Showing papers on "Shared resource published in 1982"

A comparison of two network-based file servers

Abstract: This paper compares two working network-based file servers, the Xerox Distributed File System (XDFS) implemented at the Xerox Palo Alto Research Center, and the Cambridge File Server (CFS) implemented at the Cambridge University Computer Laboratory. Both servers support concurrent random access to files using atomic transactions, both are connected to local area networks, and both have been in service long enough to enable us to draw lessons from them for future file servers.We compare the servers in terms of design goals, implementation issues, performance, and their relative successes and failures, and discuss what we would do differently next time.

PUMPS Architecture for Pattern Analysis and Image Database Management

Abstract: The PUMPS architecture consists of P task processing units (TPU) which share a pool of special peripheral processors, VLSI functional units, and a common two-dimensional shared memory (SM) via a block transfer oriented interconnection network. A shared cache is provided between the TPU's and SM for efficient MIMD interprocessor communication. The SM is also connected via a backend database management network (BDMN) with distributed control to the file memories, which are disk-based database storage devices.

Resource sharing in a network of personal computers

Abstract: : As networks of personal computers are developed to replace centralized time-shared systems, the need for sharing resources will remain, but the solutions developed for time-sharing will no longer be adequate. In particular, the sharing of network resources is complicated by issues of security and autonomy, since a network of personal computers may be composed of nodes that are completely controlled by their owners. To facilitate sharing in this sort of environment, an operating system component called the Butler is proposed. As a host, the Butler is responsible for administering a sharing policy on its local machine. This includes authenticating sharers, granting rights in accordance with a locally established policy, and creating execution environments for guests. As an agent, the Butler negotiates with hosts on remote machines to obtain resources requested by a client, and performs authentication to discourage a remote host from exploiting the client. A number of applications for the Butler are described: these fall into the categories of information exchange, load distribution, and computational parallelism. A prototype Butler has been constructed and used in a real application demonstrating computational parallelism, and the prototype has also demonstrated the deportation of processes.

A file server for a network of low cost personal microcomputers

Abstract: This paper describes a file server which was specifically designed to be the file store for a number of personal microcomputers attached to a local area network. In particular these personal machines (clients) have a very limited amount of memory with which to interface to the file server. The filing system supports an hierarchical directory structure with a simple capability-like protection mechanism. Both the operations and communication protocol supported by the file server are chosen to make a client computer's interface code simple. Special attention is given to speeding up certain frequently used processor-bound operations. Various aspects of the implementation are discussed, together with some suggested improvements.

Invariance in resource sharing problems

Abstract: Several processes or users share a common resource. Each process is described by a two state Markov chain with a "thinking" state and a service "request" state. The problem is to select a preemptive priority scheme to assign the resource among simultaneously requesting processes so as to maximize resource utilization. When the average thinking times of the processes are the same, the utilization is invariant under the assignment schemes. The invariance result extends to certain hitting time distributions for a broader class of Markov chains.

Shared resource locking controls

Abstract: The author describes how to synchronize input/output operations directed to a shared peripheral unit, such as a direct-access storage device (DASD), in a multiple host processor environment in a more efficient manner in terms of the number of accesses and the level of granularity than the reserve/release of the individual data storage unit.

Networks and flow control

Abstract: The clear advantages inherent in dynamic sharing of resources has accelerated the development of distributed processing. The number of networks which provide resource sharing have increased steadily from the early years of ARPA network. As the number of users sharing the resources increases, there is a potential danger for degradation of services, if the demand for the resources is greater than the available resources. The maximum traffic that can be carried by a network depends on the finite network resources such as link bandwidth and storage buffers. An uncontrolled flow of traffic results in service degradation and deadlock if the traffic allowed into the network exceeds the capacity of the network.

High level description and implementation of resource schedulers

Abstract: Resource sharing problems can be described in three basically independent components.• The constraints the resource places upon sharing because of physical limitations and consistency requirements.• The desired ordering of resource requests to achieve efficiency—either efficiency of resource utilization or efficiency for processes making the requests.• Modifications to the ordering, to prevent starvation of processes waiting for requests which might otherwise never receive service.A high level description language to specify these components of resource sharing problems is introduced. An implementation that lends itself to mechanical synthesis is described. Synthesis of the scheduler code by-passes the long and error- prone process of someone doing the coding themselves. Proof techniques at the high level description level are introduced to show how to prove schedulers, synthesized from their description, are or are not deadlock and starvation free. Solutions to the classical resource sharing problems of producer/consumer, reader/ writer, and disk scheduler (to the sector level) are shown to illustrate the expressiveness of this description language.

Computer Based Library Networks

Abstract: Cooperative on-line library networks represent both the highest technical development and the most concerted cooperative effort that librarians have achieved. Over 4,000 terminals link libraries to data bases containing millions of bibliographic records and library location symbols. Millions of on-line transactions are handled weekly as library staff members search, catalog, check in journals, order books, perform subject searches, and request interlibrary loans. The network is not monolithic. More than twenty cooperative networks and three network computer systems form a structure held together by contracts and held apart by competition, political and funding realities, and state and regional interests. Although many plans for a National Library Network have been promulgated, the present system cannot be said to have been planned. It is, rather, the result of many planning efforts which have somehow, through cooperation and economic necessity, intersected and interconnected. The system evolves. In an earlier paper1 I described three critical needs that on-line networks meet: research and development, capital acquisition, and technology transfer. Networks are also mechanisms for public policy and a response to the many criticisms and challenges that librarians and libraries had come in for over the past three decades. Through resource sharing networks, libraries can contribute to many state and national programs. This paper briefly discusses some aspects of the origins, governance, and operation of library networks.2 The reader needing additional information and other points of view may wish to consult the proceedings of two recent conferences, "The Structure

Modular expansion in a class of homogeneous networks

Abstract: We consider a special class of homogeneous computer network comprising several essentially identical but independent computing systems (ICSs) sharing a single resource. Of interest here are the effects of modularly expanding the network by adding ICSs. We use a previously presented approximate queueing network model to analyze modular expansion in this class of network. The performance measure used in this analysis is the mean cycle time, which is the mean time between successive requests for service by the same job at the CPU of an ICS. In this analysis we derive an intuitively satisfying mathematical relation between the addition of ICSs and the incremental increase in the service rate of the shared resource required to maintain the existing level of system performance.