable quality degradation.
These service level requirements and targets for Cisco TelePresence are summarized in Figure 2.
With these service-level requirements in mind, the next step in QoS design is to formulate an end-to-end Differentiated Services strategy for provisioning Cisco TelePresence. The Differentiated Services (DiffServ) strategy should include an end-to-end marking value, policing guidelines, queuing guidelines, and shaping guidelines, among other elements.
Marking: The best formal guidance for Cisco TelePresence marking is provided in RFC 4594 “Configuration Guidelines for DiffServ Service Classes,” which describes a “Real-Time Interactive” service class which is intended for inelastic video flows, such as Cisco TelePresence. The recommended marking value for this application class is Class Selector 4 (CS4), which is equivalent to DSCP 32. Cisco’s implementation of RFC 4594 is summarized in Figure 3.
Policing: In general, policing Cisco TelePresence traffic should be avoided whenever possible due to its high sensitivity to packet loss, although three notable exceptions to this exist: The first place where policing Cisco TelePresence might prove beneficial in the network is at the campus access edge (this is an optional policy to limit potential network abuse). The second place Cisco TelePresence traffic may be policed is if Cisco TelePresence is assigned to a Cisco IOS Low-Latency Queue (LLQ) at the WAN/VPN edge (this is because the LLQ feature includes an implicit policer). The third place that Cisco TelePresence is likely to be policed in the network is in a MPLS VPN scenario, specifically at the service provider's provider edge (PE) routers, in the ingress direction.
Queuing: To achieve the high levels of service required by Cisco TelePresence, queuing policies must be enabled on every node along the path to provide service guarantees, regardless of how infrequently congestion might occur on certain nodes. RFC 4594 specifies the minimum queuing requirement of the Real-Time Interactive service class to be a rate-based queue (in other words, a queue that has a guaranteed minimum bandwidth rate, such as a Cisco IOS Class-Based Weighted Fair-Queue [CBWFQ]). However, RFC 4594 also makes an allowance that while the PHB for Real-Time Interactive service class should be configured to provide high bandwidth assurance, it might be configured as a second EF PHB (that is, a second strict priority service treatment, such as a second Cisco IOS LLQ); although Cisco TelePresence would continue to use a CS4 marking value.
Shaping: With respect to shaping guidelines, it is recommended to avoid shaping Cisco TelePresence flows unless absolutely necessary. This is because the objective of shapers is to delay traffic bursts above a certain rate and to smooth out flows to fall within contracted rates. Therefore, shapers could have a negative effect on jitter values for Cisco TelePresence. However, in certain situations, shapers are a necessity. One such case is sub-line-rate circuits, where the only way to force queuing policies to engage at a sub-line rate is to implement a hierarchical shaping policy with a nested queuing policy.
Thus, with the service levels of Cisco TelePresence clearly understood, and a strategic DiffServ policy in place for Cisco TelePresence marking, policing, queuing, and shaping, the network administrator can now move on to Place-in-the-Network, platform-specific configurations to implement these designs, such as those documented in the Cisco TelePresence Network Systems 2.0 Design Guide or in the upcoming Cisco Press book Cisco TelePresence Fundamentals.
About the Author:
Tim Szigeti CCIE #9794, is a technical leader at Cisco Systems within the Enterprise Systems Engineering (ESE) team, where he has spent the last decade specializing in quality of service technologies. His current role is to design network architectures for the next wave of media applications, including TelePresence, IP video surveillance, digital media systems, and desktop video. He has coauthored many technical papers, including the Cisco TelePresence Network Systems Design Guide and the Cisco Enterprise QoS Design Guide, as well as the Cisco Press books Cisco TelePresence Fundamentals and End-to-End QoS Network Design. Tim holds a Bachelor of Commerce degree in Management Information Systems from the University of British Columbia.
Cisco TelePresence Fundamentals
Tim Szigeti, Kevin McMenamy, Roland Saville, Alan Glowacki
Pub Date: 5/27/2009
US SRP $60.00
Publisher: Cisco Press