Parallel computing can basically be divided into two models, the shared memory model where processes share a common address space and the distributed memory model where processes operate in disjoint address spaces and exchange messages to interact with one another. Parallel computers use many point-to-point connections for interprocessor communication or to connect processors to the memory system.

Two distinct types of parallelism can be found in today's parallel computers - Control parallel and Data parallel. Control parallel computers achieve increased performance by taking advantage of parallelism found in the control structure of the program. In these machines each processor executes a portion of the program. Data parallel computers achieve increased performance by taking advantage of the parallelism found in the data of a problem. The machines consist of a single instruction engine and thousands of data processors each having local memory, connected to a communications network over which they may exchange information with other processors.

The development of optical interconnection technology has two aspects. One is substitute optical interconnection network. The other is the combination of optics and electronics. The former needs more effort to approach due to lack of device or too large size to be used at present time. Currently, the latter is used because it is easier to realize according to present condition. Most of these systems use fiber optics for means of communication between processors whereas internally each processor still apply electronic interconnection.

Currently, copper wires are the most cost effective and reliable interconnect in parallel machines. However as machines grow more powerful, wire density becomes critical making fiber possible alternatives because of their small wire size.

Fiber optics are used mainly to exploit its high bandwidth. On a single fiber lots of information/data can be transmitted concurrently and in parallel. Over 1000 high bandwidth (100-200 Mb/s) independant channels or busses can be supported on a single optical fiber. Furthermore multiple buses can co-exist on a single fiber. Fiber links allow a number of high speed serial links to replace a large number of electrical lines. The use of fiber is thus space saving.

The usage of multiple buses are being incorporated in many parallel computers. These systems seek to exploit the high bandwidth available in optical commmunications by supporting multiple virtual buses or selectable channel on a single fiber. Each virtual bus correspond to different assigned channel frequencies. Arbitrary interconnection patterns and machine partitions can be emulated via appropiate channel assignment. Furthermore, simultaneous execution of parallel tasks are also possible. Here Frequency Division Multiplexing(FDM) or Wave Division Multiplexing(WDM) are used to provide multiple channels. Currently, over 1000 high bandwidth (100-200 Mb/s) independant channels can be supported on a single optical fiber...... in the future it is likely to reach 1 Gb/s!

However, most of these machines are still in their experimental stages and are not yet commercially used (except for a few like the Connection Machine below). Below are three types of architechtures which uses fiber optics as multiple busses in their architechture. The first two are Control parallel computer while the third is a Data parallel computer.

The Multiple Channel Architechture (MCA) figure 1, consists of fiber optic dual cabel broadband communication system with processors, memories and I/O devices acting as nodes on the network. Each node is attached to the network via tunable laser transmitters and tunable heterodyne receivers. A passive star coupler is used to evenly divide laser transmitter power to all receivers. By selecting proper frequencies processors, memories and I/Os can arbitrarily be partitioned.

figure 1 : MCA system block diagram

The MCA seeks to exploit the high bandwidth available in fiber optics systems by creating multiple high-speed optical buses that correspond to assigned channel frequencies. It is able to assign approximately 1000 high speed data channels for interconnections among processing elements, memories and I/O devices. This is done by using widely tunable lasers and other optical components. Furthermore, it can adapt to changing topology requirements of parallel programs while supplying large amounts of processing power. The transmission and receiving mechanism are also simplified because the same structure is used (MQW-DBR lasers).

Hybrid Multiprocessing Using WDM Optical Fiber Interconnections is another such system (figure 2). The architechture is almost simmiliar to the MCA except that it uses receivers and transmitters tuned to fixed frequencies rather than the use of tunable components. To reduce contention delays, nodes are assigned to groups. The nodes within a group are assigned to a common transmission channel and have to arbitrate for access to it.

figure 2 : Hybrid Multiprocessing Using WDM Optical Fiber Interconnections

Figure 3 below shows the logical equivalent of a group of processing nodes (as shown in figure 2) if electrical connections were used instead.

figure 3 : The Hybrid system using electrical connections

The Connection Machine figure 4, consist of several front end processors, a dividable block of data processors, high speed graphic display systems and various I/O computers. The connection machine is originally used using copper wires but improved performance was shown when using optical fiber.

figure 4 : Connection Machine System

Parallel processor arrays are also implementing optical fiber in their interconnection system. A system which does so is the optoelectronical hybrid parallel processor array system figure 5.

figure 5 : Optoelectronical hybrid parallel processor array system

The system is composed of six parts: processor arrays and processor elements, optical interconnection network, data acquisition system, global memory, I/O system, and manage computer.