content

1 面向并行工程的任务分配与规划

2 CUDA

3

***

面向并行工程的任务分配与规划

杨波, 黄克正, 孙红卫

计算机集成制造系统  2002 July

摘要:

   项目的成功依赖于全体参与者的协同工作, 但在并行工作环境下, 由于网络上的多功能小组工作风格
的差异, 因而在产品开发阶段经常出现任务规划的交叠和冲突现象。为解决这些问题, 保证资源的合理利用及缩短开发时间, 本文提出了在产品开发过程中任务分解的原则, 给出了任务到团队及基于均衡- 适度原则的任务到团队内各个参与人员分配的数学模型, 该模型有效地支持了并行工程中的资源配置。

关键词: 并行工程; 任务分解; 团队; 任务分配

0 引言

   任务分配与规划: 在众多的匹配方法中寻找一个最合理的子任务分配方案

1 支持并行工程的产品开发过程

2 任务的分解

3 任务的分配    

   并行设计本质: 充分利用网络上的资源和参与设计各成员的领域知识进行协同作业

3.2 任务到团队内Agent 成员的分配

面向并行工程的任务分配与规划.pdf

[2] CUDA

五种主要多核并行编程方法分析与比较

   http://www.cnblogs.com/jpcflyer/archive/2012/02/18/2356906.html

   -- MPI(Message Passing Interface)

   -- OpenMP(Open Multi Processing)

   -- Intel IPP(Integrated Performance Primitives)

   -- Intel TBB(Threading Building Blocks)

   -- MapReduce

   并行编程初步 (张丹丹 上海超级计算中心 2011-3-4)   MPI编程初步.pdf

       MPI: 一种消息传递编程模型

           Open MPI:  Open Source High Performance Computing  http://www.open-mpi.org/

***

Concurrency and the Algebraic Theory of Effects
(Abstract)
Gordon D. Plotkin
LFCS, School of Informatics, University of Edinburgh
M. Koutny and I. Ulidowski (Eds.): CONCUR 2012, LNCS 7454, pp. 21–22, 2012.
Springer-Verlag Berlin Heidelberg 2012

The algebraic theory of effects [7,8,4] continues Moggi’s monadic approach to
effects [5,6,1] by concentrating on a particular class of monads: the algebraic ones,that is, the free algebra monads of given equational theories. The operations of
such equational theories can be thought of as effect constructors, as it is they
that give rise to effects. Examples include exceptions (when the theory is that of
a set of constants with no axioms), nondeterminism (when the theory could be
that of a semilattice, for nondeterminism, with a zero, for deadlock), and action
(when the theory could be a set of unary operations with no axioms).

Two natural, apparently unrelated, questions arise: how can exception handlers
and how can concurrency combinators be incorporated into this picture?
For the first, in previous work with Pretnar [9], we showed how free algebras
give rise to a natural notion of computation handling, generalising Benton and
Kennedy’s exception handling construct. The idea is that the action of a unary
deconstructor on a computation (identified as an element of a free algebra) is the
application to it of a homomorphism, with the homomorphism being obtained
via the universal characterisation of the free algebra. This can be thought of as
an account of unary effect deconstructors.

In general, such unary deconstructors can be defined using parameters, and simultaneously defined unary deconstructors are also possible. The more complex
definitions are reduced to the simple homomorphic ones by using homomorphisms
to power and product algebras. This is entirely analogous to treatments
of (simultaneous) primitive recursion on natural numbers, or of structural recursion
on lists.

For the second, turning to, for example, CCS, the evident theory, at least for
strong bisimulation, is that of nondeterminism plus action. Then restriction and
relabelling are straightforwardly dealt with as unary deconstructors. However
the concurrency combinator is naturally thought of as a binary deconstructor
and the question arises as to how, if at all, one might understand it, and similar binary operators, in terms of homomorphisms. This question was already posed
in [9] in the cases of the CCS concurrency and the Unix pipe combinators. In
addition, a treatment of CSP in terms of constructors and deconstructors was
given in [10], but again still leaving open the question of how to treat concurrency.
Following an idea found in the ACP literature [2], concurrency combinators
can generally be split into a sum of left and right combinators, according to
which of their arguments’ actions occur first. This leads to a natural simultaneous
recursive definition of the left and right combinators, with a symmetry
between recursion variables and parameters; however the definition is not in the
form required for unary deconstructors. We give a general theory of such binary
deconstructors, in which solutions to the equations are found by breaking the
symmetry and defining unary deconstructors with higher-order parameters.1 The
theory applies to CCS and other process calculi, as well as to shared memory parallelism.
In this way we demonstrate a possibility: that the monadic approach,
which has always included (algebraic) monads for interaction and for shared
variable parallelism—see [1,3,4]—can be fruitfully integrated with the world of
concurrency.

Concurrency and the Algebraic Theory of Effects.pdf

***

Concurrency Meets Probability: Theory and Practice
(Abstract)

Joost-Pieter Katoen

   probabilistic concurrency models -> continuous stochastic phenomena

   Markov automata model

   Markov automata

Concurrency Meets Probability  Theory and Practice.pdf