By Topic

Massively Parallel Processing Using Optical Interconnections, 1994., Proceedings of the First International Workshop on

Date 26-27 April 1994

Filter Results

Displaying Results 1 - 25 of 33
  • First International Workshop on Massively Parallel Processing Using Optical Interconnections

    Save to Project icon | Request Permissions | PDF file iconPDF (35 KB)  
    Freely Available from IEEE
  • Parallel optoelectronic multiplexer architectures

    Page(s): 252 - 254
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (204 KB)  

    Parallel optoelectronic multiplexers benefit from the large number of I/O channels available in free-space optics. The comparative advantage of parallel over serial I/O is less power dissipation per unit area and less self-inductance noise, which lead to higher performance and higher reliability, respectively. The penalty is increased communication latency, but for a given message size, the latency can be hidden with the proper choice of multiplexer architecture and optoelectronic technology View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optoelectronic parallel processing with straight-pass optical interconnections and smart pixel arrays

    Page(s): 170 - 181
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (824 KB)  

    Several fine and medium grain parallel computer architectures comprise multiple stages of 2D processing element arrays. The parallel massive interconnections between two such stages can be implemented in optics. Straight-pass interconnections, as one possible interconnection scheme, can be easily realized with a lens or a lenslet array and their simplicity and regularity permits easy scale-up. We have identified four application classes with operations that can be performed in parallel by straight pass interconnections between smart pixel arrays. These classes include numerical computations such as data comparison and sorting, associative processing, database operations in a relational database environment, and image processing operations. We propose several implementations which use vertical-cavity surface emitting lasers as emitters and heterostructure phototransistors as detectors View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Considerations for optoelectronic shared cache parallel computers

    Page(s): 241 - 251
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (712 KB)  

    Discusses the selection and use of optoelectronic devices in parallel computers to increase the effective processing rate. The coherence problem of keeping consistent copies of data in all caches is resolved by using shared optoelectronic cache memories. The authors investigate the architecture, data and signal modulation, memory management, and its physical implementation. The proposed architecture uses optics for interconnections and electronics for data processing. Thus, it provides high speed and true parallelism View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Order statistics on optically interconnected multiprocessor systems

    Page(s): 162 - 169
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (652 KB)  

    Processor arrays with an optical bus are introduced as a new computational model. We use the order statistics problem as an example to demonstrate how to design efficient parallel algorithms on such systems. Besides proposing a new algorithm on the model, some basic data movement operations involved in the algorithm are discussed. We believe that these operations can be used to design other parallel algorithms on the same model. Time analysis indicates that order statistics can be done more efficiently on a linear array with a pipelined optical bus than that with an electronic bus. The initial result is promising, and the author believes that optical buses may emerge as a powerful communication mechanism for connecting large parallel computer systems in the near future View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Holographic optical interconnections

    Page(s): 237 - 240
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (256 KB)  

    Introduces a new type of planar optical interconnect, the transverse holographic waveguides. With this type of interconnect, one dimensional, input light distribution is converted into a one dimensional output light distribution via holographic patterning along the direction of the optical wave propagation View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A hypercube-based optical interconnection network: a solution to the scalability requirements for massively parallel computers

    Page(s): 81 - 93
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (984 KB)  

    An important issue in the design of interconnection networks for massively parallel computers is scalability. Size-scalability refers to the property that the number of nodes in the network can be increased with negligible effect on the existing configuration and generation-scalability implies that the communication capabilities of a network should be large enough to support the evolution of processing elements through generations. The lack of size-scalability has limited the use of certain types of interconnection networks (e.g. hypercube) in the area of massively parallel computing. The authors present a new optical interconnection network, called an Optical Multi-Mesh Hypercube (OMMH), which is both size- and generation-scalable while combining positive features of both the hypercube (small diameter, high connectivity, symmetry, simple routing, and fault tolerance) and the mesh (constant node degree and scalability) networks. Also presented is a three-dimensional optical implementation methodology of the OMMH network View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Design considerations and algorithms for partitioning opto-electronic multichip modules

    Page(s): 59 - 69
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (860 KB)  

    There is considerable interest in developing optical interconnects for multi-chip modules (MCM). As a consequence, there is a basic need in developing a methodology for partitioning the system for effective utilization of the optical and electronic technologies. For the given netlist of a system design, key question to be answered is where to use optical interconnections. The authors introduce the computer aided design (CAD) approach for partitioning opto-electronic systems into opto-electronic multichip modules (OE MCM). They first discuss the design tradeoff issues in optoelectronic system design including speed, power dissipation, area and diffraction limits for free space optics. They then define a formulation for OE MCM partitioning and describe new algorithms for optimizing this partitioning based on the minimization of the power dissipation. The models for the algorithms are discussed in detail and an example of a multistage interconnect network is given. Different results, with the number and size of chips being variable, are presented where improvement for the system packaging has been observed when the partitioning algorithms are applied View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Reconfigurable optical interconnects for computer vision applications

    Page(s): 224 - 236
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (840 KB)  

    Evaluates the advantages of reconfigurable optical interconnects within massively parallel systems due to their ability to provide versatile application-dependent network configurations. Furthermore, they are being considered as alternatives to electronic interconnects within high-performance computers because of their advantages of high bandwidth, low wire density and low power requirement at high data rates. Fiber optic interconnects based on wavelength division multiplexing and free-space holographic interconnects are two classes of optical interconnects that can support network reconfiguration. Using computer vision applications, the authors compare these two classes of optical interconnects with electronic interconnects taking into account the combined effects of link speeds, link latencies, system size, message size and network topologies feasible with current implementation capabilities View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Free space optical message routing for high performance parallel computers

    Page(s): 37 - 44
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (596 KB)  

    We survey various electrooptical message routing systems for sending N messages between N processors and discuss the theory and practice of these systems. In particular, we compare these proposed systems with respect to various metrics including time, space, number of switches, bandwidth, energy, as well as estimates to scalability and cost in current technology. There are two classes of interconnection networks for parallel computers: multistage networks and single stage networks. Optical multistage networks are often the optical realization of conventional multistage electronic networks. Optical single stage networks use free space optical routing techniques to achieve a virtual crossbar. We describe various optical single stage networks based on some quite diverse techniques including matrix-vector multiplication, various dynamic and static holographic methods, as well as frequency multiplexing View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Photonic architectures for distributed shared memory multiprocessors

    Page(s): 151 - 161
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (904 KB)  

    This paper studies the interaction between the access protocol used to provide arbitration for a wavelength-division multiple access photonic network and the cache coherence protocol required to support a distributed shared memory environment. The architecture is based on wavelength division multiplexing which enables multiple multi-access channels to be realized on a single optical fiber. Larger blocks are supported to reduce the per bit overhead and increase the exploitation of spatial locality, while false sharing is reduced through a mechanism to provide a finer granularity of invalidation. Two main approaches have been considered to harness the enormous available bandwidth of WDMA optical networks: reservation (control-channel based) or pre-allocation media access protocols. This paper extends the function of a control channel to include broadcast support of cache-level control information, in addition to its primary role of data channel reservation, thereby enabling a snooping based coherence protocol to be considered. Larger snooping-based multiprocessors may be possible with this approach. Two major scenarios are considered through trace-based discrete-event simulation in this paper: a system with a directory based cache coherence protocol and a pre-allocation based WDMA access protocol is compared to a system with a snooping based cache coherence protocol and a reservation based WDMA access protocol View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Fan in loss in electrical and optical circuits and systems

    Page(s): 197 - 201
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (436 KB)  

    Analogies between the fan-in limitations of linear electrical circuits and linear optical circuits are explored. In both cases, the second law of thermodynamics sets the limits as to the amount of power that can be coupled from multiple inputs into a single output. The limitation applies to the power coupled into a single (properly defined) output “mode” from a set of input modes. While fan-in limitations of optics have generally been stated in terms of power coupled into a single outgoing spatial mode from incoming spatial modes, a more general view makes use of temporal and polarization modes as well. It is then possible to show that the usual fan-in restrictions stated with respect to only spatial modes can be violated, while the more general restriction in terms of space-time-polarization modes remains satisfied. The results have direct application to wavelength multiplexed optical interconnects View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Performance evaluation of optical systems for multistage interconnection networks

    Page(s): 213 - 223
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (676 KB)  

    Some guidelines of the performance evaluation of an all-optical architecture are presented. This basic architecture supports bursty traffic. It can be used as the switching building block far a number of interconnection networks. The power analysis of an all optical buffering architecture that can be interfaced with such an element is discussed. The performance evaluation is given in terms of the switch model, the multistage model and the power performance of the optical buffer. The architecture uses bistable optical devices such as Fabry-Perot etalons, SEED and S-SEED. A number of important issues remain to be addressed, such as the use of different nonlinear optical devices, synchronization, and modeling requirements View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A fine-grain, high-throughput architecture using through-wafer optical interconnect

    Page(s): 27 - 36
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (904 KB)  

    The author present a highly parallel, three-dimensionally interconnected system to process high-throughput stream data such as images. Optical interconnect at wavelengths to which silicon is transparent is used to create the 3D system. Thin film InP/InGaAsP-based emitters and detectors operating at 1.3 microns are bonded to the silicon circuitry, and emit through the silicon wafer to create the vertical optical interconnect. Foundry-fabricated Si circuits are post processed using standard, low cost, high yield microfabrication techniques to integrate the thin film devices with the circuits. In order to meet off-chip I/O requirements, a high-bandwidth, three-dimensional optical network is also being designed. Using through-wafer optical interconnect, a new offset cube topology has been created, and naming and routing schemes have been developed. Its performance is comparable to that of a three-dimensional mesh. A processing architecture has also been defined that minimizes overhead for basic parallel operations. A complete processing node for high-throughput, low-memory applications can be implemented using a fraction of a chip View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optical, processing-less massively parallel processing interconnection networks

    Page(s): 323 - 330
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (72 KB)  

    The authors attempt to form a coherent picture that may suggest possible future directions of using the benefits optics has to offer, yet avoiding its drawbacks for optical interconnection networks. Possible questions addressed include: What is the nature of (some) parallel applications that may be used to avoid optical processing in the network? How can reconfiguration be used in an efficient way to permit using processing-less optical communication? What are the devices issues that need to be addressed? What opto-electronic devices can be used? What is the expected performance of such devices (density, bandwidth)? What are the connection needs to the electronic processing elements? What are the aspects of packaging? Will a distributed layout be better than a centralized one (i.e., using one central device vs. multiple devices)? How will such a reconfigurable system be used in a real environment (multi-users, partitioning of jobs, distribution of global data)? What applications can benefit from such an architecture and how general purpose can it be (business applications: databases, transaction processing; scientific applications; engineering applications; future applications)? View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optical interconnection system using stacked thin film modulator and detector

    Page(s): 140 - 150
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (604 KB)  

    Proposes a new optical connection system using thin film modulators and detectors on top of silicon VLSI chips. The advantage of this system is that the entire chip area could be dedicated to processor and memory circuits. Such a system could offer large bandwidth parallel communication channels for multiprocessors, based entirely on proven, compatible materials and processes and could be accomplished in near future. Thin film polymer etalon structures can be used as the modulators. The electrically stored information can be changed to optical information and then broadcast to receiving arrays. The amorphous silicon can be deposited on top of the silicon chip as a receiving array. The concept and important technical issues are discussed. The initial experimental results suggest that this new system is promising View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hybrid multiprocessing using WDM optical fiber interconnections

    Page(s): 182 - 196
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1112 KB)  

    We present the design of a wavelength division multiplexed fiber optic bus for multiprocessors that allows both the shared memory model and the distributed memory model to be supported efficiently. We establish this by presenting the results of a fairly detailed trace-driven simulation of the performance of a system using the proposed interconnection. We also discuss some of the engineering issues involved in the design of such a WDM fiber bus and briefly mention approaches for cutting down on the cost of the optonic interface components. Our results seem to indicate that efficient performance in the shared and distributed memory modes of operation stems from the operational concurrency of the WDM channels and the high data rate and low latency possible from the use of the new fiber channel optical receivers and transmitters View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optical interconnection and massively parallel processing

    Page(s): 118 - 132
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (660 KB)  

    The optical interconnection plays an important role of realization in massively parallel computation. This is well recognized, but it is not clear how the optical interconnection works in parallel processing. The authors revisit the relationships between optical interconnection and parallel data processing. Their view is originally architecture-oriented, but it will be profitable also for device-oriented research View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Optical interconnects based on two-dimensional VCSEL arrays

    Page(s): 202 - 212
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (884 KB)  

    Free space optical interconnects offer an exciting alternative to conventional packaging for electronic systems which uses planar board and backplane structures. The alternative, which shall be called 3D packaging, involves using the space perpendicular to the processing planes for free-space optical beams that transport the data between the planes. 3D packaging is very appealing for tightly-coupled fine-grained parallel computing where the need for massive numbers of interconnections is severely taxing the capabilities of the planar structures. This paper describes an effort to develop the enabling technologies for these 3D systems, and to demonstrate a functional computer based on this technology. The approach is to realize the optical interconnections through the use of two-dimensional optoelectronic (smart pixel) arrays based on vertical-cavity surface-emitting laser arrays View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A model for a reconfigurable fine-grained optoelectronic processor

    Page(s): 19 - 26
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (608 KB)  

    A model for a dataflow based processor is described in which a program written in a high level language is mapped directly to hardware. The concept is to reconfigure the interconnection network among an array of processing elements (PEs) to match the natural form of a computation, as represented by a dataflow graph. Communication among PEs is handled optically using free-space interconnects. A group of vertical cavity surface emitting lasers (VCSELs) is dedicated to each output port of a PE, which corresponds to an arc in a dataflow graph. Outputs of the VCSELs are imaged through a reconfigurable optical permutation network that redirects beams to their destinations. This combination of optics and electronics may support fine-grained parallelism while balancing time spent in communication with time spent in computation View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Fabrication issues for free-space optics at the board packaging level

    Page(s): 45 - 51
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (412 KB)  

    The successful application of optical interconnects to electronic processing systems requires careful consideration of the: electronic system in which it will be used; optical system and electro-optical device performance; electrical-electro optic interface characteristics; packaging of micro optic components and electro-optic devices with electronic substrates; and the reliability of optical interconnects. In addition, the optical system must be performance and cost competitive with advanced electrical connection methods. We consider several of the above issues at the board packaging level, and the design of relevant optical components for these systems View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Design considerations for optical interconnects in parallel computers

    Page(s): 306 - 322
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1224 KB)  

    Communication complexity and latency is a critical problem in multiprocessor systems. A significant portion of communication latency is associated with the interconnect network. Optics has many advantages for achieving low latency, scalable interprocessor communication. The author identifies significant ways in which optical technology can boost network functionality and performance when key architectural and implementation design issues are considered. A high bandwidth, reconfigurable optical interconnect capable of increased network throughput and optimal processor-memory connectivity can result from this approach View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Data motion and high performance computing

    Page(s): 1 - 18
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1288 KB)  

    Efficient data motion has been of critical importance in high performance computing almost since the first electronic computers were built. Providing sufficient memory bandwidth to balance the capacity of processors led to memory hierarchies, banked and interleaved memories. With the rapid evolution of MOS technologies, microprocessor and memory designs, it is realistic to build systems with thousands of processors and a sustained performance of a trillion operations per second or more. Such systems require tens of thousands of memory banks, even when locality of reference is exploited. Using conventional technologies, interconnecting several thousand processors with tens of thousands of memory banks can feasibly only be made by some form of sparse interconnection network. Efficient use of locality of reference and network bandwidth is critical View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Multiprocessor interconnection networks using partitioned optical passive star (POPS) topologies and distributed control

    Page(s): 70 - 80
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (928 KB)  

    The authors present a scalable electro-optical interconnection network architecture which is suitable for tightly coupled multiprocessors. The architecture is called a partitioned optical passive star (POPS). It is a type of multiple passive star topology in which only constant and symmetric coupler fanouts are used and in which exactly one coupler is traversed on any path through the network. Control is based on the state sequence routing paradigm which multiplexes the network between a small set of control states and defines a control operation to be a transformation of those states. These networks have highly scalable characteristics for optical power budget, resource count, and message latency. Optical power is uniformly distributed and the size of the system is not hard limited by the power budget. Resource complexity grows with asymptotic complexity O(n) for the couplers, O(n√n) for transceivers, and O(√n log(n)) for control. We present a static analysis and a simulation of dynamic performance which demonstrates the ability of a POPS design to support 1024 nodes using current device and coupler technology View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Efficient communication operations on passive optical star networks

    Page(s): 52 - 58
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (452 KB)  

    We show how to use the wavelength division multiple access capabilities of passive optical star networks for efficiently implementing communication operations that are widely used in parallel applications. We propose algorithms for multiple broadcasting, scattering, gossiping and multi-scattering, which are very close to the lower bound for these problems View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.