Remember The Term: Dense Virtual Routed Networking.

It could change IP networking forever.

Imagine a network where unlimited bandwidth is free and speed is a non-factor. Imagine a network where IP is freed from the constraints of routers with their static routing tables; a network where data moves at wireline speeds through "virtual routers" which perform complex functions far beyond today's simplistic match-the-IP-address-with-a-geographic-location computations.

Such a network is difficult to imagine today because the Internet, as we know it, is based on a 30-year old technology model in which lowest-cost, least-bandwidth routing functions are the basis for all networking decisions, be they business or technological. But advances in optical technology--and particularly Dense Wave. Division Multiplexing-- mean that future IP-based networks will be so fast as to eliminate bandwidth allocation from the equation: speed will be limitless. In these networks, when there are no stratified levels of service or packet prioritization schemes, what becomes of the router, a device designed from the start to be bandwidth-efficient above all else?

A secretive start-up called Crescent Networks thinks it has the answer and hopes to change the IP networking industry with a new networking paradigm called Dense Virtual Routed Networking (DVRN). This technology hopes to do for optical routing what Dense Wave Division Multiplexing (DWDM) did for fibre optic transport. Virtual routing, in concept, changes everything we have come to understand and expect about networking, from the business models to the technologies behind them.

Fibre Diet

To understand, how DVRN works, it's helpful to first understand the change that DWDM brought to the optical networking market. Before wave division technology, one physical strand of fibre could carry only one wavelength of light--and hence one stream of data. This meant that, to increase carry capacity (and throughput), backbone providers had to physically add fibre cable to increase capacity. As you might imagine, this process was expensive and time consuming.

To combat this problem, service providers introduced time division multiplexing, which increases bit rates by dividing time into smaller increments so more data can be transmitted per second. But this solution means that providers must implement it across the entire network, which means adding extra capacity when and when it may not be needed which, in business terms, is a waste of money and resources.

The solution of choice for optical networks today is Dense Wave Division Multiplexing, which combines multiple signals, amplifies them together, and then sends them over a single strand of fibre. According to the International Engineering Consortium, a non-profit industry group, each signal carried via DWDM can be at a different rate (OC-3, 12, 24, and so on) and in a different format (SONET, ATM, etc.). In effect, DWDM creates what some call "virtual strands" which are essentially links of fibre that exist via software and optical amplifiers. So, one physically separate fibre strand can carry multiple wavelengths of light multiplexed together.

Today, some observers liken the state of IP-based networks to the pre-DWDM optical networking days. Capacity is increased via faster routers and switches or via wider links (greater physical bandwidth) or--more likely--a combination of both. But such solutions can be very time-consuming and require huge investments in upgrading the entire network infrastructure. While software schemes like packet-based prioritization and QoS guarantees have made some strides in increasing speeds through better network resource management, these solutions generally depend on more router complexity.

Crescent Networks of Lowell, MA wants to do for IP networks what DWDM did for fibre optic networking: Essentially, increase capacity by adding a dynamic routing "layer" that intelligently and seamlessly provisions the network on demand. In theory (and, it must be said, this is theory, since the company won't even say when its products will be announced) DVRN will allow enterprises to use multiple services from multiple providers, using many different types of network infrastructures; the company calls this "Collaborative IP Services". Under such a model, an enterprise might have any combination of intranets, extranets, VPNs, ASP-based apps (including storage and content apps), and outsourced Internet services, all tied together and managed as if they were on a local network (see Fig). Intelligent virtual routers would, unlike today's devices, have multiple routing policies, allowing traffic to be tightly controlled and managed. Application-specific services would be provisioned in a matter of hours, not week s or months.

"Our products will be a combination of software running on servers in the network operations center and hardware devices at the POP that basically collapse the racks and racks of routers and subnet switches in use today," says Curt Newton, senior director of product marketing at Crescent. "We want to allow provisioning of incremental services while at the same time reducing the number of people and the time required to do so."

The current Collaboration Model

Under today's model, companies use a mixture of Internet, VPNs, Frame Relay, ATM, and leased-line technologies to take advantage of off-site resources, be they suppliers, employees, or hosted applications. As Crescent sees it, while the trend toward outsourcing and collaboration accelerates, companies will need to dynamically build and tear down fat, shared connections among one another and eliminate the data bottleneck created by the routers on these connections. DVRN uses virtual routers that are isolated from one another, allowing customized routing and multiple fully routed networks on the same DVRN system. Virtual routers will also be fully scalable and will support multiple network protocols.

"Clearly, these guys are building some kind of device that will fit between the customer and the optical core," says Tom Nolle, president of market consultancy CIMI Corp. and one of the few analysts familiar with Crescent's technology. "They are addressing the design of future networks, where you don't want or need to build a router-based network with DWDM running between the routers. The optical network of the future is not going to be built to deal with bandwidth consumption and capacity issues. Least-cost routing won't be an issue.

Nolle believes that Crescent and other companies with similar technology will initially enter the market at the network's edge, offering DVR devices as a ring around--but connecting to--the old, router-based network, with optical transport running at the core. Eventually, we're likely to see routed IP move fully to virtual routing, where the routing of packets is distributed throughout the entire network. When this day comes, the static, table-based router will be removed from the equation. (Crescent's Newton admits that the company's solution will run as an edge service but says it doesn't demand an optical core.)

DVRN will also offer application-aware, packetlevel QoS, which Crescent calls "Application QoS", or AQS. Unlike traditional packet-sniffing QoS, which uses packet headers to help allocate bandwidth to particular data streams, AQS will, according to the company, "sense" a packet's contents, make detailed, dynamic provisioning choices based on application state awareness and policies, and collect granular stats for SLAs and billing purposes. AQS will do all of this at wire-speed on ports running at multi-gigabits per second.

"QoS to date had been geared toward bandwidth-constrained environments, especially in the last mile," says Newton. "We believe that advanced queuing techniques can be put to use for applications, tracking the services that are being used (and when users come onto the network) on a transactional basis."

Crescent's AQS is fundamentally different in concept from traditional QoS. In today's networks, QoS prioritizes packets using two assumptions: that some data is more important than other data and that set bandwidth for this data should be reserved (sometimes referred to as "guaranteed"). Application Quality of Service running on optical networks will presumably not have to deal with bandwidth allocation since throughput is virtually limitless (in terms of speed). This allows for an entirely new form of data management where packets from multiple applications can go in multiple directions depending on the application's (and the user's) need at the moment.

A few other companies are rolling out similar--albeit not as ambitious--services as alternatives to costly frame relay and the management overhead associated with VPNs. For example, Tenor Networks, also in the Boston area, is creating optical "service" switches that offer end-to-end, software-driven provisioning and service measurement. Tenor is also rolling out its ServiceAware Port, which supports processing of IP, TDM, ATM, or frame relay services, all on the same universal hardware.

Other companies like Sycamore Networks, Corvis Corp., Equipe Communications, and Tachion Networks are working on different aspects of socalled intelligent optical switching; most belong to the Optical Domain Service Interconnect (ODSI) Consortium which Sycamore founded in January. ODSI (www.odsi-coalition.com) is seeking to define a standard method of interoperability which will enable electrical devices (routers and switches) to make intelligent bandwidth requests directly to an optical network, via optical switches. The first ODSI interoperability test event will take place in early December in San Jose. (Like Crescent, many of the companies in the Consortium are long on technology announcements and short on shipping products.)

Other members of the ODSI group are concentrating on the last mile, bringing optical speeds to the curb wire via intelligent provisioning. Yet another Boston-area start-up (must be something in the water) is Quantum Bridge Communications, which in October announced a multiyear agreement with Comcast to install the company's Passive Optical Networking (PON) technology. PON has the potential to revolutionize optical network provisioning by getting rid of high-cost SONET nodes. PON eliminates the need for active electronics and SONET configurations by using passive splitters/couplers linked to Intelligent Optical Terminals (IOT) on the customer premises.

While SONET rings can serve multiple customers, these rings generally have 4-6 nodes each and must be deployed in anticipation of demand. Quantum's Optical Access Networking approach utilizes PON, an Optical Access Switch, plus the IOT to allow a single pair of fibre strands to fan out to as many as 32 businesses within a 20km radius; the IOTs and OASs are added as demand dictates. According to Quantum, the system can serve 8 PON extensions from a single OAS, providing access to more then 200 customers.

Taking a different approach to the last mile, DSL provider DSL-networks recently introduced its 3PN service (Layer 3 Private Network) which creates what is basically a hybrid LANWAN over a SONET backbone. 3PN connects a company's Layer 3 network to DSL service to securely link remote sites. DSLnetworks maintains datacenter and content aggregation points across the U.S., including facilities in San Francisco, LOS Angeles, Chicago, Dallas, Atlanta, and New York. These facilities are grouped and segmented into Western, Central, and Eastern regions with redundancy between datacenters within each region. The access points are linked together by an OC-12 core backbone, which can be upgraded when added capacity is needed. DSLnetworks maintains multiple regional connections to its CLEC and ILEC partners nationwide, which connect directly into DSLnetworks' regional aggregation points. Customer traffic is then routed to its destination, either onto DSLnetworks' backbone or to the Internet.

Unlike a VPN, the service requires no hardware investment and does not send data over the public Internet (unless that's the destination). DSLnetworks runs and manages the entire thing, offering SLAs and security via IPSec and Triple DES. 3PN is in pilot phase until Q1 2001. It will be available for deployment in more than 80 metropolitan areas nationwide. The company will initially market the service to medium and large businesses and will make the service available to resellers beginning in Q2 2001.

All of these new technologies, architectures, and yet-to-be-announced products point to a huge shift in the way future IP-based services will be distributed, adding intelligence and dynamic provisioning at the user and application level. "Today's network management capabilities give you, say, a frame relay connection, and that's it," says Crescent's Newton. "But users are focused on applications and application-specific subnets. You have to move from just addressing the bits and bytes of a network to a strategy that includes subscribers and services as well." Look for Crescent's products early in 2001.