Rainbow 2 all-optical network exploits ?ber speed prowess

Rainbow 2 all-optical network exploits ?ber speed prowess

George Kotelly

Researchers from the IBM research division and its development partner Los Alamos National Laboratory demonstrated IBM`s Rainbow 2 next-generation, all-optical network at Supercomputing `94. Linked by commercial-grade singlemode fiber-optic cables, the prototype network comprised three computing nodes, each running at 1 gigabit per second.

The all-optical network aims at meeting the high-speed demands of scientific, medical, academic and research facilities that work with gigabit data throughput rates. It is also facilitating the development of powerful data-handling applications, such as those dealing with multimedia, full-motion video and supercomputing imaging.

"IBM is exploiting the performance potential of fiber-optic networking with the Rainbow 2 network prototype," says Paul Green Jr., manager for the advanced optical networking group at IBM`s Thomas J. Watson Research Center in Hawthorne, NY. "Rainbow 2 is positioned at the development forefront of a new era of supercomputing networking and is opening the door to the next generation of optical fiber technologies," he adds.

The Rainbow 2 optical network architecture implements an array of existing and new optical technologies that exploit the capabilities of optical fiber:

Wavelength-division multiplexing supports 32 wavelength channels on one fiber. Each network node transmits data at a preassigned wavelength and requires no switching. Data is broadcast and available to all network nodes.

A circular search protocol matches a transmission to its destination node. Each node is assigned one of 32 wavelengths and an address-identification code for transmitting data. Inactive nodes continually scan all network frequencies for broadcast messages and establish a connection with the sending node upon detecting its identification code.

The high-performance parallel interface protocol is employed for point-to-point communications among supercomputers and peripheral devices. In addition, Rainbow 2`s optical hardware adds flexibility to Hippi by extending the network`s range without sacrificing speed. When used in electronic-based networks, Hippi has a 25-meter length limitation before extenders, which slow down the system, are needed. Optical fiber allows the interconnection of computers and peripherals over 5 to 10 kilometers.

A protocol off-load technique accelerates the processing speeds of the connected computers and supercomputers. It overcomes the transmission roadblocks encountered because the optical network can function faster than the interconnected computers.

A simple host intersocket protocol takes advantage of increased bandwidth to off-load and implement network protocols efficiently at the node, thus enabling applications to run at gigabit speeds. The Ship protocol was developed jointly by IBM, Intel and Los Alamos National Laboratory.

An optical network adapter box eases the node connection to the network by handling the processing and protocol off-load functions. It resembles an external modem that has a fiber connection at one end and a Hippi connection at the other end. Electronic switches are used only at the end nodes.

Rainbow 2 shines

According to Green, "The historical significance of Rainbow 2 concerns two primary optical network advances." The first one deals with real-world applications, distance extensions and dynamic switching in supercomputer interconnections. It provides supercomputing users with distance extensions to 5 or more kilometers if fewer than 32 nodes are used. Dynamic switching allows a supercomputer to change connections with high-end workstations in just a few milliseconds.

"The second key advancement," says Green, "centers on the protocol off-load technique using Ship software. This software has made Rainbow 2 much more reliable than Rainbow 1."

The next step, predicts Green, is Rainbow 3, which will target packet switching and rapidly tunable filters. It will improve on Rainbow 2`s circuit switching because the receiver cannot be retuned any quicker than a few tens of milliseconds.

Looking over the rainbow

Approximately a year after developing the Rainbow 1 all-lightwave network (see Lightwave, March 1991, page 1), IBM realized that it had an advanced optical network solution in search of an application. In looking for network applications, IBM researchers met with engineers from the Los Alamos National Laboratory--the leading center for very-high-speed communications work. Laboratory investigators have extensive experience in gigabit connections among terminals, supercomputers and networks. In fact, the Hippi protocol was developed at the Los Alamos National Laboratory. At meetings, both sides expressed mutual interest in expanding their abilities in optical networking.

According to Green, "To determine real-world optical-network applications, IBM preferred working with end users, as represented by the laboratory, rather than with other computer manufacturers."

The ensuing partnership, which is partially sponsored by the Department of Energy, agreed that IBM would structure the test site and supply the necessary hardware. For its part, Los Alamos would determine what supercomputing connectivity problems matched all-optical networking solutions and would help develop the needed software.

Rainbow 1 was developed as a totally protocol transparent network. On the other hand, Rainbow 2 is designed specifically to meet the industry standard for gigabit communications, namely Hippi.

Explains Green, "Fiber-optic technology delivers bit streams into computers at a blinding rate. But there the data dies in the software path links, the number of execution cycles per message exchanged. In these links, the number of cycles executed per exchange, even in the transmission control protocol, runs into thousands and can bring down the network."

For many network applications, users typically have generated and invested in huge amounts of code that they do not want to change. The Ship intersocket interface allows users to implement their existing code.

The demonstration

At Supercomputing `94 in Washington, DC, last November, three interconnected network nodes were used; however, Rainbow 2 can accommodate 32 nodes. Two nodes contained IBM RS-6000 Unix workstations with plug-in cards that connected with the Hippi protocol. The third node used an IBM SP/2 high-end multiprocessing server.

The three computers were connected to a central star coupler in a regular broadcast network configuration via fiber-optic cables. Rainbow 2 is not designed to be a long-range network. It runs at 1 Gbit/sec, but, at that speed, it has a range of 5 km.

The demonstrated applications involved the transmission and reception of voice and pictorial information at extremely high resolution and rapidly refreshed motion video--a process known as supercomputer visualization.

Says Green, "Fewer workstations and shorter link distances translate into higher bit rates. The wrinkle here is that the distributed feedback lasers used are good only up to 1 Gbit/sec. However, using more expensive lasers with better isolators could achieve 2.4-Gbit/sec speeds."

Computer networks are usually characterized by multiple bit streams at medium rates. Most computer end users do not need to send information at more than a few hundred megabits per user. Only supercomputer applications, such as those performed at the Los Alamos National Laboratory, require 1-Gbit/sec rates. q

More in DWDM & ROADM