Caching Single-Assignment Structures to Build a Robust Fine-Grain Multi-Threading System

نویسندگان

  • Wen-Yen Lin
  • Jean-Luc Gaudiot
  • José Nelson Amaral
  • Guang R. Gao
چکیده

We present the design, implementation, and evaluation of single assignment data structures and of a software controlled cache in an existing multi-threaded architecture platform { the E cient Architecture for Running Threads (EARTH). In EARTH, split-phase operations enable the tolerance of remote communication latency by a decoupling between the initiators and the receivers of communication/synchronization transactions. Under EARTH, the requester and receiver of a split-phase transaction are actually placed in separate threads. We focus on split-phase memory operations under a global shared address space. A software-controlled cache is proposed to exploit the temporal and spatial locality of singleassignment structure memory references. Our experimental evaluation using a set of benchmarks indicates that the EARTH system with both the single-assignment storage and its caching mechanism is more robust. As a consequence the system can be ported to a wider range of machine platforms and deliver speedup for both regular and irregular application.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Multigranular Thread Support in WaveScalar

WaveScalar is a recently proposed scalable microarchitecture. The original WaveScalar research developed and evaluated an ISA and microarchitecture that efficiently executes a single, coarse-grain thread. In this paper, we expand that design to support multiple, simultaneously executing threads. Four mechanisms make this possible: (1) instructions that enable and disable wave-ordered memory; (2...

متن کامل

Different Network Performance Measures in a Multi-Objective Traffic Assignment Problem

Traffic assignment algorithms are used to determine possible use of paths between origin-destination pairs and predict traffic flow in network links. One of the main deficiencies of ordinary traffic assignment methods is that in most of them one measure (mostly travel time) is usually included in objective function and other effective performance measures in traffic assignment are not considere...

متن کامل

THROOM — Running POSIX Multithreaded Binaries on a Cluster

Most software distributed shared memory systems (SW-DSMs) lack industry standard interfaces that limit their applicability to a small set of shared-memory applications. In order to gain general acceptance, SW-DSMs should support the same look-and-feel of shared memory as hardware DSMs. This paper presents a runtime system concept that enables unmodified POSIX P1003.1c (Pthreads) compliant binar...

متن کامل

Exploiting fine-grain thread parallelism on multicore architectures

In this work we present a runtime threading system which provides an efficient substrate for fine-grain parallelism, suitable for deployment in multicore platforms. Its architecture encompasses a number of optimizations that make it particularly effective in managing a large number of threads and with low overheads. The runtime system has been integrated into an OpenMP implementation to allow f...

متن کامل

A novel approach in robust group decision making for supply strategic planning

Long-term planning is a challenging process for dealing with problems in big industries. Quick and flexible process of responding to the existing variable requirements are considered in such problems. Some of important strategic decisions which should be made in this field are, namely the way that manufacturing facilities should be applied as well as assignment and design the system of delivery...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2000