Chien, A. A., and Dally, W. J. Experience with concurrent aggregates (CA): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference (Apr. 1990), SIAM, pp. 1040--1049. 25  Home/Search   Document Not in Database   Summary   Related Articles   Check

....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. synchronization. There ....

Andrew A. Chien and William J. Dally. Experience with Concurrent Aggregates (CA): Implementation and programming. In Proc. of the 5th Distributed Memory Computer Conf., Charleston, SC, April 9-- 12, 1990.

....in a timestamped trace. Computation time was measured using a microsecond timer and scaled to simulate a processor executing approximately ten million instructions per second. A similar process was used to trace the execution of a multigrid algorithm implemented using Concurrent Aggregates (CA) [7], a concurrent programming language developed at MIT. Although Concurrent Aggregates and the Chare Kernel implement different computing environments, both use a similar set of abstractions, such as task creation and message transmission. The programs were traced by recording the occurrence of the ....

Chien, A. A., and Dally, W. J. Experience with Concurrent Aggregate (CA): Implementation and Programming. In Proceedings of the 5 th Distributed Memory Computing Conference (April 1990), Association For Computing Machinery, pp. 1040--1049.

....objects as multi ported objects and to each thread s version as a port. Implementations of this kind are prevalent in implementations of shared and distributed objects but we know of no other work that describes the specifications of the procedures that access these object. See [Ell85, Her90, CD90, WW90] for some examples of objects that contain thread specific data. The specification of a multi ported object typically has an interference specification that depends on the port being used, which is important because it means that a multi ported object looks different than a normal object, ....

....objects behave as if it were a single shared object. The most common example of a distributed object is a memory cell in a multiprocessor with caches, but many others have been designed and implemented for both distributed systems [Ell85, BT88, BHJ 87] and for multiprocessors [Luc87b, Dal86, CD90] To access a distributed object, these systems provide sophisticated run time support so that each object is given a single name; the run time system must determine what node in the distributed object should be used for a particular operation. This determination can be done using various ....

[Article contains additional citation context not shown here]

Andrew A. Chien and William J. Dally. Experience with concurrent aggregates (CA): Implementation and programming. In Proceedings of the Fifth Distributed Memory Conference, April 1990.

....size renders programs less likely to be efficient on most parallel supercomputers because they will not vectorize well and do not expose enough parallelism to take advantage of large numbers of processors. Extending these models to capture fine grained parallelism is an area of active research [23]. 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 Time per element (usec) 8K 16K 32K 64K 128K 256K 512K 1M 2M 4M 8M Problem size CM 2 Fortran CM 2 NESL Figure 15: CM 2 median: NESL vs. CM Fortran. 7 Conclusions The purpose of nested data parallel languages is to provide the advantages of data ....

Andrew A. Chien and William J. Dally. Experience with concurrent aggregates (CA): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference, pages 1040--1049. SIAM, April 1990.

....grain size renders programs less likely to be efficient on most parallel supercomputers because they won t vectorize well and don t expose enough parallelism to take advantage of large numbers of processors. Extending this model to capture fine grained parallelism is an area of active research [19]. 7 Conclusions The purpose of nested data parallel languages is to provide the advantages of data parallelism while ex tending their applicability to algorithms that use irregular data structures. The main advantages of data parallelism that should be preserved are the efficient ....

A. A. Chien and W. J. Dally. Experience with concurrent aggregates (CA): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference. SIAM, Apr. 1990.

....communication [19] would be grossly inefficient for applications of the degree of complexity as N body simulations. This comment applies to other work on object oriented parallel programming such as ObjectOriented Interface (OOI) 20] Mentat Run time System [13] and Concurrent Aggregates (CA)[7]. Our work is also related to various run time systems for parallel programming such as PARTI[11] and Kali[14] Again, neither simple run time pointer interpretation, nor optimization such as hashed caches of PARTI, is sufficient for tackling the class of applications at hand. A high level ....

A. Chien and W. Dally. Experience with concurrent aggregates (CA): Implementation and programming. In 5th DMCC, 1990.

....to be a reusable abstraction that presents a concurrent interface. This combination supports reusable libraries of concurrent abstractions. 4. 4 Discussion Collections in ICC represent a unification of collections as distributed arrays of objects as in [29, 10] and the aggregate approach as in [18]. The array approach is more compatible with the preexisting C notion of arrays and offers the advantage of separating the collection and constituent types. This can allow distinct members to be defined upon each type. A drawback to the independence of the types is that the element members have ....

Andrew A. Chien and William J. Dally. Experience with concurrent aggregates (ca): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference, Charleston, South Carolina, April 8-12 1990. SIAM.

....a concurrent interface. This combination makes libraries of concurrent abstractions possible, a key technology for concurrent programming. 4. 4 Discussion Collections in ICC represent a unification of collections as distributed arrays of objects as in [32, 12] and the aggregate approach as in [21]. The array approach is more compatible with the preexisting C notion of arrays and offers the advantage of separating the collection and constituent types. This can allow distinct members to be defined upon each type. A drawback to the independence of the types is that the element members have ....

Andrew A. Chien and William J. Dally. Experience with concurrent aggregates (ca): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference, Charleston, South Carolina, April 8-12 1990. SIAM.

No context found.

Chien, A. A., and Dally, W. J. Experience with concurrent aggregates (CA): Implementation and programming. In Proceedings of the Fifth Distributed Memory Computers Conference (Apr. 1990), SIAM, pp. 1040--1049. 25

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