Recently, lightweight thread libraries have become a common entity to support concurrent programming on shared memory multiprocessors. However, the disparity between primitives offered by operating systems creates a challenge for those who wish to create portable lightweight thread packages. What should be the interface between the machine-independent and machine-dependent parts of the thread library ? We have implemented a portable lightweight thread library on top of Unix on a KSR-1 supercomputer, BBN Butterfly multiprocessor, SGI multiprocessor, Sequent multiprocessor and Sun 3/4 family of uniprocessors. This paper first compares the nature and performance of the OS primitives offered by these machines. We then present a procedure-level abstraction that is efficiently implementable on all the architectures and is a sufficient base upon which a user-level thread package can be built. College of Computing Georgia Institute of Technology Atlanta, Georgia 30332--0280 1 Introduction T...

Blocking locks are commonly used in parallel programs to improve application performance and system throughput. However, most implementations of such locks suffer from two major problems -- latency and scalability. In this paper, we propose an implementation of blocking locks using scheduler adaptation which exploits the interaction between thread schedulers and locks. By experimentation using wellknown multiprocessor applications on a KSR2 multiprocessor, we demonstrate how such an implementation considerably reduces the latency and improves the scalability of blocking locks.

This paper presents the case for a run-time operating system to provide predictable and controllable resource management for parallel applications. Developments in compilers for parallel applications, which depend on analysis of program behaviour, require corresponding developments in predictable and controllable operating systemlevel resource management. A customisable system, Arena, is described which provides operating system-level resource management at user-level, where it is accessible to application and run-time system developers. This resource management is accessed via libraries, and effectively becomes part of the application run-time system. In such a system the application gets only the resource management that it needs. A low-policy hardware-dependent executive provides a lowlevel interface presenting an abstraction of processor hardware to the hardware-independent resource managers. The use of Arena on a Sparc-based distributed store multicomputer is described, and three...

ions: A Case Study of Multiprocessor Locks Bodhisattwa Mukherjee (bodhi@cc.gatech.edu) Karsten Schwan (schwan@cc.gatech.edu) GIT--CC--94/39 10 June 1994 Abstract Operating system kernels typically offer a fixed and limited set of primitives and underlying mechanisms for use by application programs. However, the attainment of high performance for a variety of parallel applications may require the availability of additional primitives or of variants of existing primitives best suited for specific applications. Furthermore, operating system mechanisms must also cope with changes over time in the behavior of individual applications. One approach to addressing these needs is to construct configurable operating system kernels, where configurability of operating system mechanisms permits their change to suit varying application needs. This paper addresses operating system configuration by considering specific abstractions. Specifically, we first motivate dynamic configuration by experim...

. The flexible job-oriented program model and distributed top-layer architecture described in this paper represent a novel top-layer approach to real-time software design and implementation, which can achieve portable, adaptable, fault-tolerant, and predictable high-value performance for a significant class of large-scale real-time systems with dynamic requirements, resources and constraints. Following a brief introduction to real-time software control issues and current layer-by-layer trends and limitations, we present an adaptive toplayer alternative to program modeling and control which can efficiently guarantee dynamic hard and soft requirements on a distributed platform, while providing best-effort system values for arbitrary system state sequences. Practical and scalable control algorithms have been devised, analyzed, and tested, which strongly suggest that such an approach is viable even for dynamic large-scale and complex systems where conventional layer-by-layer alte...

Abstract. The exible job-oriented program model and distributed top-layer architecture described in this paper represent anovel top-layer approach to real-time software design and implementation, which can achieve portable, adaptable, fault-tolerant, and predictable high-value performance for a signi cant class of large-scale real-time systems with dynamic requirements, resources and constraints. Following a brief introduction to real-time software control issues and current layer-by-layer trends and limitations, we present an adaptive toplayer alternative to program modeling and control which can e ciently guarantee dynamic hard and soft requirements on a distributed platform, while providing best-e ort system values for arbitrary system state sequences. Practical and scalable control algorithms have been devised, analyzed, and tested, which strongly suggest that such an approach is viable even for dynamic large-scale and complex systems where conventional layer-by-layer alternatives fail. We also show howthe proposed top-layer architecture might e ciently accommodate known levels of non-deterministic behavior within the platform and the environment. A musical ATLAS testbed is being developed to illustrate the feasibility of this approach.

Abstract not found