Two trends have accelerated change in carrier networks. The deployment of high-speed optical transport technology, such as DWDM and OC-192 lasers, has increased available bandwidth. Simultaneously, worldwide demand for bandwidth has increased with the growth of IP data traffic, fueled mainly by the Internet and the demand for private IP network service.
As carrier networks evolve to take advantage of new transport technologies and meet the demand for IP bandwidth, networking equipment needs to provide converged solutions enabling cost-effective scaling of networks. The new network model should reduce upfront capital expense while providing ongoing improvements in operational expense and network reliability.
Of course, there are multiple visions of how best to marry the various technologies. The main question appears to be whether carriers should control their networks using only routers, or some combination of routers and optical cross-connect equipment.
In what may be called a "hybrid solution" (see "Hybrid Solution" diagram, above), all optical links coming into a PoP are terminated on an optical cross-connect. The cross-connect switches transit traffic (data destined for another PoP) directly to outbound connections, while switching access traffic (data destined for this PoP) to ports connected to an access router. The granularity at which data can be switched is, in itself, a matter of some debate. Existing digital cross-connects are optimized to switch in units of DS-1 or DS-3, while newer optical layer cross-connect (OLXC) entrants are suggesting the abundance of wavelengths provided by DWDM systems means customers need only worry about managing at units of OC-48 and higher.
A router-only solution assumes that IP-based traffic is driving all the growth in the network, and the most cost-effective solution is to terminate all optical links directly on the router (see "Router Solution" diagram, below). The granularity at which switching is now done is at the packet level.
In both of these cases, there is an assumption that DWDM equipment is available to provide the underlying bandwidth. The debate is really about the efficiency of a pure packet-switched network vs. a hybrid, which packet switches only at the access point and circuit switches through the network.
Benefits and Shortcomings
The trade-off between the designs appears straightforward. Router-only advocates stress that as the majority of traffic becomes IP (predicted to be more than 95 percent by early next decade), it is more efficient to build a network strictly optimized for IP. This minimizes the cost by requiring fewer boxes in the PoP and decreases the network management load. Hybrid advocates insist routing is expensive and should be used only for access, because circuit switching is better for asynchronous traffic and restoring network failures at the optical layer is well understood.
The downside to the router approach is that carriers need to map multiple protocols to IP, and routed networks are not considered as reliable or as capable of fast restoration. Hybrid networks have the disadvantages inherent to a circuit switch model--they waste bandwidth and introduce an N2 problem, e.g., to connect 30 routers in the east with 30 routers in the west requires more than 900 connections. Finally, as it is assumed they will need a router, hybrid solutions include another device to manage.
Cost Comparison
While some of these trade-offs are difficult to quantify, the cost differences between the two approaches are relatively easy to define.
In a routed network, the router handles all traffic in a PoP. All packets go through the router whether they are accessing the network through that PoP or entering and exiting via backbone connections. Assuming that 80 percent of connections are for transit traffic and 20 percent are for access, the cost of the router may be estimated as follows:
The total number of ports equals the number of access ports plus the number of transit ports. Since the number of transit ports is 4 times the number of access ports (on an equal bandwidth basis), the total number of ports is 5 times the number of access ports. This implies the cost of a router is 5 times the cost of a single port (see "Equation 1," Cost Evaluation Formulas Chart).
In a hybrid network, the router handles only the packets accessing the network through that PoP. Packets entering and exiting via backbone connections are handled by a cross-connection, either at a time-division multiplexing (TDM) level or as an entire wavelength. All connections will need to terminate on the cross-connect to provide reconfigurability of the network. Making the same assumption that 80 percent of ports are for transit and 20 percent for access traffic, the cost may be estimated as follows:
The total number of ports equals the number of router ports plus the number of OLXC ports. The total router ports is equal to 2 times the number of access ports (those required for access plus the number used to connect to the OLXC). Because this scenario no longer takes advantage of the packet-multiplexing capabilities of the router, the OLXC port count equals the number of transit ports plus twice the number of access ports (one set to get traffic into the PoP and one to connect to the router). Using the fact that the number of transit ports is 4 times the number of access ports, the total number of OLXC ports is 6 times the number of access ports. This leads to a total cost of 2 times the cost of a router port plus 6 times the cost of an OLXC (see "Equation 2," Cost Evaluation Formulas Chart).
Putting these two calculations together, the hybrid solution proves more cost-effective than the router whenever the port cost of the router is more than twice the port cost of the OLXC (see "Equation 3," Cost Evaluation Formulas Chart).
Device Design Considerations
All switching devices currently available are opaque switches. That is, they convert photonic data flows into electronic flows that are then operated on by the switch. Currently available products will switch based on packet identifiers (frame relay, ATM, IP routers, multiprotocol label switching switches), time slots (TDM, digital cross-connect) or wavelengths.
Given this fundamental similarity, all devices can be mapped against a single architecture (see "Switch Components" diagram, below). The three constants in all of these devices are the need to transmit and receive photons, convert these to electrons and switch the resulting electron streams across a fabric.
There are differences in what happens to the three data types once they have been converted to electrons. Wavelength cross-connects do the least amount of additional processing--a stream of photons are converted to electrons and switched. The electronic information stream is not modified in any way.
In a TDM cross-connect, a framer uses timing information imbedded in the electronic flow to divide the information into multiple channels. Each channel contains less than the full bandwidth of the connection. Some examples of this are using an OC-12 connection divided into four parts of OC-3 bandwidth.
Finally, in a routed flow, each channel may be examined for packet-header information.
Given the evolution of application-specific integrated circuit (ASIC) technology and the emergence of single-chip framers and routers, it is a reasonable guess that routing will add no more than 25 percent to 50 percent to the cost of the basic switching capability. Thus, it is likely that integrated switch/routers will prove to be more cost-effective than solutions that separate the switching and routing functions between two devices.
Growth--The One Constant
What are the key goals in designing switching and routing systems, and what will differentiate systems? The one constant is the continuing growth in traffic, fueled mostly by he Internet.
The "Bandwidth Growth" chart (above) gives an example of the increasing demand at a single PoP, assuming about 4 times growth per year. The straight line indicates the cross-sectional switching bandwidth in the PoP. Possible link rates may be used to provide the necessary connectivity, starting with OC-12 and migrating to OC-192 and beyond. By taking the bandwidth required and dividing it by the optimal link speed, the curved lines, which represent the required port count (or wavelengths), are obtained. Note the number of ports rapidly rises to hundreds of OC-48s and beyond.
The key differentiator will be how well a switching system scales to support this number of high-speed links, and how well a network management system provides the capability to define the processing required on each link.
The PoP of the future will need to support hundreds or thousands of ports, switching information at tens to hundreds of terabits per second (tbps). The connections will be optical with optional support for TDM channelization to support lower speed connections. Optional packet processing will support either access traffic or sub-rate packet switching using multiprotocol label switching (MPLS). Network management systems will allow the carrier to provision end-to-end connections to take advantage of a mix of packet, TDM and wavelength switching at the edge as well as in the core of the network, to deliver the appropriate level of service.
Peter A. Chadwick is vice president of product management for Avici Systems Corp. (www.avici.com). He can be reached at pchadwick@avici.com and (978) 964-2080.
