The IP everywhere reality

Radical improvements in the cost performance of fiber optic transmission-together with the success of packet-based services-are causing many in the communications industry to proclaim that the new order will be "IP everywhere."

Over time, Internet protocol will be everywhere, but as new IP services evolve, some of the most important applications will require real-time communications. Most computer users will want to connect and interact with others using the same power and media richness they now experience working at their own desktops.

Conversational computing-a new class of real-time, connection-oriented communication applications that allows users to communicate and provides live, real-time access to stored multimedia information-is driving the need for new service delivery capabilities from the communications infrastructure.

As a result, IP services inevitably will be grouped into service tiers priced according to quality of service (QOS) and perceived by users as "good," "better" and "best." Best-quality IP services will be defined by users' expectations of having "circuit-quality" end-to-end latencies and throughputs when used for real-time, connection-oriented services.

To date, however, effective, real-time, connection-oriented communication services have been successfully scaled to massive levels (but limited to narrow bandwidth connections) only in the public network.

Lessons learned from the public network can be applied directly to the introduction and scaling of real-time, connection-oriented services on IP networks. In particular, two important lessons from the public network are:

* The value of implementing an end-to-end, out-of-band signaling network.

* The value of centralized service control point (SCP) intelligence for coordinating the real-time management of massive connectivity.

However, what was critical for scaling-up the public network-and what IP networks currently lack-is a comprehensive signaling capability with an end-to-end view of the network. Best-quality, real-time IP communication services can benefit from end-to-end connectivity management models, but the signaling technologies implemented in the public network today are limited in function and will serve only a peripheral role in enabling large-scale IP connection services. Rather, broadband connection-oriented IP services demand real-time bandwidth mapping and allocation, IP address resolution, and national directory and security management capabilities not present in today's public network systems.

QOS challenges with a routers-only approach Packet tagging promises to provide an initial, intermediate solution for improved IP service. This solution "tags" packets that must be delivered through the end-to-end network with small and predictable delays. Each switch or router that touches this packet flow is expected to give these tagged packets a higher priority over normal packets.

The theory is that when providing this capability over "party line" circuits that have been engineered to provide significant excess capacity (made possible by cheap, fiber optic transport technology), and by using extremely fast routers to process the packets, carriers can provide circuit-grade QOS for specific packet flows over the same infrastructure that is providing successful best-effort packet services today.

Although this approach will result in "better-effort," high-priority services, it probably will not consistently deliver large-scale, end-to-end, connection-oriented IP services with true circuit quality performance any time soon. This is because high-priority packets must still contend with each other at each router node, resulting in unpredictable latency, or jitter. Also, most of the cost advantages resulting from packet interleaving on shared circuits are lost with the significant over-provisioning of bandwidth.

More important, massively distributed router intelligence can't easily pre-negotiate end-to-end constant packet flows with the knowledge that packet flow reservations actually will be honored at every node through which a connection passes. Nor is it an easy task to keep such distributed intelligence synchronized with all the knowledge they should have about the entire end-to-end network. And of course, the problem gets even worse across large multicarrier public networks.

In addition, the original economic driver behind the evolution of packet services (conserving scarce and costly transport resources) has largely disappeared. It is somewhat ironic that today's router-based Internet is now experiencing an issue similar to one the public network faced in the 1960s and 1970s: It needs an end-to-end view of the network for the delivery and scaling of premium quality connection-oriented services.

Best of both worlds For many IP services, distributed routers are a wonderful, cost-effective and practical solution. But when it comes to premium IP services that demand circuit-quality connections to be effective and useful, we are forced to ask, "What's wrong with circuit switching?"

Clearly, a switched, end-to-end connection, dedicated to a single user, surpasses all other approaches for providing premium quality IP services. Although any switched connection wastes some bandwidth during idle periods, this needs to be balanced against the basic management inefficiencies and complexities of proposed distributed reservation schemes. Also, given the plummeting cost of bandwidth, the quality, simplicity and reliability with which connection-oriented services are delivered will become much more significant in the marketplace than the efficiency of bandwidth usage. This will be especially true for premium services. Effectively managing massive connection setup requests in real time is another issue that must be addressed.

The prevailing "IP everywhere" belief is that all IP services-including connection-oriented services requiring constant packet flows-should be implemented using only routers with fully distributed intelligence. But the IP technical community's historical antipathy toward the concepts of centralized authority may be causing it to miss out on the lessons learned from the massive scaling of the public network's global connection-oriented voice services.

The perceived limitations and benefits of end user-controlled circuit switching may have been colored by the specific narrowband limitations of Class 5 switches and the end-to-end signaling complexities of the SS7-based circuit switching infrastructure and addressing models as they are implemented in the public network today. As previously stated, public network connectivity solutions should be used not as specific technologies, but rather as a generalized model for how to massively scale and account for connection-oriented IP services.

Peaceful coexistence between the emerging demand for circuit-quality IP services and the demand for best- (or better-) effort connectionless IP services will not be resolved by battling it out over the merits of switches vs. routers, or even time division multiplexing vs. asynchronous transfer mode vs. packets. Each of these switching, routing and transport technologies brings specific benefits to specific tiers of IP services and networks. The important answers lie elsewhere.

Given the appropriate network signaling and control mechanisms, IP addresses can be used to control switched IP connections as effectively as routed IP connections (Figure 1). In light of these results, the emerging universal IP network architecture will combine network control capability and transport infrastructure. This combination will include both routing and switching capabilities-a combination sufficient to provide support for circuit-quality IP "connections" when needed, without diminishing the application flexibility that makes IP so compelling.

Through this universal IP network architecture, IP truly becomes an application-independent communications enabler. Network users will no longer need to sacrifice reliability for flexibility, and they can have as much of both in any application. And the efficiencies of the IP addressing and transport model-along with its independence from any particular infrastructure-will continue to be the major components that drive the IP everywhere reality.

Back to the future The challenge for connection-oriented IP services is managing massive connectivity in real-time. What has been missing is a highly scalable, user-controllable, end-to-end signaling capability that can reserve truly guaranteed constant packet flow paths for IP connections of arbitrary bandwidths.

Providing such on-demand IP connections, especially with circuit-quality service levels and security, requires two major network capabilities:

* Switches and routers capable of delivering high volumes of non-blocking, non-contending, constant packet flow IP connections from their input ports to their output ports (for example, digital cross-connects, ATM virtual circuit switches or QOS routers).

* A real-time, highly scalable centralized intelligence capable of mapping and allocating the bandwidth that is needed-and that is available (Figure 2).

Such centralized intelligence is similar in basic function to the role filled by SCPs in today's connection-oriented public network topology. But an IP-SCP must support significantly different processing capabilities and computational intensity than what has been required from conventional voice network SCPs.

The challenge, then, was to develop a real-time connection management model for IP that could support connection-oriented, broadband IP services at a massive scale (Figure 3). A secondary, but equally important, goal was to make such services appear as "natively IP" and as standards-compliant as any best-effort routed service.

Fortunately, two decades of Moore's Law have made the cost of raw processing power significantly less expensive than when public network SCPs were designed. This has made it possible to develop methods for continually adding incremental processors to an IP-SCP as service demand escalated. WarpSpeed expanded system scale in this manner without the loss of efficiency that usually accompanies such an approach.

The resulting IP-SCP solution harnesses inexpensive and reliable raw processing power. This power makes the IP-SCP's connection management task, for millions of potential end users, a manageable challenge by pre-calculating information about possible paths through the physical network and storing this static information in a very fast database. Then the IP-SCP calculates-in real time and at the request of end users-the key pieces of dynamic data that it needs to establish a dedicated Layer 1 or Layer 2 connection that can provide the user requested bandwidth: the connection between each endpoint and through the middle of the network.

The IP-SCP does this by determining which portions of the pre-calculated path information can best serve this connection request based on the IP addresses of the connection endpoints and according to pre-defined selection criteria. The IP-SCP merges the path information with the real-time state and bandwidth availability for every node on every path that can potentially support the connection request-resulting in a connection path that will be able to satisfy the connection request.

This is an enormous amount of data to process in milliseconds. What makes it possible to funnel all of this information into a real-time decision, and an action, is the unique and highly scalable internal processing architecture of the IP-SCP.

The net result of this effort is an IP connection-oriented network service capability based on a scalable end-to-end signaling and bandwidth allocation solution, which enables end users to establish usage-priced, point-to-point, circuit-quality extranet IP connections, with user selectable bandwidths, on demand.

A future of classes As IP begins to be used for delivering all types of services-and as carriers are driven to differentiate themselves-significant demand for differentiated IP services will emerge.

These differentiated services will divide into traditional categories: "first class," or circuit-quality, on-demand connections; "business class," or priority-queued services with average throughput and burst guarantees; and "economy," or best-effort service. Providing multiple levels of service on demand, with different pricing choices, benefits users and rewards service providers.

In satisfying the emerging demand for premium connection-oriented, circuit-quality IP services at national or global scale, there is much to be learned by applying architectural lessons from the successful scaling of similar connection-oriented services in the public network.

Although the concepts of end-to-end signaling and a scalable service-managing central intelligence are essential for large-scale, high-quality, connection-oriented IP services, existing SS7 technologies don't lend themselves well to this job. They were not designed from the ground up to process connections based on IP addressing, and they don't allow user requests for bandwidth to be allocated among particular paths through network elements.

A properly designed central intelligence with sufficient scalable transaction processing power, deployed in an IP network, provides a comprehensive new solution for policy, security, billing and settlements issues surrounding connection-oriented IP services in a multicarrier environment.