Showing papers on "Scalability published in 1986"

A scalable dataflow structure store

Abstract: A design for a highly parallel data structure store for the prototype Manchester Dataflow Computer is presented. The main design objective is to allow all storage functions to be performed concurrently. The functions include space allocation and garbage collection, as well as operations for reading and writing the store. The resulting design is scalable in that an incremental increase in performance for any function can be achieved by adding appropriate hardware modules into the system. A relative balance in performance between the different functions can therefore be achieved. In the hardware structure store, the logical and the physical function units are designed separately. This increases flexibility for configuring multiprocessor dataflow systems. A single hardware module performing all the functions has been constructed and attached to the prototype Manchester Dataflow Machine. The performance of this configuration is discussed.

Mokken Scale: A Pascal Program for Nonparametric Stochastic Scaling

Abstract: A nonparametric Mokken scale analysis (a stochastic elaboration of the deterministic Guttman scale) and program is described. Focus has been placed on three procedures of scaling: a search procedure, evaluation of the whole set of items as a scale and extension of an existing scale. All procedures provided a coefficient of scalability for all the items together that satisfy the criteria of the Mokken model, and an item coefficient of scalability for every item and the statistic delta star which provides information whether the null hypothesis of random response may be rejected. The program is written in Pascal, and it is usable on many microcomputers and larger systems.

Multi-Flex Machines Preliminary Report

Abstract: The multi-FLEX machines use FLEX/32 multi-computer modules for larger machines. These machines provide wide flexibility and scalability. the innovative ingredients of the design are: (a) use of locally shared memory in addition to local and global memory, (b) very high internal and external bandwidth, (c) no use of network prolocals for communication, Cd) distribution of external I/O throughout the machine. The 64 node FLEX-<:ube would have 1500 processors and 700 megawords of memory, operate at 2.5 gigaflops and have external I/O bandwidth of 4 Gigabytes/second. It is conjectured that it could service 40,000 tenninals or service one job to do weather forecasts 1000 times faster than real time (covering the entire nonhern hemisphere with about 500 million unknowns.) •Supported in part by Air Foroe OffiC& of Scientific RllSClIrcl1 gnml AFOSR-84-0385 and AlTTIY Resea.rcl1 Office conlT1lCl = 2I. RATIONALE FOR THE MULTI·FLEX MACHINES This main rationale for the class of machines described here is to provide wide flexibility and scalability in applications. The FLEX-cube described can, it is conjectured, either do a weather forecast 1000 times faster than real time for the northern hemisphere or service 40,000 terminals (it is a 1500 processor, 2.5 Gigaflops machine). The innovative ingredients of this design are a) Memory heirnrchies of local, locally shared and global (there might be more than one level of locally shared memory) b) Very high internal and intermodule communication bandwidth. c) No use of network protocols for communication. d) External 110 distributed throughout the machine. The design is based on the existing commercial product, the FLEX/32 of Flexible Computer Corp. Other desirable properties of this design are a high level of modularity. fault tolerance and the use of ordinary packaging. At this point this is entirely a "paper" machine. However, real machines of this type can be constructed quickly (given the money) using existing machines. The main barrier to the use of the multi-FLEX machines is in the software systems. However, the situation here is no better nor worse than for other designs. IL THE FLEX/32 MODULE The multi-FLEX machines are built with the existing FLEX/32 machines as modules. We refer to [FLEX 86] for details of this machine, its characteristics are summarized here. Figure 1 shows a block diagram of the machine. For concreteness, we

A Rewrite Rule Machine. Architectural Options and Testbed Facilities for the Rewrite Rule Machine.

Abstract: : The Rewrite Rule Machine (RRM) project unites advanced architectural concepts with advanced software concepts. The unification is achieved through a novel model of computation, called concurrent tree rewriting, which supports both ultra high level programming and extreme concurrency of execution. This report presents some options for RRM implementation. We expect to use custom VLSI design to place many small processors on a single chip. With a suitable high-level architecture, this will allow unprecedented concurrency, with many thousands of small processors cooperating on a preduction task. Our performance estimate for a reasonable prototype is one million MIPS. The proposed RM architecture is described at four different levels. The individual tokens that constitute trees are stored in cells, while nodes consist of many cells with a common controller. Clusters include many nodes sharing a common name space for tokens, and several clusters together constitute the network level. Scalability and fault tolerance are also discussed, as is a testbed to ease the validation of architectural concepts. (Author)

Replication, reconfiguration, and the Argus mail repository

Abstract: The Argus Distributed Mail Repository is being implemented to gain experience with a sizable distributed application, to experiment with techniques for building distributed applications that are highly available and scalable, and to evaluate the programming language Argus when used for such applications.This note briefly describes Argus, the goals for the distributed repository, the project's status, and experience to date with the system.