A walk through the architectural challenges in delivering wireless data

" State" is information about a user and the connections made by a network that is relevant to making the user's experience convenient. This user state information and the relevant service information for a user is called "context." This includes not only the physical state of the user, for example, presence and connectivity to the network, but also logical properties, including service preferences and charging policies based on the type of network access being used at a specific point in time.

In contrast, mobile data networks deliver personalized services to individual roaming subscribers. Each subscriber may have more than one service connection — for example, e-mail, virtual private network, instant messaging, and content filtering — and the network must maintain awareness of the real-time state of each connection, along with the associated subscriber service policy and charging policy, in order to accurately bill for services. Since a typical mobile network will handle up to 2 million subscribers per point of presence, mobile data architectures are for the first time forcing IP network planners to maintain large amounts of per-subscriber state.

Unlike stateful voice networks, IP networks are largely state-less. By distributing processing intelligence with robust protocols, IP networks have grown organically without end-to-end state management and control embedded within the network. A significant benefit of this paradigm has been the rapid decline in the cost and price of IP network elements, to the point where it is more economical per-bit than traditional circuit-switched-voice network technology that IP networks overlay.

Because of the state-less nature of IP, networks and applications do not tend to tailor the services delivered based on the state of the user, making effective delivery of services a challenge in some mobile environments. For example, when you fire-up Windows/Outlook over a mobile data network link, the application may be useless because it doesn't know the context of your connection, nor can it effectively adapt to the change in bandwidth context or the fact that you may be intermittently connected.

Another challenge lies in the network transparency across circuit and IP network access technologies. As native IP access technologies are becoming more economical (Wireless LAN 802.11), user and service transparency across access technologies becomes more important in delivering a convenient user experience. The user should expect to receive the same service, but perhaps a different delivery experience and charging profile, while his or her location and context changes across GSM / GPRS and WLAN 802.11 access networks.

Designing and building the infrastructure to manage differences in service "context" based on the subscriber state is key to enhancing and customizing network service for the end-user.

A significant technical challenge in managing user and service "context" in mobile networks is the problem of rapidly switching user contexts based on dynamic state information and thereby ensuring the user content (packets) is processed appropriately. With mobile voice telephony there was basically one application (voice) with branches of subordinate services ( speed dialing, call forwarding). With mobile data services, there are many more possible applications and no single killer application.

Further complexities arise due to the unique scale associated with mobile networks. With over 1 billion mobile users in the world, mobile data networks must scale to support high subscriber capacities and high traffic density, and must be resilient to network and service disruptions, intermittent connections, and other anomalies that naturally occur in large-scale networks. Managing user and service context with such diversity, capacity, and reliability goes beyond the abilities of today's state-of-the-art general-purpose processors.

Bandwidth limits
General-purpose processors have followed Moore's Law and have increased orders of magnitude in processing speed over the decades. However, a server using general-purpose processors in a big Web farm today may handle only hundreds of simultaneous user contexts instead of tens of thousands. While memory density has increased and costs have decreased dramatically as processor speeds have improved, the bandwidth of getting the bits in and out of memory has not improved proportionally. The limit of hundreds of simultaneous contexts in a general processor is due to the memory I/O and "context" switching limits of getting program and user data into the CPU chip (cache), so it can be used by the high speed CPU.

"Context" switching speeds have improved over the years with real-time OS and silicon economic improvements, but the memory bandwidth and context-switching speeds of general-purpose CISC processors have not improved at the pace of optical transmission speeds, CPU speeds, and memory density. So, when faced with the challenge of selecting any one of hundreds of thousands of user contexts at the next instant, traditional processor technologies and solutions, which involve racks of general purpose processors, have difficulty correlating one user's context and adapting it to one of many differential services at any instant.

A mobile service delivery platform uses programmable silicon for processing all of the bits of a packet while accessing uncorrelated subscriber policy and context from memory in real-time. For each incoming packet, a programmable processor correlates the user's source and destination addresses with multiple application and protocol types. Source: Megisto Systems

A purpose-built mobile service delivery platform employs programmable silicon specifically designed to efficiently support high memory bandwidths and handle hundreds of thousands of user contexts. The net result is lower cost network processing with greater richness of service delivery, charging, and high fault-tolerant reliability.

The programmable silicon in such a system processes all of the bits of a packet while accessing uncorrelated subscriber policy and context from nearby memory in real-time. For each incoming packet, the programmable ingress processor subsystem correlates the user's source and destination addresses from a hundred thousand or more, with multiple application and protocol types, as well as with personalized accounting/billing data.

The programmable silicon reads and processes information deep into every user packet — a significant scaling challenge for commercial processors and even state-of-the-art network processors. This type of processing is required in order to charge and bill a user for services and applications delivered, versus just billing for raw bit-pipe usage. To improve the user's experience, the silicon can adapt the flow of packets to and from a user by filtering, trans-coding and other real-time content optimization on the entire packet. The result is highly scalable processing of both user state and context.

Managing user connection state and context changes in a manner that is transparent to the user is the heart of how mobile networks have added profitable value to voice telephony over the last 15 years, growing to a market of more than 1 billion users.

Because of the quantity and complexity of state information required to create successful mobile data services, operators should look carefully at how state and context information is processed. By choosing a solution that is purpose-built to evolve mobile core networks to data, using programmable silicon that applies state intelligence deep into user packets, operators can minimize migration headaches in the short term, and be well positioned for growth in the long term.