The proceedings contains 53 papers from the twenty-secondannual ACM symposium on principles of distributedcomputing, PODC 2003. The topics discussed include: simple and fast optimistic protocols for fair electronic ...
详细信息
The proceedings contains 53 papers from the twenty-secondannual ACM symposium on principles of distributedcomputing, PODC 2003. The topics discussed include: simple and fast optimistic protocols for fair electronic exchange;distributed error confinement;broadcasting in undirected ad hoc radio networks;scalable and dynamic quorum systems;routing networks for distributed hash tables;asynchronous resource discovery;and scalable public key tracing and revoking.
second-order optimization methods have been developed to enhance convergence and generalization in deep neural network (DNN) training compared to first-order methods like Stochastic Gradient Descent (SGD). However, th...
详细信息
ISBN:
(纸本)9798400714436
second-order optimization methods have been developed to enhance convergence and generalization in deep neural network (DNN) training compared to first-order methods like Stochastic Gradient Descent (SGD). However, these methods face challenges in distributed settings due to high communication overhead. Gradient compression, a technique commonly used to accelerate communication for first-order approaches, often results in low communication reduction ratios, decreased model accuracy, and/or high compression overhead when applied to second-order methods. To address these limitations, we introduce a novel gradient compression method for second-order optimizers called COMPSO. This method effectively reduces communication costs while preserving the advantages of second-order optimization. COMPSO employs stochastic rounding to maintain accuracy and filters out minor gradients to improve compression ratios. Additionally, we develop GPU optimizations to minimize compression overhead and performance modeling to ensure end-to-end performance gains across various systems. Evaluation of COMPSO on different DNN models shows that it achieves a compression ratio of 22.1x, reduces communication time by 14.2x, and improves overall performance by 1.9x, all without any drop in model accuracy.
The 26 papers in this volume deal with principles of distributedcomputing. Topics covered include distributed Networks;Concurrency Controls;distributed Databases;and Operating Systems.
ISBN:
(纸本)0897911105
The 26 papers in this volume deal with principles of distributedcomputing. Topics covered include distributed Networks;Concurrency Controls;distributed Databases;and Operating Systems.
The most significant strategic development in information technology over the past year has been 'trusted computing'. This is popularly associated with Microsoft's 'Palladium' project, recently ren...
详细信息
The most significant strategic development in information technology over the past year has been 'trusted computing'. This is popularly associated with Microsoft's 'Palladium' project, recently renamed 'NGSCB'. In this paper, I give an outline of the technical aspects of 'trusted computing' and sketch some of the public policy consequences.
The latest specifications for dynamic group communication in a distributed system are discussed. The reliable broadcast in a dynamic system is called reliable multicast, which is defined by the two primitives rmultica...
详细信息
The latest specifications for dynamic group communication in a distributed system are discussed. The reliable broadcast in a dynamic system is called reliable multicast, which is defined by the two primitives rmulticast and rdeliver. The properties of the reliable multicast include validity, uniform agreement, uniform integrity and uniform same view delivery. For the validity property, if a correct process executes rmulticast(m), then it eventually rdelivers m.
The notions of refinement for branching time based on stuttering simulation and bisimulation are developed. A local proof rule called well-founded simulation is introduced. It is shown that if one system refines anoth...
详细信息
The notions of refinement for branching time based on stuttering simulation and bisimulation are developed. A local proof rule called well-founded simulation is introduced. It is shown that if one system refines another, then a refinement map always exists. The added generality is obtained by using relations to relate implementation states with specification states and by requiring the use of well-founded relations.
In 2004, Klavins et al. introduced the use of graph grammars to describe-and to program-systems of self-assembly. It turns out that these graph grammars can be embedded in a graph rewriting characterization of distrib...
详细信息
ISBN:
(纸本)9781605583969
In 2004, Klavins et al. introduced the use of graph grammars to describe-and to program-systems of self-assembly. It turns out that these graph grammars can be embedded in a graph rewriting characterization of distributed systems that was proposed by Degano and Montanari over twenty years ago. By applying techniques obtained from this observation, we prove a generalized version of Soloveichik and Winfree's theorem on local determinism. We also obtain a canonical method to simulate asynchronous constant-size-message-passing models of distributedcomputing with systems of self-assembly.
The extension of Structured Query Language (SQL) access control approaches in deriving data from distributed systems was discussed. The three component features of the discussed SQL model included: applicability of de...
详细信息
The extension of Structured Query Language (SQL) access control approaches in deriving data from distributed systems was discussed. The three component features of the discussed SQL model included: applicability of derived object concept in any architecture;privileges usable only within a particular derivation;and the use of factored privileges in modeling different execution concerns. The stated features were combined in representing the policies specifiable in researchers' models. The use of the three components of the model in exporting the data of a national network of hospitals to a data warehouse was discussed.
暂无评论