Converged Networks of the Future
Krishan Sabnani, Bell Labs, Holmdel, NJ 07733
It is widely accepted that services offered over circuit switched networks and packet-switched networks will converge onto a few multi-service networks supported by an IP/MPLS core. These converged networks will transport voice, video, and best-effort data, and their blended combinations. Figure 1 shows such a converged network. End-users will connect to the network through a variety of access networks via specialized access boxes. In most cases these access boxes perform all processing specific to the access technology and provide a common data stream to the IP/MPLS core network. The benefits of such a converged network are great: reduced operational and capital expenses (opex and capex, respectively), access to a common, consistent set of services regardless of access technology and location, and basic services, such as security and location services, provided across networks. Many challenges remain to realize this vision. Below we provide a sampling of projects that address challenges in the access network, core transport network, and core services control.
Access Network. The main challenge for access networks is to fully terminate access functions specific to the access technology in the access box. By doing this, a single common core network can be used to provide homogeneous end-to-end solutions for routing, quality of service, security, etc. While many access technologies are straight-forward in this respect, current wireless cellular third generation (3G) networks are not.
In these networks, users are connected to the backbone network through complex radio access networks (RANs) comprised of base stations, backhaul concentrators, radio network controllers, and finally gateway boxes that convert access protocols into the backbone network format, i.e., IP packets or voice frames for the PSTN. The access network functions are spread amongst these boxes, and access technology-specific functions propagate all the way into the gateway boxes (PDSNs for CDMA2000 networks and SGSNs for UMTS networks) which connect to the core IP/MPLS.
We are currently performing research to replace these complex RANs with simple access boxes called base station routers (Figure 2) (BSR). These BSRs terminate all air interface specific functions and provide mobility support at the network layer. An analogy can be drawn between these networks and simple 802.11 networks: management is simple, a variety of backhaul layer 2 technologies can be used (e.g., Ethernet, T1, SONET), and end-to-end protocols can naturally terminate at the network edge improving network efficiency. This approach specifically:
- Reduces end-to-end latency leading to enhanced support for consumer and business applications (for example gaming or remote access to applications such as MS Office). The reduction in network delay allows the admission of higher loads while still maintaining delays low enough to support real-time services. For a VoIP system, up to 30% more voice users may be supported.
- Reduces CapEx by approximately 30% through lower equipment costs, savings in site deployment, and network integration compared to a traditional UMTS network roll-out.
- Reduces OpEx by simplifying network management and easing the adoption of alternative backhaul technologies, for example Ethernet or xDSL. Backhaul savings of up to 75% can be achieved using alternative packet-based backhaul versus tradition transmission lines.
This approach poses several unique and challenging research problems because of the characteristics of CDMA and WCDMA networks. First, tight power control requiring feedback from the network to the terminal must occur on the order of thousands of times per second. Second, soft handoff must be supported in which a terminal transmits through multiple base stations simultaneously and only a single copy of the radio frame is selected to be forwarded into the core network. Both of these functions require tight bounds on transport jitter. In addition, while these algorithms were performed in a central site in traditional 3G networks, the disaggregation of the BSR system requires communication between BSRs to implement the distributed algorithms.
As new wireless interfaces emerge, some of the RAN functions required in CDMA and W-CDMA functions may no longer be required. However, given that later versions of the 802.16 standard (802.16e, for example) still propose using soft handoff, the
fundamental results on packets based RANs and BSR technology will enable a migration to end-to-end IP/MPLS networks.
Core Transport. Building the IP/MPLS core for these converged networks poses several significant challenges. Current IP networks are best-effort, poorly managed, and not secure. For carrier-grade performance, the core networks will be called upon to provide an increasing number of services such as QoS-support, high levels of security, and more flexible manageability. Attempts to add more services to existing router platforms have been met with mixed results because of the tension between of keeping routers simple and optimized for packet forwarding, and the inevitable complexity that is introduced by adding software control functionality.
We are working on a unique way of building such core IP networks based on softrouters. In the softrouter approach, routers are disaggregated into simple forwarding elements and control elements (Figure 3 depicts the logical view and Figure 4 depicts the physical view of the architecture) that execute most of the network and services control software. This approach enables easy addition of new value-added functions into IP networks and eliminates the dependency linking the location of a service to the routing path of a packet.
An example application in the wireless domain is that of mobile IP support. In existing implementations, the home agent is often implemented as a router feature, where both the control (e.g., registration, authentication) as well as transport (e.g., encapsulation and tunneling) functions are supported in the same box. This creates two problems. First, because of the processing and memory limitations of the typical router controllers, such an implementation can present significant scalability issues and are not cost efficient. Second, because control and forwarding are linked, inefficient triangular routes are present.
Due to the relatively low bit rate of wireless data traffic as compared to broadband wireline access, the forwarding engine of a typical router may be underutilized when deployed as a home agent. The naïve solution for using the forwarding resource more efficiently is to support more users on a single home agent. However, each user requires a significant amount of memory in the router controller to maintain its location information and timers to keep state fresh. The controllers must process frequent updates to support mobility and
long-lived sessions. This mismatch between the large required growth in processing and memory requirements per user, with the relatively low increase in demand on the forwarding engine of the router, results in an inefficient router deployment.
The SoftRouter allows the complex control software to be offloaded to an external dedicated server optimized to process control messages and maintain per user state. Forwarding elements may be deployed as dictated by bit rate requirements. This allows each portion to grow independently to optimize the network deployment. In addition, the forwarding elements need not be co-located with the control elements, so the impact of triangular routing is removed.
The research issues solved in the soft router project will have a large impact on other transport-affecting services in the Internet.
Services Support. In these converged networks, common application level functions such as single sign-on, personalization, global roaming, and always-on are provided by a common service enablement layer. These functions are used by all services as building blocks and are accessible from all access networks. In particular, by providing this common layer, many applications may be built using the always-on characteristic.
Some examples of such “always-on” service are push-to-game, push-to-see-what-I-see, and push-to-traffic-information. These services will be very easy to use and provide a user perception of being “always connected” to the network.
In order to enable these new services, which are commonly referred to as push-to-connect services, the service intelligence must be built into the network. However, currently deployed push-to-connect services use the network as a dumb bit-pipe. This results in (1) long setup and delivery delays due to large number of over-the-air messages between the network and the user, and (2) large execution delays and resource waste caused by non-optimized and non-scalable implementation of common functions. In order to solve this problem we are working on creating an intelligent networking infrastructure with a goal to provide always-on services across multiple networks and services. We propose to use an “overlay” network which is independent of the underlying network technology (3G, 4G, WiFi) and the administrative boundaries. The overlay can
provide always-on services either to the customers of a specific wireless carrier, across disparate link layer technologies, or to the customers of a Mobile Virtual Network operator (MVNO) across multiple 3G wireless service providers’ networks.
The overlay network is created using two network elements, namely, Always on Gateway (AOG) and Always on Redirector (AOR). The goal is to bundle always on specific functionalities into the AOG so that the existing network elements in a service provider’s network do not need to be changed. The purpose of AOR is to transparently route traffic from various access networks into the overlay network. Once the traffic is hauled into the overlay network, the AOG uses a resident user agent to provide the necessary functions thereby reducing over-the-air communication. By using a common server, a uniform user experience is provided across different services.
Summary. To fully realize the goal of a converged backbone network, three challenges must be met. First, access networks must provide a homogeneous bit stream to the core network. Second, transport and control in the core network should be separate so each function can be optimized. Finally, core services must be ubiquitously provided so that consistent services may be accessed regardless of location.