Parallel Networks for Machine Vision

The amount of computation required to solve many early vision problems is prodigious, and so it has long been thought that systems that operate in a reasonable amount of time will only become feasible when parallel systems become available. Such systems now exist in digital form, but most are large and expensive. These machines constitute an invaluable test-bed for the development of new algorithms, but they can probably not be scaled down rapidly in both physical size and cost, despite continued advances in semiconductor technology and machine architecture. Simple analog networks can perform interesting computations, as has been known for a long time. We have reached the point where it is feasible to experiment with implementation of these ideas in VLSI form, particularly if we focus on networks composed of locally interconnected passive elements, linear amplifiers, and simple nonlinear components. While there have been excursions into the development of ideas in this area since the very beginnings of work on machine vision, much work remains to be done. Progress will depend on careful attention to matching of the capabilities of simple networks to the needs of early vision. Note that this is not at all intended to be anything like a review of the field, but merely a collection of some ideas that seem to be interesting.

Parallel Computation of Vernier Offsets, Curvature and Chevrons in Humans

A vernier offset is detected at once among straight lines, and reaction times are almost independent of the number of simultaneously presented stimuli (distractors), indicating parallel processing of vernier offsets. Reaction times for identifying a vernier offset to one side among verniers offset to the opposite side increase with the number of distractors, indicating serial processing. Even deviations below a photoreceptor diameter can be detected at once. The visual system thus attains positional accuracy below the photoreceptor diameter simultaneously at different positions. I conclude that deviation from straightness, or change of orientation, is detected in parallel over the visual field. Discontinuities or gradients in orientation may represent an elementary feature of vision.

Parallel Simulation of Subsonic Fluid Dynamics on a Cluster of Workstations

An effective approach of simulating fluid dynamics on a cluster of non- dedicated workstations is presented. The approach uses local interaction algorithms, small communication capacity, and automatic migration of parallel processes from busy hosts to free hosts. The approach is well- suited for simulating subsonic flow problems which involve both hydrodynamics and acoustic waves; for example, the flow of air inside wind musical instruments. Typical simulations achieve $80\\%$ parallel efficiency (speedup/processors) using 20 HP-Apollo workstations. Detailed measurements of the parallel efficiency of 2D and 3D simulations are presented, and a theoretical model of efficiency is developed which fits closely the measurements. Two numerical methods of fluid dynamics are tested: explicit finite differences, and the lattice Boltzmann method.

A Coupled Multi-ALU Processing Node for a Highly Parallel Computer

This report describes Processor Coupling, a mechanism for controlling multiple ALUs on a single integrated circuit to exploit both instruction-level and inter-thread parallelism. A compiler statically schedules individual threads to discover available intra-thread instruction-level parallelism. The runtime scheduling mechanism interleaves threads, exploiting inter-thread parallelism to maintain high ALU utilization. ALUs are assigned to threads on a cycle byscycle basis, and several threads can be active concurrently. Simulation results show that Processor Coupling performs well both on single threaded and multi-threaded applications. The experiments address the effects of memory latencies, function unit latencies, and communication bandwidth between function units.

Reliable Interconnection Networks for Parallel Computers

This technical report describes a new protocol, the Unique Token Protocol, for reliable message communication. This protocol eliminates the need for end-to-end acknowledgments and minimizes the communication effort when no dynamic errors occur. Various properties of end-to-end protocols are presented. The unique token protocol solves the associated problems. It eliminates source buffering by maintaining in the network at least two copies of a message. A token is used to decide if a message was delivered to the destination exactly once. This technical report also presents a possible implementation of the protocol in a worm-hole routed, 3-D mesh network.

A Parallel Crossbar Routing Chip for a Shared Memory Multiprocessor

This thesis describes the design and implementation of an integrated circuit and associated packaging to be used as the building block for the data routing network of a large scale shared memory multiprocessor system. A general purpose multiprocessor depends on high-bandwidth, low-latency communications between computing elements. This thesis describes the design and construction of RN1, a novel self-routing, enhanced crossbar switch as a CMOS VLSI chip. This chip provides the basic building block for a scalable pipelined routing network with byte-wide data channels. A series of RN1 chips can be cascaded with no additional internal network components to form a multistage fault-tolerant routing switch. The chip is designed to operate at clock frequencies up to 100Mhz using Hewlett-Packard's HP34 $1.2\\mu$ process. This aggressive performance goal demands that special attention be paid to optimization of the logic architecture and circuit design.

Design and Evaluation of the Hamal Parallel Computer

Parallel shared-memory machines with hundreds or thousands of processor-memory nodes have been built; in the future we will see machines with millions or even billions of nodes. Associated with such large systems is a new set of design challenges. Many problems must be addressed by an architecture in order for it to be successful; of these, we focus on three in particular. First, a scalable memory system is required. Second, the network messaging protocol must be fault-tolerant. Third, the overheads of thread creation, thread management and synchronization must be extremely low. This thesis presents the complete system design for Hamal, a shared-memory architecture which addresses these concerns and is directly scalable to one million nodes. Virtual memory and distributed objects are implemented in a manner that requires neither inter-node synchronization nor the storage of globally coherent translations at each node. We develop a lightweight fault-tolerant messaging protocol that guarantees message delivery and idempotence across a discarding network. A number of hardware mechanisms provide efficient support for massive multithreading and fine-grained synchronization. Experiments are conducted in simulation, using a trace-driven network simulator to investigate the messaging protocol and a cycle-accurate simulator to evaluate the Hamal architecture. We determine implementation parameters for the messaging protocol which optimize performance. A discarding network is easier to design and can be clocked at a higher rate, and we find that with this protocol its performance can approach that of a non-discarding network. Our simulations of Hamal demonstrate the effectiveness of its thread management and synchronization primitives. In particular, we find register-based synchronization to be an extremely efficient mechanism which can be used to implement a software barrier with a latency of only 523 cycles on a 512 node machine.

Sparsely Faceted Arrays: A Mechanism Supporting Parallel Allocation, Communication, and Garbage Collection

Conventional parallel computer architectures do not provide support for non-uniformly distributed objects. In this thesis, I introduce sparsely faceted arrays (SFAs), a new low-level mechanism for naming regions of memory, or facets, on different processors in a distributed, shared memory parallel processing system. Sparsely faceted arrays address the disconnect between the global distributed arrays provided by conventional architectures (e.g. the Cray T3 series), and the requirements of high-level parallel programming methods that wish to use objects that are distributed over only a subset of processing elements. A sparsely faceted array names a virtual globally-distributed array, but actual facets are lazily allocated. By providing simple semantics and making efficient use of memory, SFAs enable efficient implementation of a variety of non-uniformly distributed data structures and related algorithms. I present example applications which use SFAs, and describe and evaluate simple hardware mechanisms for implementing SFAs. Keeping track of which nodes have allocated facets for a particular SFA is an important task that suggests the need for automatic memory management, including garbage collection. To address this need, I first argue that conventional tracing techniques such as mark/sweep and copying GC are inherently unscalable in parallel systems. I then present a parallel memory-management strategy, based on reference-counting, that is capable of garbage collecting sparsely faceted arrays. I also discuss opportunities for hardware support of this garbage collection strategy. I have implemented a high-level hardware/OS simulator featuring hardware support for sparsely faceted arrays and automatic garbage collection. I describe the simulator and outline a few of the numerous details associated with a "real" implementation of SFAs and SFA-aware garbage collection. Simulation results are used throughout this thesis in the evaluation of hardware support mechanisms.

A Parallel Processing Model of Musical Structures

Euterpe is a real-time computer system for the modeling of musical structures. It provides a formalism wherein familiar concepts of musical analysis may be readily expressed. This is verified by its application to the analysis of a wide variety of conventional forms of music: Gregorian chant, Mediaeval polyphony, Back counterpoint, and sonata form. It may be of further assistance in the real-time experiments in various techniques of thematic development. Finally, the system is endowed with sound-synthesis apparatus with which the user may prepare tapes for musical performances.

Pi: A Parallel Architecture Interface for Multi-Model Execution

This thesis defines Pi, a parallel architecture interface that separates model and machine issues, allowing them to be addressed independently. This provides greater flexibility for both the model and machine builder. Pi addresses a set of common parallel model requirements including low latency communication, fast task switching, low cost synchronization, efficient storage management, the ability to exploit locality, and efficient support for sequential code. Since Pi provides generic parallel operations, it can efficiently support many parallel programming models including hybrids of existing models. Pi also forms a basis of comparison for architectural components.

Parallel Methods for Synthesizing Whole-Hand Grasps from Generalized Prototypes

This report addresses the problem of acquiring objects using articulated robotic hands. Standard grasps are used to make the problem tractable, and a technique is developed for generalizing these standard grasps to increase their flexibility to variations in the problem geometry. A generalized grasp description is applied to a new problem situation using a parallel search through hand configuration space, and the result of this operation is a global overview of the space of good solutions. The techniques presented in this report have been implemented, and the results are verified using the Salisbury three-finger robotic hand.

Thread Scheduling Mechanisms for Multiple-Context Parallel Processors

Scheduling tasks to efficiently use the available processor resources is crucial to minimizing the runtime of applications on shared-memory parallel processors. One factor that contributes to poor processor utilization is the idle time caused by long latency operations, such as remote memory references or processor synchronization operations. One way of tolerating this latency is to use a processor with multiple hardware contexts that can rapidly switch to executing another thread of computation whenever a long latency operation occurs, thus increasing processor utilization by overlapping computation with communication. Although multiple contexts are effective for tolerating latency, this effectiveness can be limited by memory and network bandwidth, by cache interference effects among the multiple contexts, and by critical tasks sharing processor resources with less critical tasks. This thesis presents techniques that increase the effectiveness of multiple contexts by intelligently scheduling threads to make more efficient use of processor pipeline, bandwidth, and cache resources. This thesis proposes thread prioritization as a fundamental mechanism for directing the thread schedule on a multiple-context processor. A priority is assigned to each thread either statically or dynamically and is used by the thread scheduler to decide which threads to load in the contexts, and to decide which context to switch to on a context switch. We develop a multiple-context model that integrates both cache and network effects, and shows how thread prioritization can both maintain high processor utilization, and limit increases in critical path runtime caused by multithreading. The model also shows that in order to be effective in bandwidth limited applications, thread prioritization must be extended to prioritize memory requests. We show how simple hardware can prioritize the running of threads in the multiple contexts, and the issuing of requests to both the local memory and the network. Simulation experiments show how thread prioritization is used in a variety of applications. Thread prioritization can improve the performance of synchronization primitives by minimizing the number of processor cycles wasted in spinning and devoting more cycles to critical threads. Thread prioritization can be used in combination with other techniques to improve cache performance and minimize cache interference between different working sets in the cache. For applications that are critical path limited, thread prioritization can improve performance by allowing processor resources to be devoted preferentially to critical threads. These experimental results show that thread prioritization is a mechanism that can be used to implement a wide range of scheduling policies.

Parallel Coupled Micro-Macro Actuators

This thesis presents a new actuator system consisting of a micro-actuator and a macro-actuator coupled in parallel via a compliant transmission. The system is called the Parallel Coupled Micro-Macro Actuator, or PaCMMA. In this system, the micro-actuator is capable of high bandwidth force control due to its low mass and direct-drive connection to the output shaft. The compliant transmission of the macro-actuator reduces the impedance (stiffness) at the output shaft and increases the dynamic range of force. Performance improvement over single actuator systems was expected in force control, impedance control, force distortion and reduction of transient impact forces. A set of quantitative measures is proposed and the actuator system is evaluated against them: Force Control Bandwidth, Position Bandwidth, Dynamic Range, Impact Force, Impedance ("Backdriveability'"), Force Distortion and Force Performance Space. Several theoretical performance limits are derived from the saturation limits of the system. A control law is proposed and control system performance is compared to the theoretical limits. A prototype testbed was built using permanenent magnet motors and an experimental comparison was performed between this actuator concept and two single actuator systems. The following performance was observed: Force bandwidth of 56Hz, Torque Dynamic Range of 800:1, Peak Torque of 1040mNm, Minimum Torque of 1.3mNm. Peak Impact Force was reduced by an order of magnitude. Distortion at small amplitudes was reduced substantially. Backdriven impedance was reduced by 2-3 orders of magnitude. This actuator system shows promise for manipulator design as well as psychophysical tests of human performance.

ADAM: A Decentralized Parallel Computer Architecture Featuring Fast Thread and Data Migration and a Uniform Hardware Abstraction

The furious pace of Moore's Law is driving computer architecture into a realm where the the speed of light is the dominant factor in system latencies. The number of clock cycles to span a chip are increasing, while the number of bits that can be accessed within a clock cycle is decreasing. Hence, it is becoming more difficult to hide latency. One alternative solution is to reduce latency by migrating threads and data, but the overhead of existing implementations has previously made migration an unserviceable solution so far. I present an architecture, implementation, and mechanisms that reduces the overhead of migration to the point where migration is a viable supplement to other latency hiding mechanisms, such as multithreading. The architecture is abstract, and presents programmers with a simple, uniform fine-grained multithreaded parallel programming model with implicit memory management. In other words, the spatial nature and implementation details (such as the number of processors) of a parallel machine are entirely hidden from the programmer. Compiler writers are encouraged to devise programming languages for the machine that guide a programmer to express their ideas in terms of objects, since objects exhibit an inherent physical locality of data and code. The machine implementation can then leverage this locality to automatically distribute data and threads across the physical machine by using a set of high performance migration mechanisms. An implementation of this architecture could migrate a null thread in 66 cycles -- over a factor of 1000 improvement over previous work. Performance also scales well; the time required to move a typical thread is only 4 to 5 times that of a null thread. Data migration performance is similar, and scales linearly with data block size. Since the performance of the migration mechanism is on par with that of an L2 cache, the implementation simulated in my work has no data caches and relies instead on multithreading and the migration mechanism to hide and reduce access latencies.

On-the-Fly Maintenance of Series-Parallel Relationships in Fork-Join Multithreaded Programs

A key capability of data-race detectors is to determine whether one thread executes logically in parallel with another or whether the threads must operate in series. This paper provides two algorithms, one serial and one parallel, to maintain series-parallel (SP) relationships "on the fly" for fork-join multithreaded programs. The serial SP-order algorithm runs in O(1) amortized time per operation. In contrast, the previously best algorithm requires a time per operation that is proportional to Tarjan’s functional inverse of Ackermann’s function. SP-order employs an order-maintenance data structure that allows us to implement a more efficient "English-Hebrew" labeling scheme than was used in earlier race detectors, which immediately yields an improved determinacy-race detector. In particular, any fork-join program running in T₁ time on a single processor can be checked on the fly for determinacy races in O(T₁) time. Corresponding improved bounds can also be obtained for more sophisticated data-race detectors, for example, those that use locks. By combining SP-order with Feng and Leiserson’s serial SP-bags algorithm, we obtain a parallel SP-maintenance algorithm, called SP-hybrid. Suppose that a fork-join program has n threads, T₁ work, and a critical-path length of T[subscript ∞]. When executed on P processors, we prove that SP-hybrid runs in O((T₁/P + PT[subscript ∞]) lg n) expected time. To understand this bound, consider that the original program obtains linear speed-up over a 1-processor execution when P = O(T₁/T[subscript ∞]). In contrast, SP-hybrid obtains linear speed-up when P = O(√T₁/T[subscript ∞]), but the work is increased by a factor of O(lg n).