Project HORNET begins second phase of next-generation Internet research

Th 1802lwe109

Stanford University's optical-networking research remains bright with new funding for developing optoelectronic packet switching for the metro area.

IAN WHITE, Stanford University

Project HORNET (Hybrid Optoelectronic Ring Network) is a research project conducted by the Stanford University Optical Communications Research Lab or atory (OCRL). The project is making a case for constructing an optoelectronic packet-switching metropolitan-area network (MAN) and how it may be advantageous for providers to construct such a network (see Lightwave, July 2000, page 1). The project's first phase has since concluded, and a new phase of the project is underway.

HORNET was exciting enough to grab the attention of the Next Generation Internet (NGI) program, a consortium overseen by the Defense Advanced Research Projects Agency (DARPA). HORNET will now be funded under the NGI project through the end of the U.S. government's support of NGI.

The pursuit to stay ahead of the demand for bandwidth led many researchers toward increasing the amount of information one fiber-optic link can transmit. More recently, however, some researchers have begun complementing that research by developing and investigating new architectures and protocols to allow optical networks not only to support more bandwidth, but also to use it intelligently.

Project HORNET and others are doing exactly that within the NGI program, focusing on new technologies for the metropolitan area. The HORNET architecture uses widely tunable semiconductor lasers in the transmitter of each node in combination with a wavelength-routed ring network that allows all nodes on the ring to exchange packets in a fully meshed architecture. This configuration is in contrast to the static, logically star-based architecture used by most conventional metro networks.

The HORNET architecture distributes intelligence at the optical layer to all the nodes on the ring network. Conventional ring networks used in the metro area are generally based on a static, or slowly dynamic, hub-and-spoke configuration (star) where the point of presence (PoP) is the hub on the ring and each wavelength routes traffic from the PoP to a particular node, like spokes. Each node on the ring has a wavelength drop and single wavelength optical transmitter. This configuration seems logical because conventional MANs are used as distribution networks in which the PoP collects traffic from the nodes and sends it to the long-haul network, while conversely distributing the traffic it receives from the long-haul network to the nodes.Th 1802lwe109

Researchers at Stanford University's Optical Communications Research Laboratory set up an experiment to characterize the performance of the tunable transmitter and fast bit-synchronizing receiver.

However, OCRL researchers predict that conditions placed on metro networks will change significantly over the next few years. The networking community is pushing for caching information throughout the Web, instead of storing it on a centralized server. Secondly, there is a movement to develop programs that run over the Web in a distributed manner (e.g., Napster). On top of those two trends is the increase of packet-based traffic that will occur in the metro area due to new wireless Internet Protocol (IP) devices as well as voice over IP.

The expected result of the sum of these Web trends is that packet-based communications within the metro network will increase, causing metro networks to be more than just distribution networks. They will need to operate with more of a fully connected, intelligent architecture that can transmit packets directly from source node to destination node. Conventional metro-ring networks are not built for that, but the HORNET architecture is.

The HORNET project began in mid-1998 with financial support and guidance from Sprint's Advanced Technology Laboratories (ATL-Burlingame, CA). Stanford's OCRL has been collaborating with the ATL since the mid-1990s and will continue to do so for the foreseeable future. The principal investigator of the HORNET project is Leonid Kazovsky, a Stanford professor for 10 years and founder of Stanford's OCRL. Prof. Kazovsky is also known for his years of pioneering research at Bellcore (now Telcordia Technologies-Morristown, NJ) in the 1980s and more recently for founding metro startup Alidian Networks (Mountain View, CA).

The HORNET project is much more than just a newly proposed architecture. The project also aims to develop all new technologies necessary for the architecture's implementation. The two most interesting new technologies investigated were a fast tunable optical transmitter and a fast bit-synchronizing packet receiver. The tunable transmitter is used to enable a node to transmit on any network wavelength it chooses, allowing it to transmit a packet directly (at the optical layer) to any node on the network.

It is crucial that the tunable transmitter has the ability to tune from one wavelength to another almost instantly, because the transmitter will tune between consecutive packet transmissions. On OC-48 (2.5-Gbit/sec) and OC-192 (10-Gbit/sec) networks, typical IP packets can be less than 1 microsec. To keep overhead low, the transmitter must have a tuning duration that is very small compared to a packet duration (i.e., only a few nanoseconds). Using currently available tunable semiconductor lasers, the OCRL designed and constructed a fast-tunable transmitter that does exactly that.

A serious technical issue arises in a node's receiver with the use of the tunable transmitter. The HORNET architecture allows a transmitter to send consecutive packets to two different receiving nodes. By the same token, a receiving node can receive two consecutive packets from a different transmitting node. This situation does not arise in conventional networks because that architecture requires the transmitting node and receiving node to maintain a permanent optical connection. The nodes synchronize with each other and then use phase-locking circuits to stay synchronized (synchronization refers to the receiver matching its sampling clock to the bit stream entering it).

In the HORNET architecture, this "bit-synchronization" must take place at the beginning of every received packet. Conventional technology allows the receiver to synchronize in a microsecond or so-far too long, considering the duration of a packet may be less than 1 microsec. Therefore, it is imperative to establish synchronization in only a few nanoseconds. Currently, the HORNET project uses an analog radio-frequency (RF) technique to quickly extract the sampling clock from incoming packets.

The experimental demonstration of these two new technologies as well as all other technologies developed for HORNET has been the central focus of the research. A large proportion of that research went into developing and demonstrating the fast-tunable transmitter, because it is the most crucial component of the architecture.

Researchers Kapil Shrikhande and Matt Rogge were able to construct and demonstrate a transmitter that could tune throughout the conventional telecommunications band (C-band), with a maximum tuning duration of approximately 15 nsec. The transmitter consists of a simple programmable-logic processor, two digital-to-analog converters (DACs), and a tunable semiconductor laser from Altitun Inc. (Irvine, CA).

The processor selects the wavelength and sends 10-bit digital values to the DACs, and the DACs convert the digital values to the appropriate electronic currents to tune the laser. Ultimately, the laser-tuning time can be decreased, if the package of the laser is optimized for high-speed currents at the tuning inputs.Th 1802lwe108

Tunable transmitter used in HORNET's experiments, including an Altitun tunable laser and a tuning controller printed circuit board consisting of a programmable processor and digital-to-analog converters.

To evaluate the performance of the fast-tunable-transmitter subsystem, bit-error-rate measurements were conducted by placing the transmitter back-to-back with the fast bit-synchronizing packet receiver with a wavelength filter between the two. The tunable transmitter was programmed to tune through a series of wavelengths, transmitting a packet between tuning events. The wavelength filter matched one of the wavelengths in the series, allowing the packet receiver to receive the packets transmitted on that wavelength. The results of the experiments showed that the transmitter and receiver subsystems had excellent error-rate performances and that very little penalty was incurred when changing the system from a fixed-wavelength, synchronous transmission system to a system with a fast-tuning transmitter and a packet receiver accepting packets asynchronously.

One outstanding conclusion from the experimental results is that the semiconductor laser exhibited no problems of tuning to the incorrect wavelength or of hopping out of the correct wavelength after reaching the correct wavelength. That means no feedback control system is required to aid the tunable laser in finding and holding the correct output wavelength for a packet transmission.

The success of Project HORNET's first two years led to the official funding from NGI during the summer of 2000. Inclusion in the NGI program gives HORNET a tremendous boost because of the strong funding and especially because of the wealth of knowledge and research in the NGI community.

This system of knowledge exchange will have a positive impact on the success of HORNET, because there are other NGI projects that have a similar nature to HORNET. A prime example is the research conducted by the University of California at Santa Barbara (USCB) led by Prof. Daniel Blumenthal (more information can be found at http://ocpn.ece.ucsb.edu/darpangi/darpapage.html). The USCB research group has built and demonstrated an all-optical packet switch for ultra-high-capacity networks. The packet switch uses tunable lasers and all-optical wavelength conversion as a means for switching packets without converting them to electronic form.

Another project investigating the use of tunable transmitters is the Helios project (see http://www.anr.mcnc.org/projects/Helios/Helios.html), led by Dr. Daniel Stevenson at the Microelectronics Center of North Carolina. Telcordia is conducting two interesting projects that are experimentally investigating IP-over-WDM transmission over high-capacity long-haul networks. One project is led by Dr. Gee-Kung Chang (http://govt.argreenhouse.com/ngi) and another by Dr. Nim Cheung (http://govt.argreen house.com/BLTN). Additionally, one of the newest NGI projects, led by Prof. Ben Yoo at the University of California at Davis, will be investigating similar issues, as well (see http://sierra.ece.ucdavis.edu).

The NGI funding will allow OCRL researchers to make tremendous steps in the HORNET project. There is a relatively long list of items OCRL researchers believe are imperative to accomplish for HORNET to be a success. These items include necessary enhancements and analysis of HORNET's media access control (MAC) protocol, further results from the fast-tunable transmitter, more advanced receiver bit-synchronization techniques, a survivability architecture and protocol, and development of operations, administration, maintenance, and provision (OAM&P) details.

The original design of the HORNET network bundled IP packets into a short, fixed-sized cell, similar to IP over ATM. This assumption simplified the design of the architecture, particularly for the MAC protocol. But recently, the networking community has lobbied against the use of IP over ATM because of the excessive overhead caused by forcing IP packets into the fixed-sized cell (known as the ATM cell tax). Therefore, OCRL researchers are looking for a low-overhead way to deal with the variable-sized packet nature of IP. The best-case solution will not divide or group the IP packets, but rather transmit them around the ring in their natural Layer 3 form. Ultimately, the complexity of such a scheme will be weighed against the overhead of the fixed-sized cell scheme to determine the best architecture.

The MAC protocol is not the only HORNET technology that can benefit from improvements; OCRL researchers believe a lot of progress can be made on the fast bit-synchronizing receiver. As mentioned, the receiver in a node on the HORNET network must have the ability to synchronize itself with the incoming packet in almost negligible time. Currently, the HORNET experiments have been using analog RF techniques to quickly recover the sampling clock from the incoming packet. These techniques have proven to be a bit slow and cause too much of a power penalty, but they are very inexpensive and simple to implement.

Nonetheless, the OCRL would like to investigate other receiver designs-particularly digital designs-that will almost instantly recover the synchronization without causing a large power penalty and without being too complicated or expensive. A graduate student at Stanford created one design with very-large-scale integration. This research, however, went largely unnoticed by the optical communications community. OCRL researchers hope to experiment with this technique in the HORNET project. In addition, the OCRL will be studying other proposed designs such as that created in the Telcordia project mentioned earlier.

Historically, one of the most overlooked technologies by next-generation network researchers has been the survivability of the network. But due to the dependence of commerce on the Internet, any network deployed today must be resilient to accidental cuts, equipment failures, and breaks due to maintenance. With that in mind, the HORNET project would not be complete without demonstrating a survivability architecture and protocol.

HORNET could potentially use a typical architecture and protocol similar to those being discussed and proposed for SONET rings, such as 2-fiber unidirectional line-switched ring (2FULSR), and 4-fiber bidirectional path-switched ring (4FBPSR). However, OCRL researchers are not satisfied with the amount of overhead associated with these architectures. In general, to be fully survivable, it is assumed that the carrier will need to bury twice the bandwidth the network will use.

On the contrary, an architecture has been developed for HORNET that deploys the same amount of bandwidth but uses twice as much during regular operation. In the case of a cut or maintenance, many of the nodes on the ring remain nearly unaffected, while some may suffer a cut in accessible bandwidth that will take them down no lower than the level at which conventional survivability architectures operate. The architecture is unproven as of yet, but it is expected that its advantages will be proven before the end of the HORNET project.

The development and implementation of these and other technologies in the HORNET project have the potential to make a significant impact on future generations of the Internet. The architecture and protocols developed in the project may or may not be deployed; however, it can be certain that many of the technologies developed will be implemented commercially within the next decade. At the very least, expect to see fast-tunable transmitters, fast-synchronizing packet receivers, and MAC protocols similar to those developed for HORNET.

When Project HORNET began in 1998, OCRL researchers were only trying to determine if some of the technologies were feasible. By 2002, the telecommunications industry should be anxiously awaiting many of the exciting technologies developed and maybe even preparing for the deployment of the next generation of metropolitan networks for the Internet.

Ian White is a Ph.D. student in the electrical engineering department at Stanford University and a member of Stanford's Optical Communications Research Laboratory. He can be reached at ianwhite@stanford.edu. More information on Project HORNET can be found at http://ocrl.stanford.edu/hornet/hornet.html.

More in Packet Transport