Achieving sustainable revenue growth requires tight controls on quality of service across a range of services - and carriers must choose the right platform
Telecommunications service providers are seeking ways to increase Average Revenue Per User (ARPU) through value-added services such as IPTV, Video on Demand (VoD), and video collaboration applications such as on-line gaming. The services must be delivered in conjunction with VoIP and Internet access, over a single subscriber connection. With the cost of customer acquisition at ~$300 and annual per-subscriber multi-play service revenues at over $1000, subscriber loyalty is critical. Poor Quality of Service (QoS) results in subscribers switching to competitive offerings (cable, satellite, etc). Therefore, adequate service quality must be maintained. Of course service offerings must be price-competitive. Therefore, tight control over costs (CapEx and OpEx) is important.
Early approach to QoS
The basic challenge of delivering multi-play services arises because the services have different characteristics. Voice has stringent latency and jitter requirements, although bandwidth per channel is low. High quality video has low jitter tolerance, however, bandwidth per channel is much higher. Latency tolerance varies with the video service - VoD tolerates higher latency than video gaming. Data traffic (e.g., email, browsing) is largely agnostic to jitter and latency.
Prior to the era of ubiquitous rich content, social networking and file sharing, it was assumed that Internet traffic would remain a tiny component of overall traffic. Hence it was believed that multi-play QoS could be guaranteed with minor over-provisioning of bandwidth.
Bandwidth no QoS guarantee
With initial multi-play service deployments, it became apparent that though Internet use generated far less traffic than voice or video, it is bursty. With increasing rich content, the spikes get bigger. With recent widespread use of bandwidth-hogging applications (YouTube, eDonkey, Bittorrent etc), the situation has worsened. Several users accessing such applications simultaneously places a significant load on the network. In networks where traffic is not intelligently managed, excess traffic is dropped when provisioned bandwidth is exceeded. Indiscriminate traffic drops result in information loss for all types of traffic, and potentially for all subscribers, even those not using those applications. From the perspective of the end-user, loss in Internet data traffic is hardly perceptible, because re-transmission mechanisms recover losses before presentation. Conversely, even small traffic losses cause unacceptable quality degradation of voice and video.
Bandwidth over-provisioning at very high levels may guarantee QoS, however, it is often infeasible in many access networks and certainly not affordable.
Looking ahead, there is no crystal ball to predict what new services will be deployed, what new applications will be invented, and how usage patterns will evolve. What is certain is that for the foreseeable future, Internet traffic and temporal unpredictability will increase. According to studies published by Cisco Visual Networking Index, Internet video (excluding P2P video file sharing) already constitutes a third of all consumer Internet traffic, and is expected to grow at an annual rate of 58% over the next 4 years. New security threats will emerge.
Networks must be engineered to continually adapt to changing conditions. Wholesale equipment replacement is unaffordable. Software upgradeability is an imperative.
Attributes of NGNs
Networks must be built with platforms that offer adaptability, intelligence, security, control, and scalability.
Adaptable platforms enable the creation of "learning" network elements. With learning network elements, better traffic management, different protocols and services can be supported without hardware upgrades. Rather, improvements are enabled by software upgrades. For example, new traffic management algorithms are deployed as software upgrades to adapt to new traffic patterns. New protocols are also handled with software upgrades. An upcoming example is the imminent migration of access networks to IPv6, driven by the Internet address shortage problem of existing IPv4 based access nodes.
Intelligence encompasses service and subscriber isolation, traffic management (buffer management, policing, traffic shaping and scheduling), and the ability to dynamically configure algorithms for different network conditions. Service and subscriber isolation involves identifying traffic based on service type or origin, subscriber, etc, and separating it into distinct queues. Traffic management algorithms make discard decisions and regulate traffic flow in various queues to meet service-specific requirements modulated with subscriber-specific SLAs. Many algorithms have been devised for each function, e.g. Weighted Round Robin (WRR) for traffic scheduling. Software-based implementations of these algorithms have the advantage that they can be refined and dynamically tuned for specific network characteristics. Ideally, it should be possible to select from a menu of algorithms, so that service providers can appropriately select the algorithm and tune the parameters for each node in their network.
In broadband access, security often refers to stopping Denial of Service (DoS) attacks. Security is implemented through a variety of mechanisms - Access Control Lists, rate control of host-directed traffic, etc. An adaptable platform is essential to accommodate new threats, protocols and services.
Controls typically limit network misuse with respect to SLAs or regulatory frameworks. Policing and shaping algorithms guarantee users do not exceed bandwidth allotment of their SLAs. Service providers also want to better manage traffic related to bandwidth-hogs. In these cases, application recognition is used to identify the specific traffic that is filtered or de-prioritized.
Scalable platforms address the development of cost-effective portfolios for a wide range of performance and functionality, deployable across a broad spectrum of service providers.
A simple case study demonstrates the importance of these attributes. The WRR algorithm ensures that queue-specific weights specify relative priorities of services. But experience has shown that WRR is inadequate for triple-play QoS. To guarantee triple-play QoS, LSI recently implemented a sophisticated multi-level hierarchical scheduling algorithm, making it possible to deploy it as a software upgrade on existing nodes built on the LSI APP communications processor.
Choosing the right platform
Fixed function and Ethernet switch devices have significant drawbacks in many respects. Both include hardwired traffic management, most do not meet even today's requirements. Adaptability and functional scalability are non-existent or highly restrictive. Hence they are not suitable for subscriber-facing linecards in next generation network elements. Note that Ethernet switches are suitable, and often used in network elements for other purposes such as internal interconnects.
Programmable platforms offer these desirable attributes and are recommended for building learning network elements. However, they differ greatly in their degree of support, hence a deeper assessment is recommended. All the attributes should be affordable from a total cost of ownership (TCO) perspective. (TCO includes cost of development, maintenance and upgrade of software through the product lifecycle.). The architecture should support predictable performance in a variety of scenarios. Programmable, hardware-based scheduling with multi-level hierarchy support is critical. For known standards and world-wide service provider requirements (DSL Forum TR-101, IPv6 enablement, a menu of traffic management algorithms, etc), pre-packaged, platform-optimized software must be available. For adaptability, it is equally important that the platform vendor is committed to invest in a software roadmap. For differentiation, the architecture must be easy to program, and source code with modification rights complemented with robust tools must be available. For long-term requirements, the programmable platform evolution roadmap must not only consider new hardware functions but also incorporate a simple software migration strategy.
An example of a platform with all the desirable attributes is the LSI Broadband Access Platform Solution including APP communications processors and Broadband Access Software. The LSI Tarari Content Processors represent a good example of a platform to implement application-specific controls.
It is indeed possible to build cost-effective, future-proofed, next-generation networks that meet the requirements of multi-play services.
About the Author: Sindhu Xirasagar is Product Line Manager, Networking Components Division, LSI