Parallel Optical Link Organization

Excerpts from the video can be accessed by clicking on the sample images. These highlights range from 980 kB to 4.2 MB and are 15 to 45 seconds long.

The parallel optical link organization is an ARPA sponsored consortium consisting of Hewlett-Packard, AMP, Dupont, the University of Southern California and SDL. The consortium is developing low cost, high-performance parallel optical interconnect modules for workstation clusters, multimedia, and high-speed switching systems. This presentation describes POLO Technologies and technologies at the University of Southern California.
POLO leverages the performance of Gb/s fiber-optic technology and multiplies it by ten using fiber ribbon. Ribbon fiber from Siecor is used for parallel data links over distances of a few meters to a kilometer. The technology migration path evolves from telephone switches to super computers. USC is leap-frogging past these early applications and addressing future markets for advanced scalable servers and workstation clusters.Click here to view a video clip (2.49 MB)

The Hewlett Packard POLO module has ten Vertical Cavity Surface Emitting Lasers and ten receivers. The module uses Polyguide technology developed by DuPont and manufactured by AMP. AMP is also developing the Polyguide-to-fiber array connection. The assembled POLO-module is a low-cost multi-Gbs physical layer for advanced networks. Network technologies developed at USC include the high-performance CMOS Link Adapter Chip needed to interface between the POLO-module and the Host computer bus adpater. USC will also make the complete POLO Network Interface Card. The LA Surfboard will be used to attach the network to workstations.

The Chip is integrated into the LA Surfboard and connects between the workstation and the POLO module. The LA Surfboard advances network technology beyond experimental networks such as AfterBurner and JetStream developed by HP Bristol Labs. With node latencies of less than 100 ns, an efficient Media Access Controller, VCI, and ATM compatibility, the LA Chip creates a new type of high-performance network for multimedia applications.

Advanced CMOS ciruits in the LA Chip require sophisticated test and measurement capability. At USC we make use of mulitGb/s electrical packaging, test pattern and error rate testing to develop the LA Chip.

Click here to view a video clip (978 kB)   


An initial demonstration uses two series 700 workstations and a POLO-1 module. Each workstation contains a JetStream newtork interface. This board connects to the main processor and core I/O of the workstation via a standard graphics interface bus. Both transmit and receive channels of JetStream have differential one Gb/s signals.

Click here to view a video clip (4.19 MB)

All of the interfaces have been combined into a POLO-1 module to produce a networked data flow of 4 Gb/s. In our experiments we send data over 500 meters of ultra low-skew ribbon fiber. Image format data is passed over the network and visualized on one of the workstations.

Measurements of application-to-application throughput as a function of message size show JetStream is capable of one hundred times greater sustained throughput compared to ethernet and ten times higher throughput compared to a FDDI network. With the new link adapter chip from USC the POLO network is projected to deliver an additional order-of-magnitude improvement to network throughput.

A dramatic difference in Quality of Service can be seen by passing high-resolution images over a JetStream network using the POLO module. In this scene the impact on delivery of high resolution images is compared. On the left you see a rotating volume-rendered medial image being served between two workstations. On the right you see the same image with a much lower level quality of service over the Ethernet network.Click here to view a video clip (1.30 MB)
In a more advanced application we demonstrate remote visualization and distributed computation using a network of three workstations. The graphical user interface has been designed to view a volume rendered medical image from different angles. One machine on the network serves this image to the remote display. Another machine calculates the volume slice allowing the user to interactively view internal structures of the rendered image.