Search

In this paper, we consider the performance of alternative mass storage services for a client-server style distributed system. Some qualitative arguments are presented on the ramifications of implementations of mass storage services at various levels of the storage semantics hierarchy. This paper concentrates, in particular, on contrasting disk and files services. The functionality of disk and file services are distinguished by their primitive operations: individual disk block access, for the disk service; and individual file block access, for the file service. This difference results in different partitionings of the computation between the client and server, as well as different network communication requirements. To understand the ramifications of such differences between the services, this paper presents performance estimates for basic disk and file services. To provide a commensurate comparison of the performance of the two services, a uniform closed queueing network model was developed that accurately portrayed this computation/communication tradeoff. Detailed parameterization of the model was done through measurements of prototype systems. The workload applied to the mass storage service is based on measurements of timesharing systems performed in earlier studies. In addition to the basic comparisons, performance estimates for several design alternatives are also presented. Some of these performance enhancing alternatives were: caching of file/disk blocks and file-opens, kernel-based services, and using a network interface able to do gather write to eliminate data copies. [ABSTRACT FROM AUTHOR]

Elaborates on a processor design called simultaneous multithreading (SMT), which exploits both thread-level and instructional-level parallelism. Information on how SMT works; Illustration of an SMT model; Implementation of a simulated environment; Results of the simulation exercise. INSET: Alernative approaches to exploiting parallelism..

Sparse tensor algorithms are becoming widespread, particularly in the domains of deep learning, graph and data analytics, and scientific computing. Current high-performance broad-domain architectures, such as GPUs, often suffer memory system inefficiencies by moving too much data or moving it too far through the memory hierarchy. To increase performance and efficiency, proposed domain-specific accelerators tailor their architectures to the data needs of a narrow application domain, but as a result cannot be applied to a wide range of algorithms or applications that contain a mix of sparse and dense algorithms. This article proposes Symphony, a hybrid programmable/specialized architecture that focuses on the orchestration of data throughout the memory hierarchy to simultaneously reduce the movement of unnecessary data and data movement distances. Key elements of the Symphony architecture include (1) specialized reconfigurable units aimed not only at roofline floating-point computations but also at supporting data orchestration features, such as address generation, data filtering, and sparse metadata processing; and (2) distribution of computation resources (both programmable and specialized) throughout the on-chip memory hierarchy. We demonstrate that Symphony can match non-programmable ASIC performance on sparse tensor algebra and provide 31× improved runtime and 44× improved energy over a comparably provisioned GPU for these applications. [ABSTRACT FROM AUTHOR]

The article describes improvements in computing power can claim a large share of the credit for many of the things that we take for granted in modern lives. Topics include cellphones that are more powerful than room-sized computers; drug discoveries enabled by powerful supercomputers; and semiconductor miniaturization is running out of steam as a viable way to grow computer performance.