Andrew Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. 2nd Ed., 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....actors by sending messages, and all actor actions are in response to these messages. Specific actor methods are invoked to process each type of message, and actors are not allowed to block or explicitly make receive requests from other processors. Also concurrent abstractions known as aggregates [18] are available in ProperCAD II to support a multi access interface to groups of actors. 1.2 Preliminaries In this section we review some basic definitions and then describe the motivation for the efficient State Assignment of Finite State Machines. 1.2.1 Definitions A variable is a symbol ....

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. Cambridge, MA: The MIT Press, 1993.

....as in a standalone environment. We further hypothesize that in the future, multicomputer applications will communicate more than current multicomputer applications. This hypothesis is based on two factors: First, if fine grained parallel programming languages such as Id [1] Concurrent Aggregates [6], and Multilisp [20] gain more acceptance, there will be an increased reliance on communication. And second, as massively parallel computers become commonplace, each mode has at least the potential for increased communication (i.e. there are more nodes with which to communicate) But, as Figures ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....parallel apply operations performed over data collections (typically vectors) nested parallelism is allowed, but is compiled into non nested, vector parallel operations. None of these languages is well suited to distributing objects non uniformly over a multiprocessor. Concurrent Aggregates [10] and Concurrent Smalltalk [24, 23] are languages specifically intended for the J machine. Each is based on the notion of essentially replicating important data structures spatially in order to provide means for parallel access. The languages do not explicitly reveal distributed object placement to ....

Andrew Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993.

....parallel apply operations performed over data collections (typically vectors) nested parallelism is allowed, but is compiled into non nested, vector parallel operations. None of these languages is well suited to distributing objects non uniformly over a multiprocessor. Concurrent Aggregates [10] and Concurrent Smalltalk [24, 23] are languages specifically intended for the J machine. Each is based on the notion of essentially replicating important data structures spatially in order to provide means for parallel access. The languages do not explicitly reveal distributed object placement to ....

Andrew Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993.

....the Actor model of computation have been developed in di erent object oriented languages. Three examples of these frameworks are the Actor Foundry [29] Actalk [7] and Broadway [33] Several actor languages have also been proposed and implemented to date, including ABCL [39] Concurrent Aggregates [12], Rosette [34] and Thal [25] There are several advantages associated with directly using an actor programming language, as compared to using a library for actors: Semantic constraints: Certain semantic properties can be guaranteed at the language level. For example, an important property is ....

A. Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993.

....focused on designing parallel extensions for C [21] Several groups have joined to define standard library and language extensions for writing portable, parallel C applications. In Europe, the Europa consortium [16] has defined a model of parallel C computation based on Active Objects [1, 6, 8] and a meta object protocol derived from the work on reflection in the programming language research community [12, 17] In Japan, the Real World Computing Partnership has established the MPC programming system [11] which provides broad and powerful mechanisms for user level extensions to a C ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, 1993.

....mechanisms alone cannot describe the irregular blocking structures that arise in adaptive mesh refinement and recursive coordinate bisection [44] Data Parallel C Languages The pC [30, 31, 144] programming language is a data parallel extension of C . It implements a concurrent aggregate [47] model in which a parallel operation is applied simultaneously to all elements of a data aggregate called a collection. 4 Information about the second High Performance Fortran standardization effort can be found at World Wide Web address ftp: hpsl.cs.umd.edu pub hpf bench index.html. 40 Each ....

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs, MIT Press, 1993. 202

....be the T3D. Information, the language report, and the current release of the Concert software can be found at: http: www csag.cs.uiuc.edu ftp: cs.uiuc.edu pub csag Email addresses: group Gamma concert red herring cs.uiuc.edu Andrew A. Chien Gamma achien cs.uiuc.edu References: 65] [66] [67] 68] 69] 70] 71] 129] 177] 2.29 ConcurrentSmalltalk Developer: Description: oo. memory model. parallelism. Asynchronous method call plus futures (CBox) Post processing. Synchronous messages are also available. The caller decides which mode to use. scheduling. mapping. ....

Andrew A. Chien, editor. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press Cambridge, Massachusetts, London, England, 1993.

....object oriented model allows data abstraction to be used via modularity and information hiding to manage the complexity. Our concurrent object oriented model is extended with aggregates, parallel collections of objects, which enable the convenient expression of data parallelism as task parallelism [8]. The underlying compiler and runtime are responsible for providing a global object namespace that spans the machine, distributing the parallel collections of objects, and scheduling the dynamically created threads on the nodes of the parallel machine. We believe this programming model is ....

....model provides a shared object namespace, expression of both task and data parallelism, and dynamic thread creation. This allows the programmer to express parallelism conveniently and portably. We give a brief description of the programming model here, more detailed discussions can be found in [8, 9, 4]. 3.2 Programming Model The Concert programming model is embodied in the language, Concurrent Aggregates [9] The language provides a simple concurrent object oriented programming model. Program data is divided into objects, and the computation is expressed as operations on objects, much as in a ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....not allowed to block or explicitly make receive requests from other processors. The runtime system on each processor picks the next available actor thread with some priority and that thread is then allowed to run to completion without interruption. Also concurrent abstractions known as aggregates [27] are available in ProperCAD II to support a multi access interface to groups of actors. The implementation of the actor interface is defined in terms of the APA, a low level interface which provides an interface and implementation which can be used to describe and utilize resources needed by any ....

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. Cambridge, MA: The MIT Press, 1993.

....is a first step towards providing a formal semantics for open distributed systems, and a basis for specifying and reasoning about properties of and interactions between components of such systems. Application concurrency is provided by paradigms such as concurrent object oriented programming [11, 177, 3, 49]. However, in a distributed system, various applications share a common underlying set of resources. The focus of the research reported in this thesis is on resource management concurrency that allows us to exploit concurrency in using distributed resources (for performance and QoS) without ....

A. A. Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993.

....target problems. 1.3 Summary of Contributions The primary contributions of this thesis are: 1. A class library interface for actor based parallelism in a statically typed language and based on built in type mechanisms. 2. An implementation of aggregates providing the functionality described in [5] with additional meta programmability features. 3. An interface supporting composable meta programming of an actor system on contemporary microprocessor based machines. 4. An open implementation supporting application specific customization of the run time support system. 5. An abstract parallel ....

....expressing concurrency. The purpose of the high level interface is to insulate the application and by extension, the developer from the details of the underlying hardware without reducing parallelism. The AIF is based closely on the Actor model [51] and supports such extensions as aggregates [5] and meta programmability. In this chapter, we consider the basic actor interface; in Chapter 5, we consider advanced meta programmability features which facilitate tuning the operation and performance of the run time support. To exemplify use of the Actor Interface, we will use examples from ....

[Article contains additional citation context not shown here]

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. Cambridge, MA: The MIT Press, 1993.

....When an object of a subclass receives a message, it tries to respond to it using its own methods. If it fails, it climbs up the inheritance tree to respond to the message. Delegation implements the prototype approach to sharing knowledge in object oriented systems. It appears in actor languages [88, 135, 29] and several Lisp based object oriented systems such as Director [72] T [67] Orbit [113] and others. An object shares knowledge with a prototype by simply having the prototype as its acquaintance; the object may also keep its personal behavior idiosyncratic to itself. When it receives a ....

....its presence manifest only when a message is scheduled. 2. 3 Actor Based Languages The Actor model was first introduced by Carl Hewitt [53] refined by many others [55, 54, 30, 56] and defined in its current standard form by Agha [1] Particularly, since its introduction many actor based languages [121, 13, 88, 16, 11, 85, 135, 89, 70, 29] have been proposed for programming concurrent computation. Act1 [88] was an early actor language which was implemented in Lisp. It supported a number of abstractions which are still found in other contemporary actor based languages. For example, it used continuations to support bidirectional ....

[Article contains additional citation context not shown here]

A. A. Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. MIT Press, 1993.

....methods for accessing the continuations. An example of the CTableData class is show in Figure 4. An instantiation of the CTable class is a member of the UserInterface actor. This is the object which is shown being initialized in Figure 3. As soon as this initialization is completed, an aggregate [8, 18] is created, with the continuation table as a data member. As soon as the UserInterface knows that all actors have been successfully created, it creates the Continuation Table aggregate, with a representative on each processor. Once this has been created, the initSim actor on each of the ....

A. A. Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. The MIT Press, 1993.

....which communicate by executing continuations, the parallel, object oriented extrapolation of remote procedure calls. The high level actor model obviates the need for application designers to deal with low level mutual exclusion primitives. The actor model has been extended to include aggregates [17], collections of actors, individually called representatives. The ProperCAD II library supports actor semantics via Actor and Aggregate base classes. For example, in ProperPROOFS the main computation aggregate class is FaultSimulatorAggregate, derived from the ProperCAD II Aggregate class and ....

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. Cambridge, MA: The MIT Press, 1993.

....Individual actors express neither internal parallelism nor data distribution. Collection types, based on aggregates with explicit distributions, allow both object internal concurrency as well as data distribution. An aggregate is simply a collection or group of actors which share a common name [19]. An example of an aggregate would be a distributed array where different elements are stored on different actors. The use of aggregate representations removes the serializa7 tion step that would be required because of a gateway actor. The interface is similar to that of the Actor class, and the ....

....for parallel cell placement described in Chapter 6. 1.2. 3 Related work in parallel software environments Several other researchers have produced work in environments to support irregular applications in object oriented environments, such as Charm [20] CC [21] Concurrent Aggregates Concert [19, 22, 23], pC [24] and SDDG DAGH [25] Charm provides similar run time support for message driven applications to ProperCAD II. The primary differences are Charm s lack of support for static message typing, as represented by first class continuations, and composability. Concurrent Aggregates is a ....

[Article contains additional citation context not shown here]

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. Cambridge, MA: The MIT Press, 1993.

....processors are packed. 2.2.2.2 Related work in concurrent objects Dynamic object creation is provided in ICC [17] Mentat [18] C [19] C [17] ABCL [20] and COOL [21] among others. Mentat supports coarse grained objects, while ICC and languages such as Cantor [22] HAL [23] and CA [24] support fine grained objects. Compilers for fine grained languages typically coalesce multiple objects and operations into coarser ones in order to achieve acceptable performance on available parallel machines. While Charm does not allow parallelism within an object (i.e. all methods are ....

A.A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, 1993.

....actors express neither internal parallelism nor data distribution. Collection types, based on aggregates with explicit distributions, facilitate both object internal parallelism and data distribution. Aggregates and AggregateNames: An aggregate is a collection of actors which share a common name [7]. When a continuation created from an aggregate name is executed, the appropriate actor method is executed by one or more representative actors. An example of the use of aggregates is a distributed array where non overlapping ranges of elements are stored in individual actors. A distributed array ....

....nature of a continuation is inherited from the name used to create the continuation, the unicast broadcast choice can be made by the creator of the name. By contrast, in the original aggregate model the unicast broadcast choice must be made at the point where the continuation is called [7]. The ability to encapsulate this information in continuation objects which are then passed to independent packages has considerable potential in the design of reusable library modules. Distributions: Distribution of aggregate representatives is specified via an optional argument supplied when ....

[Article contains additional citation context not shown here]

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. The MIT Press, 1993.

....providing much needed modularity in concurrent programs. Fine grained COOP languages are particularly promising since they expose a great deal of concurrency by allowing its expression at the object level. Numerous researchers have been exploring fine grained object oriented approaches [1, 2, 3, 4, 5]. Implementations of COOP languages face not only the challenges of object oriented languages, but also several additional challenges related to concurrency and distribution. 1 In particular, COOP language implementations must provide a global object namespace, implement the communication ....

....and Section 6 summarizes the paper and the current status of the project. 2 The Concert runtime This section surveys the design of the Concert runtime system. First, we briefly describe an abstract computation model which embodies characteristics common to many COOP languages proposed so far [2, 3, 4, 9, 5]. We then present the Concert runtime interface which provides the runtime operations required by the abstract model. 2.1 Computation model of COOP languages COOP languages express computation as a set of autonomous communicating entities (objects) Generally, these objects reside in a globally ....

[Article contains additional citation context not shown here]

A. A. Chien, Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. Cambridge, MA: MIT Press, 1993.

....for several different source languages. This would demonstrate the genericity of the optimizations, as well as allow direct comparisons isolating the cost of programming language features. The specific language we are supporting initially is an extended version of Concurrent Aggregates (CA) [11]. As with most concurrent object oriented languages, CA augments the basic asynchronous message passing model with common idioms (such as RPC and tail forwarding [21] inheritance (a mechanism for code reuse) and some concurrency control constructs. Other novel aspects of Concurrent Aggregates ....

....languages can be an efficient medium in which to express a wide variety of parallel computations. The Concert system has been operational on both sequential and parallel platforms since October 1992. The system includes an optimizing compiler for an extended version of Concurrent Aggregates [11] and a high performance runtime system which runs on both Sun workstations and the Thinking Machines CM5 [25] We are currently building a second generation system with much greater analysis and optimization capability as well as much greater overall performance. This second generation Concert ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....is to develop portable efficient implementations of concurrent object oriented languages on parallel machines. This work includes a variety of research in program analysis, optimization and runtime techniques. 1 At present, the Concert system compiles the Concurrent Aggregates (CA) language [9, 10], a dynamically typed concurrent object oriented language with single inheritance as well as first class selectors, continuations, and messages for execution on the Thinking Machines CM5 [21] All program examples are written in Concurrent Aggregates. 2.2 Polymorphism We differentiate data ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....transport is the essence of messaging and flow control (buffer management) ensures that there is space for messages at the destination. Flow control is required by general messaging models such as ones supported by PVM [13] or MPI [12] and programming models which produce concurrency dynamically [6, 4]. Note that primitives such as PUT GET only specify data movement flow control is still necessary, the source and destination processors must interact to decide where the message must be buffered. If there is no space, the destination must be able to block the source. Programming models with ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....run time type errors in dynamically typed languages. Program examples in the running text will be in a syntax derived from [14] The target language of our implementation and in which the benchmarks are written is the dynamically typed concurrent object oriented language Concurrent Aggregates [7, 8] which includes first class selectors, messages and continuations. The algorithm is very general and can be applied easily to a wide range of languages. Before we begin, let us clarify some terms. We differentiate two types of polymorphism: data and functional. Data polymorphism includes ....

....a compiler and runtime for concurrent object 7 A constant number of recursive calls can be split at each site without affecting termination, however our implementation allows only one recursive call. oriented languages. The front end currently supports the language Concurrent Aggregates (CA) [7, 8], a dynamically typed concurrent object oriented language with single inheritance as well as first class selectors, continuations, and messages. We have tested the type inference system on more than 20,000 lines of CA code. The results on a variety of real and synthetic programs appear in Table 1. ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

....creates the callee s context, saves local state into it, and propagates the fall back by returning this context to its caller which then sets up the linkage. 3.2. 3 Continuation Passing: Lazy Continuation Creation Explicit continuation passing can improve the composability of concurrent programs [33, 6]. However, when continuation passing occurs, invocations on the stack are complicated because the callee may want its continuation. If the call is being executed on the stack, the callee s continuation is implicit. Since one of our goals is to execute forwarded invocations [20] on the stack, lazy ....

Andrew A. Chien. Concurrent Aggregates: Supporting Modularity in Massively-Parallel Programs. MIT Press, Cambridge, MA, 1993.

No context found.

Andrew Chien. Concurrent Aggregates: Supporting Modularity in Massively Parallel Programs. 2nd Ed., 1992.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC