Zvi Kedem

On shared memory parallel computers (SMPCs) it is natural to focus on decomposing the computation... more On shared memory parallel computers (SMPCs) it is natural to focus on decomposing the computation (mainly by distributing the iterations of the nested Do-Loops). In contrast, on distributed memory parallel computers (DMPCs) the decomposition of computation and the distribution of data must both be handled|in order to balance the computation load and to minimize the migration of data. We propose and validate experimentally a method for handling computations and data synergistically to optimize the overall execution time. The method relies on a number of novel techniques, also presented in this paper. The core idea is to rank the \importance" of data arrays in a program and de ne some of the dominant. The intuition is that the dominant arrays are the ones whose migration would be the most expensive. Using the correspondence between iteration space mapping vectors and distributed dimensions of the dominant data array in each nested Do-loop, we are able to design algorithms for determining data and computation decompositions at the same time. Based on data distribution, computation decomposition for each nested Do-loop is determined based on either the owner computes rule or the owner stores rule with respect to the dominant data array. If all temporal dependence relations across iteration partitions are regular, we use tiling to allow pipelining and overlapping the computation and communication time. However, to use tiling on DMPCs, we needed to extend the existing techniques for determining tiling vectors and tile sizes, as they were originally suited for SMPCs only. The method is illustrated on programs for the 2D heat equation and for the 2D fast Fourier transform both on a linear processor array.

Download

Persistent Storage Device Data Image Management System and Method

Methods for Handling Faults and Asynchrony in Parallel Computation

E cient and Secure Information Sharing in Distributed, Collaborative Environments

Algorithms and Techniques for Parallel Computers

Exploiting Application Tunability for Efficient, Predictable Parallel Resource Management

Parallel computing is becoming increasing central and mainstream, driven both by the widespread a... more

Knitting-Factory: An infrastructure for distributed web applications

... Additionally, if any Quality of Service parameters are given, the searcher script will only c... more

A theory of correct locking protocols for database systems

Proceedings of the Seventh International Conference on Very Large Data Bases Volume 7, 1981

Haya Shida, Subscribe (Full Service), Register (Limited Service, Free), Login. Search: The ACM Di... more

Data image management via emulation of non-volatile storage device

CALYPSO Goes to Wall Street: A Case Study

Page 1. Appeared in the 3rd International Conference on Artificial Intelligence Applications on W... more

Modeling Dynamics of Databases with Relational Discrete Event Systems and Models

Page 1. MODELING DYNAMICS OF DATABASES WITH RELATIONAL DISCRETE EVENT SYSTEMS AND MODELS Alexande... more

Fast Algorithms for Discovering the Maximum Frequent Set

... I would like to thank my parents, Chin-Yung Lin and Yeh San-May Lin, for v Page 6. their unco... more

Comments on the All Nearest-Neighbor Problem for Convex Polygons

Information Processing Letters, 1979

Predetermining Vis-ibility Priority in 3-D Scenes

Acm Siggraph Computer Graphics, 1979

Download

Parallel su x-pre x matching algorithm and application

Siam Journal on Computing, 1996

Exploiting Application Tunability for Efficient, Predictable Resource Management in Parallel and Distributed Systems

Jpdc, 1999

ABSTRACT Parallel computing is becoming increasing central and mainstream, driven both by the wid... more ABSTRACT Parallel computing is becoming increasing central and mainstream, driven both by the widespread availability of commodity SMP and high-performance cluster platforms, as well as the growing use of parallelism in general-purpose applications such as image recognition, virtual reality, and media processing. In addition to performance requirements, the latter computations impose soft real-time constraints, necessitating efficient, predictable parallel resource management. In this paper we propose a novel approach for increasing parallel system utilization while meeting application soft real-time deadlines. Our approach exploits the application tunability found in several general-purpose computations. Tunability refers to an application&#39;s ability to trade off resource requirements over time, while maintaining a desired level of output quality. We first describe language extensions to support tunability in the Calypso system, then characterize the performance benefits of tunability, using a synthetic task system to systematically identify its benefits. Our results show that application tunability is convenient to express and can significantly improve parallel system utilization for computations with predictability requirements

The concurrency control protocol: non-two-phase locking in general databases

Tods, 1990

Predetermining visibility priority in 3-D scenes (Preliminary Report)

Proceedings of the 6th Annual Conference, Aug 1, 1979

Download

The Milan Project: A New Method for High-Assurance and High-Performance Computing on Large-Scale Distributed Platforms

, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Manage... more

Download

Mapping Nested Loop Algorithms into Multi-Dimensional Systolic Arrays

Icpp, 1989

Consideration is given to transforming depth p-nested for loop algorithms into q-dimensional syst... more Consideration is given to transforming depth p-nested for loop algorithms into q-dimensional systolic VLSI arrays where 1&amp;amp;gt;or=q&amp;amp;gt;or=p-1. Previously, there existed complete characterizations of correct transformation only for the cases where q=p-1 orq=1. This gap is filled by giving formal necessary and sufficient conditions for correct transformation of a p-nested loop algorithm into a q-dimensional systolic array for any q,1&amp;amp;gt;or=q&amp;amp;gt;or=p-1. Practical

Automatic data and computation decomposition on distributed memory parallel computers

Acm Transactions on Programming Languages and Systems, 2002

Download

Persistent Storage Device Data Image Management System and Method

Methods for Handling Faults and Asynchrony in Parallel Computation

E cient and Secure Information Sharing in Distributed, Collaborative Environments

Algorithms and Techniques for Parallel Computers

Exploiting Application Tunability for Efficient, Predictable Parallel Resource Management

Parallel computing is becoming increasing central and mainstream, driven both by the widespread a... more

Knitting-Factory: An infrastructure for distributed web applications

... Additionally, if any Quality of Service parameters are given, the searcher script will only c... more

A theory of correct locking protocols for database systems

Proceedings of the Seventh International Conference on Very Large Data Bases Volume 7, 1981

Haya Shida, Subscribe (Full Service), Register (Limited Service, Free), Login. Search: The ACM Di... more

Data image management via emulation of non-volatile storage device

CALYPSO Goes to Wall Street: A Case Study

Page 1. Appeared in the 3rd International Conference on Artificial Intelligence Applications on W... more

Modeling Dynamics of Databases with Relational Discrete Event Systems and Models

Page 1. MODELING DYNAMICS OF DATABASES WITH RELATIONAL DISCRETE EVENT SYSTEMS AND MODELS Alexande... more

Fast Algorithms for Discovering the Maximum Frequent Set

... I would like to thank my parents, Chin-Yung Lin and Yeh San-May Lin, for v Page 6. their unco... more

Comments on the All Nearest-Neighbor Problem for Convex Polygons

Information Processing Letters, 1979

Predetermining Vis-ibility Priority in 3-D Scenes

Acm Siggraph Computer Graphics, 1979

Download

Parallel su x-pre x matching algorithm and application

Siam Journal on Computing, 1996

Exploiting Application Tunability for Efficient, Predictable Resource Management in Parallel and Distributed Systems

Jpdc, 1999

The concurrency control protocol: non-two-phase locking in general databases

Tods, 1990

Predetermining visibility priority in 3-D scenes (Preliminary Report)

Proceedings of the 6th Annual Conference, Aug 1, 1979

Download

The Milan Project: A New Method for High-Assurance and High-Performance Computing on Large-Scale Distributed Platforms

, 1215 Jefferson Davis Highway, Suite 1204, Arlington, VA 22202-4302, and to the Office of Manage... more

Download

Mapping Nested Loop Algorithms into Multi-Dimensional Systolic Arrays

Zvi Kedem

Related Authors

Uploads

Papers by Zvi Kedem

Log In