AMOS - Analysis and Modelling of Optical Systems

Project Objectives

(i) To specify a layered set of optoelectronic architectural models, showing the key performance parameters, at three levels of abstraction: system, optical interconnection and physical hardware layers.

(ii) To undertake modelling studies of the optoelectronic interconnect layer, taking into account the optic and electronic hardware, and in particular, the use of smart pixel technology; and validated through physical experiments.

(iii) To study the BSP model of computation and implement it on the optoelectronic architectural system layer with performance parameters determined from (ii) and evaluate the effectiveness of using commodity process-based system using a well chosen case study, namely parallel sorting.

Project Status: Grant ended August 2002. Work continues.

Topology and Reconfiguration of Optically Interconnected Systems

Abstract: This project aimed to find and validate a new design of optical interface (mapping of the emitters and receivers). In order to achieve this new design, we first set a computational model of how a group of links (beams) behaves from one node to another simultaneously. This is the crucial part since it figures out the main issue due to the imaging system the optical highway represents. This step enables us to set the design of the optical interface.

Y. Tissot, Topology and Reconfiguration of Optically Interconnected Systems, School of Engineering and Physical Sciences, Heriot-Watt University, Edinburgh, UK, May 2003.

Download the MSc dissertation in PDF format (2.57MB).

Experimental Investigation Of The Optical Highway Using Large Free Space Optical Components

Abstract: A multi-source laser diode system emitting in the 670nm range is aligned through an optical system to deliver beams through a free space optical configuration consisting of three relay stages. Signals were detected at each stage verifying the validity of the theoretical AMOS model and the nature of the image orientation at each relay stage.

R. Raheem, Y. Tissot, G. A. Russell, C. J. Moir, K. J. Symington and J. F. Snowdon, "Experimental Investigation Of The Optical Highway Using Large Free Space Optical Components", Proceedings of PREP 2003, Exeter University, UK, April 2003.

Download a copy of the conference abstract in PDF format (97.6kB) or a copy of the poster in PDF format (382kB).

Supporting Bulk Synchronous Parallelism with a High Bandwidth Optical Interconnect

Abstract: The list of applications requiring high performance computing resources is constantly growing. The cost of inter-processor communication is critical in determining the performance of massively parallel computing systems for many of these applications. This paper considers the feasibility of a commodity processor based system which uses a free-space optical interconnect. A novel architecture, based on this technology, is presented. Analytical and simulation results based on an implementation of BSP (Bulk Synchronous Parallelism) are presented, indicating that a significant performance enhancement, over architectures using conventional interconnect technology, is possible.

I. Gourlay, P. M. Dew, K. Djemame, J. F. Snowdon and G. Russell, "Supporting Bulk Synchronous Parallelism with a High Bandwidth Optical Interconnect", Special Issue on System Performance Evaluation, Computation and Concurrency: Practice and Experience, 2003.

Journal paper accepted for publication in March 2003. Download the draft article in PDF format (194kB, no figures).

Analysis and Modelling of Optical Systems - AMOS

Abstract: This technical brief examines commodity computer architectures and the flow of data within such systems. It specifically examines I/O bus, RAM and cache architectures with attention paid to real world performance and associated limitations. The case is made for the move from electronic to optical interconnection and a detailed examination is undertaken of core optoelectronic interconnect components. Finally, an architecture is outlined that exploits the advantages of optical interconnection between commodity system called the optical highway. Conclusions are drawn as to the implications of optoelectronics and their potential impact on modern day computer architectures.

K. J. Symington, G. A. Russell, T. Lim and J. F. Snowdon, Analysis and Modelling of Optical Systems, School of Engineering and Physical Sciences, Heriot-Watt University, Edinburgh, UK, February 2003.

Download the technical brief in PDF format (2.7MB).

AMOS - Analysis and Modelling of an Optically Interconnected Commodity Cluster

Abstract: An optically interconnected Beowulf cluster has been modelled considering the optical, optoelectronic, electronic and algorithmic behaviour of the system. A Smart Pixel Array (SPA) layer was considered to alleviate the bandwidth mismatch between the commodity IO bus of the PCs and the high bandwidth in the optics. By using the SPA layer to provide load balancing for the system it has been shown that the optical bandwidth can be utilised.

J. F. Snowdon, G. A. Russell, K. J. Symington, I. Gourlay, P. Dew, "AMOS - Analysis and Modelling of an Optically Interconnected Commodity Cluster", European Optical Society Workshop on Optical Interconnections (IoG), Mannheim, Germany, September 2002.

Download the conference proceedings in PDF format (263kB) or the oral presentation in PDF format (5.66MB).

AMOS - Analysis of an Optically Interconnected Beowulf Cluster

G. A. Russell, K. J. Symington, J. F. Snowdon, I. Goulay, P. Dew, "AMOS - Analysis of an Optically Interconnected Beowulf Cluster", Technical Digest of Photon 2002, Cardiff, September 2002.

Download the conference proceedings in PDF format (123kB) or the oral presentation in PDF format (5.46MB).

Modelling and Simulation of Cache Access to Utilise a High Bandwidth Optical Processor-Memory Bus

Abstract: A planar optical interconnect will be proposed as a method of implementing an interleaved memory bus. Analysis and simulation will show the design of caches can be improved due to the higher bandwidth and connectivity.

G. A. Russell, K. J. Symington, T. Lim and J. F. Snowdon, "Modelling and Simulation of Cache Access to Utilise a High Bandwidth Optical Processor-Memory Bus", Special Issue of Optical Memory and Neural Networks (Information Optics), vol. 11, no. 3, pp. 159-163, Autumn 2002.

Download the publication in PDF format (352kB).

Bulk Synchronous Parallel Computing using a High Bandwidth Optical Interconnect

Abstract: The list of applications requiring high performance computing resources is constantly growing. The cost of inter-processor communication is critical in determining the performance of massively parallel computing systems for many of these applications. This paper considers the feasibility of a commodity processor-based system which uses a free-space optical interconnect. A novel architecture, based on this technology, is presented. Analytical and simulation results based on an implementation of BSP (Bulk Synchronous Parallelism) are presented, indicating that a significant performance enhancement, over architectures using conventional interconnect technology, is possible.

I. Gourlay, P. Dew and K. Djemame, "Bulk Synchronous Parallel Computing using a High Bandwidth Optical Interconnect", IEEE Computer Society Proceedings of PMEO-PDS 2002, Fort Lauderdale, Florida, April 2002.

Download the conference proceedings in PDF format (156kB).

A Reconfigurable Bi-Directional Optical Data Interconnect

Abstract: A 4-node reconfigurable, bi-directional optical highway was constructed showing both the use of polarisation to route signals and liquid crystal plates to change the topology. Large enough signal swings were detected at the output after 1 and 2 relay stages to suggest that such a system is feasible for many stages.

G. A. Russell, K. J. Symington, J. F. Snowdon, I. Gourlay and P. Dew, "A Reconfigurable Bi-Directional Optical Data Interconnect", Proceedings of PREP 2002, University of Nottingham, UK, April 2002.

Download the conference proceedings in PDF format (42kB) or the poster in PDF format (3.4MB).

Supporting Highly Parallel Computing With a High Bandwidth Optical Interconnect

Abstract: The list of applications requiring high performance computing resources is constantly growing. The cost of inter-processor communication is critical in determining the performance of massively parallel computing systems for many of these applications. This paper considers the feasibility of a commodity processor-based system which uses a free-space optical interconnect. A novel architecture, based on this technology, is presented. Analytical and simulation results based on an implementation of BSP (Bulk Synchronous Parallelism) are presented, indicating than a significant performance enhancement, over architectures using conventional interconnect technology, is possible.

I. Gourlay, P. M. Dew, K. Djemame, J. F. Snowdon, G. A. Russell, T. Lim and B. Layet, "Supporting Highly Parallel Computing with a High Bandwidth Optical Interconnect", University of Leeds School of Computing Research Report Series, Report 2001.19, October 2001.

Download the report without figures in PDF format (214kB) or the figures in PDF format (81kB).

AMOS - Analysis and Modelling of Optoelectronic Systems

Abstract: Not available.

G. A. Russell, K. J. Symington, J. F. Snowdon, I. Goulay and P. Dew, "AMOS - Analysis and Modelling of Optoelectronic Systems", OSI Session, Quantum Electronics and Photonics 15, Glasgow, September 2001.

Download the poster presentation in PDF format (1.8MB).

The Analysis of Multiple Buses in a Highly Connected Optical Interconnect

Abstract: Interconnection networks for parallel computers require three properties to be optimised for good performance; bandwidth, latency and scalability. The performance of the interconnection is especially important in multicomputers, such as Beowulf clusters, as interprocessor communication is in the form of message passing. Electronic interconnects are limited in bandwidth by their aspect ratio. Optical interconnection solutions have been suggested. The inherently high bandwidth and connectivity of such a system looks attractive and latency may be lowered due to fewer amplification stages and less routing. The limiting factor on the bandwidth of the optical interconnect is the efficiency of the optic used as the bandwidth is proportional to the power which reaches the detector. To compensate for this it has been suggested that extra physical links can be added for non-local communications to equalise the bandwidth with nearest neighbour links.

G. A. Russell, J. F. Snowdon, T. Lim, J. Casswell, P. Dew and I. Gourlay, "The Analysis of Multiple Buses in a Highly Connected Optical Interconnect", Technical Digest of Quantum Electronics and Photonics 15, Glasgow, pp. 75, September 2001.

Download the conference proceedings in PDF format (35kB) or the poster presentation in PDF format (1.5MB).

Modelling of Optical Interconnects for Parallel Processing

Abstract: By linking an abstract model of parallel computing with the physical parameters of an optically interconnected PC cluster it is shown that the high connectivity in the optical domain can reduce routing and increase performance in clusters of more than 55 machines for reduction and 8 for a one-to-all broadcast.

G. A. Russell , J. F. Snowdon, T. Lim, I. Gourlay and P. M. Dew, "Modelling of Optical Interconnects for Parallel Processing", Proceedings of Postgraduate Research in Electronics, Photonics, Communications and Software (PREP 2001), University of Keele, UK, pp. 29-30, April 2001.

Download the conference proceedings in PDF format (225kB) or the oral presentation in PDF format (8.4MB).

Comparison of Two Approaches for Implementing Free-Space Optical Interconnection Networks

Abstract: A particular design choice in the implementation of free-space optical interconnection networks (e.g. photonic backplanes) based on cascaded image-relay lenses is investigated. In these systems, a communication link can be implemented either by a single hop between source and destination nodes with the signal remaining in the optical domain through many image-relay stages, or by multiple hops between adjacent nodes with the signal undergoing optical-electrical conversion and vice versa at intermediate nodes (which act as repeaters). These two approaches place different demands on the optical system and the optoelectronic interface. We compare the raw bandwidth-per-link available in two example networks (the mesh and the completely connected network) using a model of the bandwidth and power consumption of an optoelectronic data channel and considerations on the aggregate bandwidth of the optoelectronic interface chip. We find that the single-hop approach provides a higher bandwidth-per-link. For example, the single-hop bandwidth-per-link is three times greater than the multiple-hop value for a mesh network of 49 nodes and for a completely connected network of 13 nodes. The advantage can increase further as the network size grows. The methodology is also applicable to the investigation of other implementation choices in optoelectronic interconnects.

B. Layet and J. F. Snowdon, "Comparison of Two Approaches for Implementing Free-Space Optical Interconnection Networks", Optics Communications, vol. 189, no. 1-3, pp. 39-46, March 2001.

Download the journal paper in PDF format (131kB).

Smart Optoelectronic Networks for Multiprocessors

Abstract: An intelligent interconnection network with fine-grain parallelism is described that has the potential to support efficient, scalable algorithms running on associated coarse-grain processors. The approach is relevant to the emerging computational cluster systems. The support of concurrent operations within the network is discussed and their mapping onto smart-pixel array network interfaces is shown. Choices are considered in the design of the free-space optical interconnect that enables the inter-smart-pixel-array communication. In particular, a system that uses multi-element macro-lenses is studied and results of detailed modeling are given that quantify the smart-pixel density. These results are used, in an illustrative case study of the sorting problem, to compare potential system architectures. This is work in progress and throughout the paper, important issues in the design and use of the intelligent interconnect are raised that require more study.

B. Layet, I. Gourlay, P. M. Dew and J. F. Snowdon, "Smart Optoelectronic Networks for Multiprocessors", Technical Digest of OSA Optics in Computing, Quebec City, Canada, pp. 278-288, June 2000.

Download the conference proceedings in PDF format (234kB) or the poster presentation in PDF format (306kB).