Terabit optical local area networks for multiprocessing systems

Area of Science:

• Computer Engineering
• Optical Networking
• High-Performance Computing

Background:

• Multiprocessing systems require high-bandwidth, scalable interconnects.
• Existing local area networks (LANs) face limitations in aggregate data throughput for demanding applications.
• Transitioning from electrical to optical interconnects is crucial for future network performance.

Purpose of the Study:

• To design and describe a scalable optical local area network (LAN) architecture for multiprocessing systems.
• To demonstrate a centralized switch core using complementary metal-oxide silicon (CMOS) technology.
• To explore the scalability of the proposed architecture towards terabit capacities.

Main Methods:

• Utilizing parallel-fiber-ribbon optical links for workstation connectivity.
• Implementing a centralized CMOS switch core on a printed circuit board (PCB).
• Employing a broadcast-and-select architecture with parallel CMOS integrated circuits (ICs).
• Fabricating and describing a prototype optoelectronic switch core.

Main Results:

• Each workstation achieves a data bandwidth of 6.4 Gbits/s using Motorola Optobus fiber technology.
• A 32-workstation switch core supports an aggregate data bandwidth of 204 Gbits/s.
• The architecture demonstrates scalability through CMOS optoelectronic ICs with optical input-output, enabling single-chip terabit capacities.

Conclusions:

• The proposed optical LAN architecture offers a scalable solution for multiprocessing systems.
• The design leverages commercially available parallel fiber technology and established network markets.
• The architecture provides a smooth transition path from electrical to optical domains, adapting to technological advancements.