| 1. |
- Maguire Jr., Gerald Q., et al.
(författare)
-
Process migration Effects on scientific computation
- 1987
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- This report describes the notion of process migration. and points out certain vulnerabilities to architectural assumptions.Computationally intensive processes, typified by scientific computations, are among the best candidates for migration. In architecturally heterogeneous systems, these processes face particular problems due to their dependence on architectural features such as "machine precision". We point out some of these problems. and suggest some solutions.
|
|
| 2. |
|
|
| 3. |
- Smith, Jonathan M., et al.
(författare)
-
Exploring "multiple worlds" in parallel
- 1989
-
Ingår i: International Conference on Parallel Processing. - New York, NY, USA : Columbia University, Department of Computer Science. ; , s. 239-245
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- We examine computing alternative solutions to a problem in parallel to improve response time.Problems with exploring multiple alternatives in parallel include (1) side-effects and (2) combinatorial explosion in the amount of state which must be preserved. These are solved by process management and an application of "copy-on-write" virtual memory management. The side effects resulting from interprocess communication are handled by a specialized message layer which interacts with process management The effect is to create "Multiple Worlds", which are different due to the alternate solution methods. but internally self-consistent.We show how such a "Multiple Worlds" scheme can be used for several applications.
|
|
| 4. |
- Smith, Jonathan M., et al.
(författare)
-
RB: Programmer Specification of Redundancy
- 1987
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- RB is a programming language for specifying redundancy in various dimensions. Avizienis's notation T I HIS, for Time I Hardware I Software, describes the different types of redundancy possible in a computation: repetition (nT I HIS), redundant hardware (T I nH I S), and program (software) (T I H InS). These can each be controlled by the programmer with RB. RB derives its name from its use of the recovery block notion to specify fault-tolerant segments of software. RB also supplies the programmer with the ability to specify degrees of replication and repetition for a given recovery block alternate; the underlying support software can then take this advice to replicate in time or hardware, based on available resources. An implementation of RB based on the C programming language is described in this paper. This implementation uses a combination of a language preprocessor for C and a runtime library to provide the desired semantics. Modification of RB to support other programming languages, or programmer specification of N-Version Programming as the decision mechanism. is straightforward.
|
|
| 5. |
- Smith, Jonathan M., et al.
(författare)
-
System support for "multiple worlds"
- 1989
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- Concurrently computing alternative solutions to a problem can be used as a method to improve response time. In order to maintain internal consistency despite differences caused by alternative solution methods, "Multiple Worlds" are created. This paper examines the operating system requirements posed by such a method. Problems include (1) side-effects and (2) combinatorial expldsion in the amount of state which must be preserved. We solve these by process management and an application of "copy-onwrite" virtual memory management. Side effects resulting from interprocess communication are handled by a specialized message layer which interacts with process management.
|
|
| 6. |
- Smith, Jonathan M., et al.
(författare)
-
The RB Language
- 1988
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- Typical algorithms for distributed or parallel computations are cooperative, meaning that the sequential component is broken down into cooperating pieces, which are distributed across available hardware. An approach wich has recently gained some attention is competitive processing, where several versions of a sequential program are distributed across available processors to gain performance from algorithmic diversity. There is also potential for fault tolerance from available hardware by executing the sequential versions, called alternatives, on a distributed configuration. Schemes for implementing competitive concurrent processing have been described in the literature, but there is little implementation experience. RB is a practical step towards gaining such experience. RB is a programming language for specifying alternative methods of per- forming a computation, where at most one of the results of the alternatives is used. Our prototype implementation uses a combination of a language preprocessor for C and a runtime library to provide the desired semantics. Using other base programming languages, e.g., Ada, or other methods of managing alternatives is straightforward.
|
|
| 7. |
- Smith, Jonathan M., et al.
(författare)
-
Transparent concurrent execution of mutually exclusive alternatives
- 1988
-
Rapport (övrigt vetenskapligt/konstnärligt)abstract
- We examine the task of concurrently computing alternative solutions to a problem. We restrict our interest to the case where only one of the solutions is needed; in this case we need some rule for selecting between the solutions. We use "fastest first," where the first successful alternative is selected. For problems where the required execution time is unpredictable, such as database queries. this method can show substantial execution time performance increases. These increases are dependent on the mean execution time of the alternatives, the fastest execution time, and the overhead involved in concurrent computation. Among the problems with exploring multiple alternatives in parallel are side-effects and combinatorial explosion in the amount of state which must be preserved. These are solved by process management and an application of "copy-on-write" virtual memory management. The side effects resulting from interprocess communication are handled by a specialized message layer which interacts with process management. In order to test the utility of the design, we show how it can be applied to two application areas, distributed execution of recovery blocks and OR-parallelism in Prolog.
|
|
