Hongki Sung

An optical multi-mesh hypercube: a scalable optical interconnection network for massively parallel computing

A new interconnection network for massively parallel computing is introduced. This network is called an optical multi-mesh hypercube (OMMH) network. The OMMH integrates positive features of both hypercube (small diameter, high connectivity, symmetry, simple control and routing, fault tolerance, etc.) and mesh (constant node degree and scalability) topologies and at the same time circumvents their limitations (e.g., the lack of scalability of hypercubes, and the large diameter of meshes). The OMMH can maintain a constant node degree regardless of the increase in the network size. In addition, the flexibility of the OMMH network makes it well suited for optical implementations. This paper presents the OMMH topology, analyzes its architectural properties and potentials for massively parallel computing, and compares it to the hypercube. Moreover, it also presents a three-dimensional optical design methodology based on free-space optics. The proposed optical implementation has totally space-invariant connection patterns at every node, which enables the OMMH to be highly amenable to optical implementation using simple and efficient large space-bandwidth product space-invariant optical elements. >

Scalable optical hypercube-based interconnection network for massively parallel computing

Two important parameters of a network for massively parallel computers are scalability and modularity. Scalability has two aspects: size and time (or generation). Size scalability refers to the property that the size of the network can be increased with nominal effect on the existing configuration. Also, the increase in size is expected to result in a linear increase in performance. Time scalability implies that the communication capabilities of a network should be large enough to support the evolution of processing elements through generations. A modular network enables the construction of a large network out of many smaller ones. The lack of these two important parameters has limited the use of certain types of interconnection networks in the area of massively parallel computers. We present a new modular optical interconnection network, called an optical multimesh hypercube (OMMH), which is both size and time scalable. The OMMH combines positive features of both the hypercube (small diameter, high connectivity, symmetry, simple routing, and fault tolerance) and the torus (constant node degree and size scalability) networks. Also presented is a three-dimensional optical implementation of the OMMH network. A basic building block of the OMMH network is a hypercube module that is constructed with free-space optics to provide compact and high-density localized hypercube connections. The OMMH network is then constructed by the connection of such basic building blocks with multiwavelength optical fibers to realize torus connections. The proposed implementation methodology is intended to exploit the advantages of both space-invariant free-space and multiwavelength fiber-based optical interconnect technologies. The analysis of the proposed implementation shows that such a network is optically feasible in terms of the physical size and the optical power budget.

3D optical interconnects for high-speed interchip and interboard communications

Metal-based communications between subsystems and chips has become the limiting factor in high-speed computing. Maturing optics-based technologies offer advantages that may unplug this bottleneck. Optical interconnects offer high-speed computers key advantages over metal interconnects. These include (1) high spatial and temporal bandwidths, (2) high-speed transmission, (3) low crosstalk independent of data rates, and (4) high interconnect densities. Although faster device switching speeds will eventually be necessary for future massively parallel computing systems, the deciding factor in determining system performance and cost will be subsystem communications rather than device speed. Free-space optical interconnects, by virtue of their inherent parallelism, high data bandwidth, small size and power requirement, and relative freedom from mutual interference of signals, already show great promise in replacing metal interconnects to solve communication problems.<<ETX>>

Efficient implementation methodology for three-dimensional space-invariant hypercube-based optical interconnection networks

A new design methodology for constructing optical space-invariant hypercube interconnection networks for connection of a two-dimensional array of inputs to a two-dimensional array of outputs is presented. The methodology permits the construction of larger hypercube networks from smaller networks in systematic and incremental fashion. It is shown that the proposed methodology greatly improves area utilization as compared with previous methods. An example network is provided that illustrates the proposed design method. Owing to their totally space-invariant nature, the resulting three-dimensional hypercube networks are highly amenable to optical implementations by use of simple optical hardware such as multiple-imaging components and space-invariant holographic techniques. We present space-invariant optical implementation technique for the realization of such networks. A theoretical analysis of the physical limitations of the implementation method is also presented. The analysis indicates that two-dimensional arrays of 512 × 512 nodes interconnected in a hypercube (18-cube) topology could be implemented.

An efficient 3D optical implementation of binary de Bruijn networks with applications to massively parallel computing

As an alternative to the hypercube, the binary de Bruijn (BdB) network is recently receiving much attention. The BdB not only provides a logarithmic diameter fault tolerance, and simple routing but also requires fewer links than the hypercube for the same network size. Additionally, a major advantage of the BdB network is a constant node degree: the number of edges per node is independent of the network size. This makes it very desirable for large scale parallel systems. However, due to its asymmetrical nature and global connectivity it is posing a major challenge for VLSI technology. Optics, owing to its three dimensional and global connectivity nature seems to be very suitable for implementing BdB networks. We present an implementation methodology for BdB networks. The distinctive feature of the proposed implementation methodology is partitionability of the network into a few primitive operations that can be efficiently implemented. We further show feasibility of the presented design methodology by proposing an optical implementation of the BdB network.

A hypercube-based optical interconnection network: a solution to the scalability requirements for massively parallel computers

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.<<ETX>>

Design methodology for three-dimensional space-invariant hypercube networks with graph bipartitioning

We present a new design methodology for constructing three-dimensional space-invariant hypercube interconnection networks. The methodology is based on a graph bipartitioning technique and permits the construction of larger hypercube networks from smaller networks in a systematic and incremental fashion. Owing to their totally space-invariant nature, the resulting three-dimensional hypercube networks are amenable to optical implementations with simple optical hardware such as multiple imaging components and space-invariant holographic techniques.

Optical binary de Bruijn networks for massively parallel computing: Design methodology and feasibility study

The interconnection network structure can be the deciding and limiting factor in the cost and the performance of parallel computers. One of the most popular point-to-point interconnection networks for parallel computers today is the hypercube. The regularity, logarithmic diameter, symmetry, high connectivity, fault tolerance, simple routing, and reconfigurability (easy embedding of other network topologies) of the hypercube make it a very attractive choice for parallel computers. Unfortunately the hypercube possesses a major drawback, which is the complexity of its node structure: the number of links per node increases as the network grows in size. As an alternative to the hypercube, the binary de Bruijn (BdB) network has recently received much attention. The BdB not only provides a logarithmic diameter, fault tolerance, and simple routing but also requires fewer links than the hypercube for the same network size. Additionally, a major advantage of the BdB network is a constant node degree: the number of edges per node is independent of the network size. This makes it very desirable for large-scale parallel systems. However, because of its asymmetrical nature and global connectivity, it poses a major challenge for VLSI technology. Optics, owing to its three-dimensional and globalconnectivity nature, seems to be very suitable for implementing BdB networks. We present an implementation methodology for optical BdB networks. The distinctive feature of the proposed implementation methodology is partitionability of the network into a few primitive operations that can be implemented efficiently. We further show feasibility of the presented design methodology by proposing an optical implementation of the BdB network.

A new compiler-directed cache coherence scheme for shared memory multiprocessors with fast and parallel explicit invalidation (abstract)

We propose a novel compiler-directed cache management scheme which allows parallel invalidation of a subset of array elements. The scheme limits non-stale data invalidations using a novel memory allocation technique. Its correctness is proved using a flow graph model. It is also shown that the scheme provides more cacheability than the previous compiler-directed ones and has lower overhead in determining read hit at runtime. A new performance parameter called unwanted invalidation ratio, for compiler-directed coherence schemes is also proposed.

An enhanced one-step C-testable design of two-dimensional iterative logic arrays

The authors present an improved approach to one-step C-testability of orthogonal two-dimensional iterative logic arrays. This is an improvement of the approach of W. Huang and F. Lombardi (1988) and H. Elhuni et al. (1986). A group of sufficient conditions to test two-dimensional iterative logic arrays with a constant number of test vectors independent of the array size (C-testability) is stated. It is proved that the proposed approach requires a smaller number of test vectors than in previous works.<<ETX>>