Amittai Aviram

Efficient system-enforced deterministic parallelism

Operating Systems Design and Implementation, Oct 4, 2010

Deterministic execution offers many benefits for debugging, fault tolerance, and security. Curren... more Deterministic execution offers many benefits for debugging, fault tolerance, and security. Current methods of executing parallel programs deterministically, however, often incur high costs, allow misbehaved software to defeat repeatability, and transform time-dependent races into input-or path-dependent races without eliminating them. We introduce a new parallel programming model addressing these issues, and use Determinator, a proof-of-concept OS, to demonstrate the model's practicality. Determinator's microkernel API provides only "shared-nothing" address spaces and deterministic interprocess communication primitives to make execution of all unprivileged code-well-behaved or notprecisely repeatable. Atop this microkernel, Determinator's user-level runtime adapts optimistic replication techniques to offer a private workspace model for both thread-level and process-level parallel programing. This model avoids the introduction of read/write data races, and converts write/write races into reliably-detected conflicts. Coarse-grained parallel benchmarks perform and scale comparably to nondeterministic systems, on both multicore PCs and across nodes in a distributed cluster.

Download

Efficient system-enforced deterministic parallelism

Communications of The ACM, May 1, 2012

Deterministic execution offers many benefits for debugging, fault tolerance, and security. Curren... more Deterministic execution offers many benefits for debugging, fault tolerance, and security. Current methods of executing parallel programs deterministically, however, often incur high costs, allow misbehaved software to defeat repeatability, and transform time-dependent races into input-or path-dependent races without eliminating them. We introduce a new parallel programming model addressing these issues, and use Determinator, a proof-of-concept OS, to demonstrate the model's practicality. Determinator's microkernel API provides only "shared-nothing" address spaces and deterministic interprocess communication primitives to make execution of all unprivileged code-well-behaved or notprecisely repeatable. Atop this microkernel, Determinator's user-level runtime adapts optimistic replication techniques to offer a private workspace model for both thread-level and process-level parallel programing. This model avoids the introduction of read/write data races, and converts write/write races into reliably-detected conflicts. Coarse-grained parallel benchmarks perform and scale comparably to nondeterministic systems, on both multicore PCs and across nodes in a distributed cluster.

Download

Literariness, Markedness, and Surprise in Poetry

Determinating Timing Channels in Compute Clouds

arXiv (Cornell University), Mar 27, 2010

Timing side-channels represent an insidious security challenge for cloud computing, because: (a) ... more Timing side-channels represent an insidious security challenge for cloud computing, because: (a) massive parallelism in the cloud makes timing channels pervasive and hard to control; (b) timing channels enable one customer to steal information from another without leaving a trail or raising alarms; (c) only the cloud provider can feasibly detect and report such attacks, but the provider's incentives are not to; and (d) resource partitioning schemes for timing channel control undermine statistical sharing efficiency, and, with it, the cloud computing business model. We propose a new approach to timing channel control, using provider-enforced deterministic execution instead of resource partitioning to eliminate timing channels within a shared cloud domain. Provider-enforced determinism prevents execution timing from affecting the results of a compute task, however large or parallel, ensuring that a task's outputs leak no timing information apart from explicit timing inputs and total compute duration. Experiments with a prototype OS for deterministic cloud computing suggest that such an approach may be practical and efficient. The OS supports deterministic versions of familiar APIs such as processes, threads, shared memory, and file systems, and runs coarse-grained parallel tasks as efficiently and scalably as current timing channel-ridden systems.

Download

Telling Rhythm: Body and Meaning in Poetry

Page 1. TELLING RHYTHM BODY AND MEANING IN POETRY Amittai E Aviram MICHIGAN Page 2. Page 3. Page ... more

Efficient System-Enforced Deterministic Parallelism

arXiv (Cornell University), May 19, 2010

Deterministic execution offers many benefits for debugging, fault tolerance, and security. Runnin... more Deterministic execution offers many benefits for debugging, fault tolerance, and security. Running parallel programs deterministically is usually difficult and costly, however-especially if we desire system-enforced determinism, ensuring precise repeatability of arbitrarily buggy or malicious software. Determinator is a novel operating system that enforces determinism on both multithreaded and multi-process computations. Determinator's kernel provides only single-threaded, "sharednothing" address spaces interacting via deterministic synchronization. An untrusted user-level runtime uses distributed computing techniques to emulate familiar abstractions such as Unix processes, file systems, and shared memory multithreading. The system runs parallel applications deterministically both on multicore PCs and across nodes in a cluster. Coarse-grained parallel benchmarks perform and scale comparably to-sometimes better than-conventional systems, though determinism is costly for fine-grained parallel applications.

Download

Lyric Poetry and Subjectivity

Intertexts, 2001

Postmodern Gay Dionysus: Dr. Frank N. Furter

The Journal of Popular Culture, Dec 1, 1992

Page 1. Postmodern Gay Dionysus: Dr. Frank N. Furter Amittai F. Aviram Now that the Rocky Horror ... more

Le rythme, pratique du ravissement ; la poésie, pontifex

Etudes Litteraires, Apr 12, 2005

Download

Deterministic Execution for Multicore and Cloud Computing

arXiv (Cornell University), Dec 4, 2009

The difficulty of developing reliable parallel software is generating interest in deterministic e... more The difficulty of developing reliable parallel software is generating interest in deterministic environments, where a given program and input can yield only one possible result. Languages or type systems can enforce determinism in new code, and runtime systems can impose synthetic schedules on legacy parallel code. To parallelize existing serial code, however, we would like a programming model that is naturally deterministic without language restrictions or artificial scheduling. We propose deterministic consistency, a parallel programming model as easy to understand as the "parallel assignment" construct in sequential languages such as Perl and JavaScript, where concurrent threads always read their inputs before writing shared outputs. DC supports common data-and task-parallel synchronization abstractions such as fork/join and barriers, as well as non-hierarchical structures such as producer/consumer pipelines and futures. A preliminary prototype suggests that softwareonly implementations of DC can run applications written for popular parallel environments such as OpenMP with low (< 10%) overhead for some applications.

Download

Determinating Timing Channels in Statistically Multiplexed Clouds

arXiv (Cornell University), Mar 27, 2010

Timing side-channels represent an insidious security challenge for cloud computing, because: (a) ... more Timing side-channels represent an insidious security challenge for cloud computing, because: (a) massive parallelism in the cloud makes timing channels pervasive and hard to control; (b) timing channels enable one customer to steal information from another without leaving a trail or raising alarms; (c) only the cloud provider can feasibly detect and report such attacks, but the provider's incentives are not to; and (d) resource partitioning schemes for timing channel control undermine statistical sharing efficiency, and, with it, the cloud computing business model. We propose a new approach to timing channel control, using provider-enforced deterministic execution instead of resource partitioning to eliminate timing channels within a shared cloud domain. Provider-enforced determinism prevents execution timing from affecting the results of a compute task, however large or parallel, ensuring that a task's outputs leak no timing information apart from explicit timing inputs and total compute duration. Experiments with a prototype OS for deterministic cloud computing suggest that such an approach may be practical and efficient. The OS supports deterministic versions of familiar APIs such as processes, threads, shared memory, and file systems, and runs coarse-grained parallel tasks as efficiently and scalably as current timing channel-ridden systems.

Download

Deterministic openmp

Deterministic OpenMP Amittai F. Aviram 2012 Researchers widely agree that determinism in parallel... more Deterministic OpenMP Amittai F. Aviram 2012 Researchers widely agree that determinism in parallel programs is desirable. Although experimental parallel programming languages have long featured deterministic semantics, in mainstream parallel environments, developers still build on nondeterministic constructs such as mutexes, leading to time-or schedule-dependent heisenbugs. To make deterministic programming more accessible, we introduce DOMP, a deterministic extension to OpenMP, preserving the familiarity of traditional languages such as C and Fortran, and maintaining source-compatibility with much of the existing OpenMP standard. Our analysis of parallel idioms used in 35 popular benchmarks suggests that the idioms used most often (89% of instances in the analyzed code) are either already expressible in OpenMP's deterministic subset (74%), or merely lack more general reduction (12%) or pipeline (3%) constructs. DOMP broadens OpenMP's deterministic subset with generalized reductions, and implements an efficient deterministic runtime that acts as a drop-in replacement for Gnu's widely used conventional OpenMP support library GOMP, on mainstream Unix platforms. We evaluate DOMP with several existing OpenMP benchmarks, each requiring under 50 source line changes and a majority requiring none, as well as several benchmarks we ported to OpenMP/DOMP. We find DOMP's efficiency and scalability comparable to GOMP in 7 of 11 benchmarks, suggesting that a deterministic model for mainstream parallel programming may be well within reach.

Download

Deterministic Consistency: A Programming Model for Shared Memory Parallelism

arXiv (Cornell University), Dec 4, 2009

The difficulty of developing reliable parallel software is generating interest in deterministic e... more The difficulty of developing reliable parallel software is generating interest in deterministic environments, where a given program and input can yield only one possible result. Languages or type systems can enforce determinism in new code, and runtime systems can impose synthetic schedules on legacy parallel code. To parallelize existing serial code, however, we would like a programming model that is naturally deterministic without language restrictions or artificial scheduling. We propose deterministic consistency, a parallel programming model as easy to understand as the "parallel assignment" construct in sequential languages such as Perl and JavaScript, where concurrent threads always read their inputs before writing shared outputs. DC supports common data-and task-parallel synchronization abstractions such as fork/join and barriers, as well as non-hierarchical structures such as producer/consumer pipelines and futures. A preliminary prototype suggests that softwareonly implementations of DC can run applications written for popular parallel environments such as OpenMP with low (< 10%) overhead for some applications.

Download

Telling Rhythm

Towards Generative Event Factuality Prediction

Findings of the Association for Computational Linguistics: ACL 2023

Deterministic Scheduling Does Not Eliminate Data Races

How to guarantee the deterministic execution of parallel programs and prevent race conditions and... more How to guarantee the deterministic execution of parallel programs and prevent race conditions and heisenbugs is now an important research topic. One obstacle is that conventional programming models rely on naturally nondeterministic synchronization abstractions. With naturally deterministic abstractions, programming logic alone, immune to timing, determines which threads synchronize, how, and at what point, making data races impossible. By contrast, nondeterministic primitives, such as mutex locks and condition variables, place the burden on the programmer to head off data races. New programming languages may address this problem, but legacy code must be rewritten. Alternatively, deterministic thread scheduling [1] handles conventional languages— making race conditions repeatable without eliminating them. A novel OS, Determinator, handles conventional languages and eliminates data races entirely, but only supports code using a handful of naturally deterministic synchronization abstr...

Abstract Deterministic OpenMP

Researchers widely agree that determinism in parallel programs is desirable. Although experimenta... more Researchers widely agree that determinism in parallel programs is desirable. Although experimental parallel programming languages have long featured deterministic semantics, in mainstream parallel environments, developers still build on nondeterministic constructs such as mutexes, leading to time- or schedule-dependent heisenbugs. To make deterministic programming more accessible, we introduce DOMP, a deterministic extension to OpenMP, preserving the familiarity of traditional languages such as C and Fortran, and maintaining source-compatibility with much of the existing OpenMP standard. Our analysis of parallel idioms used in 35 popular benchmarks suggests that the idioms used most often (89 % of instances in the analyzed code) are either already expressible in OpenMP’s deterministic subset (74%), or merely lack more general reduction (12%) or pipeline (3%) constructs. DOMP broadens OpenMP’s deterministic subset with generalized reductions, and implements an efficient deterministic...

Download

Workspace Consistency : A Programming Model for Shared Memory Parallelism

Recent interest in deterministic parallelism has yielded new deterministic programming languages,... more Recent interest in deterministic parallelism has yielded new deterministic programming languages, which offer promising features but require rewriting existing code, and deterministic schedulers, which emulate existing thread APIs but do not eliminate races from the basic programming model. Workspace consistency (WC) is a new synchronization and memory consistency model that offers a “naturally deterministic,” race-free programming model that can be adopted in both new or existing languages. WC’s basic semantics are inspired by—and intended to be as easily understood as— the “parallel assignment” construct in sequential languages such as Perl and JavaScript, where concurrent threads always read their inputs before writing shared outputs. Prototype implementations of a restricted form of WC already exist, supporting only strictly hierarchical fork/join-style synchronization, but this paper develops and explores the model in more detail and extends it to support non-hierarchical synch...

Download

Determinating Timing Channels in Statistically Multiplexed Clouds

ArXiv, 2010

Timing side-channels represent an insidious security challenge for cloud computing, because: (a) ... more Timing side-channels represent an insidious security challenge for cloud computing, because: (a) massive parallelism in the cloud makes timing channels pervasive and hard to control; (b) timing channels enable one customer to steal information from another without leaving a trail or raising alarms; (c) only the cloud provider can feasibly detect and report such attacks, but the provider's incentives are not to; and (d) resource partitioning schemes for timing channel control undermine statistical sharing efficiency, and, with it, the cloud computing business model. We propose a new approach to timing channel control, using provider-enforced deterministic execution instead of resource partitioning to eliminate timing channels within a shared cloud domain. Provider-enforced determinism prevents execution timing from affecting the results of a compute task, however large or parallel, ensuring that a task's outputs leak no timing information apart from explicit timing inputs and total compute duration. Experiments with a prototype OS for deterministic cloud computing suggest that such an approach may be practical and efficient. The OS supports deterministic versions of familiar APIs such as processes, threads, shared memory, and file systems, and runs coarse-grained parallel tasks as efficiently and scalably as current timing channel-ridden systems.

Download

Deterministic Consistency: A Programming Model for Shared Memory Parallelism

arXiv: Operating Systems, 2010

The difficulty of developing reliable parallel software is generating interest in deterministic e... more The difficulty of developing reliable parallel software is generating interest in deterministic environments, where a given program and input can yield only one possible result. Languages or type systems can enforce determinism in new code, and runtime systems can impose synthetic schedules on legacy parallel code. To parallelize existing serial code, however, we would like a programming model that is naturally deterministic without language restrictions or artificial scheduling. We propose deterministic consistency, a parallel programming model as easy to understand as the “parallel assignment” construct in sequential languages such as Perl and JavaScript, where concurrent threads always read their inputs before writing shared outputs. DC supports common data- and task-parallel synchronization abstractions such as fork/join and barriers, as well as non-hierarchical structures such as producer/consumer pipelines and futures. A preliminary prototype suggests that softwareonly impleme...

Download

Uploads

Papers by Amittai Aviram

Log In