Systems Architecture Commons^™

Scalable Correct Memory Ordering Via Relativistic Programming, Josh Triplett, Philip William Howard, Paul E. Mckenney, Jonathan Walpole

Computer Science Faculty Publications and Presentations

We propose and document a new concurrent programming model, relativistic programming. This model allows readers to run concurrently with writers, without blocking or using expensive synchronization. Relativistic programming builds on existing synchronization primitives that allow writers to wait for current readers to finish with minimal reader overhead. Our methodology models data structures as graphs, and reader algorithms as traversals of these graphs; from this foundation we show how writers can implement arbitrarily strong ordering guarantees for the visibility of their writes, up to and including total ordering.

Generalized Construction Of Scalable Concurrent Data Structures Via Relativistic Programming, Josh Triplett, Paul E. Mckenney, Philip W. Howard, Jonathan Walpole

Computer Science Faculty Publications and Presentations

We present relativistic programming, a concurrent programming model based on shared addressing, which supports efficient, scalable operation on either uniform shared-memory or distributed shared- memory systems. Relativistic programming provides a strong causal ordering property, allowing a series of read operations to appear as an atomic transaction that occurs entirely between two ordered write operations. This preserves the simple immutable-memory programming model available via mutual exclusion or transactional memory. Furthermore, relativistic programming provides joint-access parallelism, allowing readers to run concurrently with a writer on the same data. We demonstrate a generalized construction technique for concurrent data structures based on relativistic programming, …

A Comparison Of Relativistic And Reader-Writer Locking Approaches To Shared Data Access, Philip William Howard, Josh Triplett, Jonathan Walpole

Computer Science Faculty Publications and Presentations

This paper explores the relationship between reader-writer locking and relativistic programming approaches to managing accesses to shared data. It demonstrates that by placing certain restrictions on writers, relativistic programming allows more concurrency than reader-writer locking while still providing the same isolation guarantees. Relativistic programming also allows for a straightforward model for reasoning about the correctness of programs that allow concurrent read-write accesses.

The Ordering Requirements Of Relativistic And Reader-Writer Locking Approaches To Shared Data Access, Philip William Howard, Josh Triplett, Jonathan Walpole, Paul E. Mckenney

Computer Science Faculty Publications and Presentations

The semantics of reader-writer locks allow read-side concurrency. Unfortunately, the locking primitives serialize access to the lock variable to an extent that little or no concurrency is realized in practice for small critical sections. Relativistic programming is a methodology that also allows read- side concurrency. Relativistic programming uses dfferent ordering constraints than reader-writer locking. The different ordering constraints allow relativistic readers to proceed without synchronization so relativistic readers scale even for very short critical sections. In this paper we explore the diferences between the ordering constraints for reader-writer locking and relativistic programs. We show how and why the dfferent ordering …

Pvw: Designing Virtual World Server Infrastructure, Francis Chang, C. Mic Bowman, Wu-Chi Feng

Computer Science Faculty Publications and Presentations

This paper presents a high level overview of PVW (Partitioned Virtual Worlds), a distributed system architecture for the management of virtual worlds. PVW is designed to support arbitrarily large and complex virtual worlds while accommodating dynamic and highly variable user population and content distribution density. The PVW approach enables the task of simulating and managing the virtual world to be distributed over many servers by spatially partitioning the environment into a hierarchical structure. This structure is useful both for balancing the simulation load across many nodes, as well as features such as geometric simplification and distribution of dynamic content.

Xpu: A Distributed Architecture For Metaverses, Francis Chang, C. Mic Bowman, Wu-Chi Feng

Computer Science Faculty Publications and Presentations

A significant problem of designing 3D virtual worlds (such as metaverses) is developing a scalable architecture that can manage millions of simultaneous users in an interactive 3D environment. This paper presents XPU (Extremely Partitioned Universe), a hierarchical client-server architecture for developing highly scalable metaverses. This design addresses the problem of dynamically partitioning the world to manage network and computing resources.

Is Parallel Programming Hard, And If So, Why?, Paul E. Mckenney, Maged M. Michael, Manish Gupta, Philip William Howard, Josh Triplett, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Of the 200+ parallel-programming languages and environments created in the 1990s, almost all are now defunct. Given that parallel systems are now well within the budget of the typical hobbyist or graduate student, it is not unreasonable to expect a new cohort in excess of several thousand parallel languages and environments to appear in the 2010s. If this expected new cohort is to have more practical impact than did its 1990s counterpart, a robust and widely applicable framework will be required that encompasses exactly what, if anything, is hard about parallel programming. This paper revisits the fundamental precepts of concurrent …

A Pattern Language For Extensible Program Representation, Andrew P. Black, Daniel Vainsencher

Computer Science Faculty Publications and Presentations

For the last 15 years, implementors of multiple view programming environments have sought a single code model that would form a suitable basis for all of the program analyses and tools that might be applied to the code. They have been unsuccessful. The consequences are a tendency to build monolithic, single-purpose tools, each of which implements its own specialized analyses and optimized representation. This restricts the availability of the analyses, and also limits the reusability of the representation by other tools. Unintegrated tools also produce inconsistent views, which reduce the value of multiple views. This article describes a set of …

Thread Transparency In Information Flow Middleware, Rainer Koster, Andrew P. Black, Jie Huang, Jonathan Walpole, Calton Pu

Computer Science Faculty Publications and Presentations

Existing middleware is based on control-flow centric interaction models such as remote method invocations, poorly matching the structure of applications that process continuous information flows. Difficulties cultiesin building this kind of application on conventional platforms include flow-specific concurrency and timing requirements, necessitating explicit management of threads, synchronization, and timing by the application programmer. We propose Infopipes as a high-level abstraction for information flows, and we are developing a middleware framework that supports this abstraction. Infopipes transparently handle complexities associated with control flow and multi-threading. From high-level configuration descriptions the platform determines what parts of a pipeline require separate threads or …

Work In Progress: Automating Proportion/Period Scheduling, David Steere, Jonathan Walpole, Calton Pu

Computer Science Faculty Publications and Presentations

The recent effort to define middleware capable of supporting real-time applications creates the opportunity to raise the level of abstraction presented to the programmer. We propose that proportion/period is a better abstraction for specifying resource needs and allocation than priorities. We are currently investigating techniques to address some issues that are restricting use of proportion/period scheduling to research real-time prototypes. In particular, we are investigating techniques to automate the task of selecting proportion and period, and that allow proportion/period to incorporate job importance under overload conditions.

Feedback Based Dynamic Proportion Allocation For Disk I/O, Dan Revel, Dylan Mcnamee, Calton Pu, David Steere, Jonathan Walpole

Computer Science Faculty Publications and Presentations

In this paper we propose to use feedback control to automatically allocate disk bandwidth in order to match the rate of disk I/O to the real-rate needs of applications. We describe a model for adaptive resource management based on measuring the relative progress of stages in a producer-consumer pipeline. We show how to use prefetching to transform a passive disk into an active data producer whose progress can be controlled via feedback. Our progress-based framework allows the integrated control of multiple resources. The resulting system automatically adapts to varying application rates as well as to varying device latencies.

Synthetic Files: Enabling Low-Latency File I/O For Qos-Adaptive Applications, Dylan Mcnamee, Dan Revel, Calton Pu, David Steere, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Files are a tried and true operating system abstraction. They present a simple byte-stream model of I/O that has proven intuitive for application programmers and efficient for operating system builders. However, current file systems do not provide good support for adaptive continuous media (CM) applications - an increasingly important class of applications that exhibit complex access patterns and are particularly sensitive to variations in I/O performance. To address these problems we propose synthetic files. Synthetic files are specialized views of underlying regular files, and convert complex file access patterns into simple sequential synthetic file access patterns. Synthetic file construction can …

Adaptation Space: Surviving Non-Maskable Failures, Crispin Cowan, Lois Delcambre, Anne-Francoise Le Meur, Ling Liu, David Maier, Dylan Mcnamee, Michael Miller, Calton Pu, Perry Wagle, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Some failures cannot be masked by redundancies, because an unanticipated situation occurred, because fault-tolerance measures were not adequate, or because there was a security breach (which is not amenable to replication). Applications that wish to continue to offer some service despite nonmaskable failure must adapt to the loss of resources. When numerous combinations of non-maskable failure modes are considered, the set of possible adaptations becomes complex. This paper presents adaptation spaces, a formalism for navigating among combinations of adaptations. An adaptation space describes a collection of possible adaptations of a software component or system, and provides a uniform way of …

Adaptive Prefetching For Device-Independent File I/O, Dan Revel, Dylan Mcnamee, David Steere, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Device independent I/O has been a holy grail to operating system designers since the early days of UNIX. Unfortunately, existing operating systems fall short of this goal for multimedia applications. Techniques such as caching and sequential read-ahead can help mask I/O latency in some cases, but in others they increase latency and add substantial jitter. Multimedia applications, such as video players, are sensitive to vagaries in performance since I/O latency and jitter affect the quality of presentation. Our solution uses adaptive prefetching to reduce both latency and jitter. Applications submit file access plans to the prefetcher, which then generates I/O …

Cheetahs Are Fast, But Nearly Irrelevant, Calton Pu, Jonathan Walpole, Charles Consel

Computer Science Faculty Publications and Presentations

A brief paper submitted to the NSF Workshop on New Challenges and Directions for Systems Research, held in St. Louis, Mo., July 31-August 1, 1997. Outlines the case for more cooperation among software system architects, and a less reductionistic approach to software development.

Predictable File Access Latency For Multimedia, Dan Revel, Crispin Cowan, Dylan Mcnamee, Calton Pu, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Multimedia applications are sensitive to I/O latency and jitter when accessing data in secondary storage. Transparent adaptive prefetching (TAP) uses software feedback to provide multimedia applications with file system quality of service (QoS) guarantees. We are investigating how QoS requirements can be communicated and how they can be met by adaptive resource management. A preliminary test of adaptive prefetching is presented.

A Migratable User-Level Process Package For Pvm, Ravi Kunuru, Steve Otto, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Shared, multi-user, workstation networks are characterized by unpredictable variability in system load. Further, the concept of workstation ownership is typically present. For efficient and unobtrusive computing in such environments, applications must not only overlap their computation with communication but also redistribute their computations adaptively based on changes in workstation availability and load. Managing these issues at application level leads to programs that are difficult to write and debug. In this paper, we present a system that manages this dynamic multi-processor environment while exporting a simple message-based programming model of a dedicated, distributed memory multiprocessor to applications. Programmers are thus insulated …

Adaptive Methods For Distributed Video Presentation, Crispin Cowan, Shanwei Cen, Jonathan Walpole, Carlton Pu

Computer Science Faculty Publications and Presentations

This paper describes problems and solutions for delivering real-time, multi-media presentations across the Internet. A key characteristic of presentations of continuous media datatypes, such as digital video and audio, is their need for predictable real-time data delivery. For example, an NTSC quality video presentation requires video frames to be displayed every 1/30th of a second. Variations in this display rate can be observable as stalls or glitches in the video stream and reduce the quality of the presentation [6]. Delivering such presentations across the Internet is difficult because highly variable band- width and latency make it difficult to predict the …

Mist: Pvm With Transparent Migration And Checkpointing, Jeremy Casas, Dan Clark, Phil Galbiati, Ravi Konuru, Steve Otto, Robert Prouty, Jonathan Walpole

Computer Science Faculty Publications and Presentations

We are currently involved in research to enable PVM to take advantage of shared networks of workstations (NOWs) more effectively. In such a computing environment, it is important to utilize workstations unobtrusively and recover from machine failures. Towards this goal, we have enhanced PVM with transparent task migration, checkpointing, and global scheduling. These enhancements are part of the MIST project which takes an open systems approach in developing a cohesive, distributed parallel computing environment. This open systems approach promotes plug-and-play integration of independently developed modules, such as Condor, DQS, A VS, Prospero, XPVM, PIOUS, Ptools, etc. Transparent task migration, in …

Adaptive Execution Of Data Parallel Computations On Networks Of Heterogeneous Workstations, Robert Prouty, Steve Otto, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Parallel environments consisting of a network of heterogeneous workstations introduce an inherently dynamic environment that differs from multicomputers. Workstations are usually considered “shared” resources while multicomputers provide dedicated processing power. The number of workstations available for use is continually changing; the parallel machine presented by the network is in effect continually reconfiguring itself. Application programs must effectively adapt to the changing number of processing nodes while maintaining computational efficiency. This paper examines methods for adapting to this dynamic environment within the framework of explicit message passing under the data parallel programming model. We present four requirements which we feel a …

Comet: A Synthetic Benchmark For Message-Passing Architectures, Nalini Ganapati, Steve Otto, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Rapid advances in hardware technology have led to wide diversity in parallel computer architectures. This diversity makes it difficult to evaluate or compare the performance of different parallel computers. Existing benchmarks tend either to be too architecture-specific, or too high-level. Both problems can result in benchmarks that not only provide insufficient information on the performance characteristics of the computer being tested, but are also difficult to port. New benchmarking approaches are needed for new architectural classes, particularly distributed-memory, message-passing computers. This paper focuses on benchmarking distributed-memory message-passing computers. A synthetic benchmark called CoMet (COmmunication METrics), is presented. CoMet is based …

A Study Of Dynamic Optimization Techniques: Lessons And Directions In Kernel Design, Calton Pu, Jonathan Walpole

Computer Science Faculty Publications and Presentations

The Synthesis kernel [21,22,23,27,28] showed that dynamic code generation, software feedback, and fine-grain modular kernel organization are useful implementation techniques for improving the performance of operating system kernels. In addition, and perhaps more importantly, we discovered that there are strong interactions between the techniques. Hence, a careful and systematic combination of the techniques can be very powerful even though each one by itself may have serious limitations. By identifying these interactions we illustrate the problems of applying each technique in isolation to existing kernels. We also highlight the important common under-pinnings of the Synthesis experience and present our ideas on …

Constrained-Latency Storage Access: A Survey Of Application Requirements And Storage System Design Approaches, Richard Staehli, Jonathan Walpole

Computer Science Faculty Publications and Presentations

Applications with Constrained Latency Storage Access (CLSA) are those that have large storage needs and hard constraints on the amount of latency they can tolerate. Such applications present a problem when the storage technology that is cost effective and large enough cannot meet their latency constraints for demand fetching. Examples are found in the developing field of multimedia computing and, to a lesser extent, in real-time database literature. This paper examines the nature of timing constraints at the application-storage interface and defines a classification for both the synchronization constraints of the application and the latency characteristics of the storage system. …

Concurrency Control, Version Management And Transactions In Advanced Database Systems, Jonathan Walpole, Muntuck Yap

Computer Science Faculty Publications and Presentations

This document constitutes the final deliverable for the research project titled “An Investigation of Selected Issues in Transaction Mechanism Design for Object Oriented Databases.” The document describes our ideas for extending the traditional transaction concept for use in object oriented databases, and concentrates specifically on providing an underlying model to support the concurrency control and version management aspects of the problem. The ideas presented here are not however restricted to the domain of object oriented databases. They are more generally applicable to database systems that require flexibility in their versioning and concurrency control policies.

In this document we define a …

Benchmarking Parallel Machines Via A Software Architecture, Srikanth Kambhatla, Jon Inouye, Jonathan Walpole

Computer Science Faculty Publications and Presentations

The diversity in parallel architectures and the programming styles induced thereof, make benchmarking of parallel machines a hard problem. Current benchmarking techniques are limited in applicability due to the issues of portability. We present a different approach to benchmarking parallel machines, which would make the resulting benchmarks portable, easy to use and widely applicable. The approach involves the use of a software architecture to mask the diversities of the underlying architectures. The benchmarks are based on the Linda Tuple Space, and is called BeLinda. In this report, we motivate the use of a software architecture, give a specification of the …