Cypher, R., and Konstantinidou, S. Bounds on the efficiency of message-passing protocols for parallel computers. SIAM J. Comput. 25, 5 (1996), 1082--1104.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....work on efficient data transfers with limited memory. Some standard collective communication operations can be implemented in ways that limit memory usage, but the general problem we are proposing seems to be new. Cypher and Konstantinidou designed memory efficient message passing protocols [3]. However, their work addressed exchange of tokens as opposed to variable sized messages. And they didn t explicitly consider the effect of finite memory in the processors. Their work conceptually divides a process into communication and application processes. Communication processes receive ....

Cypher, R., and Konstantinidou, S. Bounds on the efficiency of message-passing protocols for parallel computers. SIAM J. Comput. 25, 5 (1996), 1082--1104.

....in what aspects of parallel machines are exposed. Some focus on dealing with asynchrony in a shared memory context (e.g. 8, 20, 21, 28, 32, 35, 49, 57, 61] Others focus on accounting for the overheads in accessing the shared memory ( 2, 3, 25, 32, 41, 44, 52, 56] or in sending messages ([5, 9, 10, 22, 23, 39, 53, 55, 69]) Several models are primarily concerned with the memory hierarchy, especially disk I O ( 6, 62, 72] Others focus on contention at the memory location ( 28, 36] or memory module ( 60, 7, 27] Finally, a few models incorporate powerful aggregate communication primitives ( 14, 18] Given this ....

R. Cypher and S. Konstantinidou. Bounds on the efficiency of message-passing protocols for parallel computers. In Proc. 5th ACM Symp. on Parallel Algorithms and Architectures, pages 173--181, June-July 1993.

....differ in what aspects of parallel machines are exposed. Some focus on dealing with asynchrony in a shared memory context (e.g. 6, 15, 16, 21, 25, 39, 47, 49] Others focus on accounting for the overheads in accessing the shared memory ( 1, 2, 20, 25, 34, 36, 43, 46] or in sending messages ([8, 9, 17, 18, 32, 44, 45, 55]) Several models are primarily concerned with the memory hierarchy, especially disk I O ( 4, 50, 57] Others focus on contention at the memory location ( 21, 30] or memory module ( 5] Finally, a few models incorporate powerful aggregate communication primitives ( 10, 13] In [26] we ....

....Standard PRAM models and network based models are described above. Localityminded PRAM models (cf. 1, 2, 20, 25, 34, 43, 46, 48] augment the PRAM model with parameters that reward locality, and may explicitly consider the partitioning of memory among the processors. Abstract network models (cf. [5, 8, 9, 17, 18, 32, 36, 44, 45, 55]) consist of a collection of processors that communicate by sending point to point messages. The underlying communication network is modeled using a series of parameters reflecting, e.g. the bandwidth and latency of the network. Many of these models either explicitly or implicitly support a ....

R. Cypher and S. Konstantinidou. Bounds on the efficiency of message-passing protocols for parallel computers. In Proc. 5th ACM Symp. on Parallel Algorithms and Architectures, pages 173--181, June-July 1993.

....differ in what aspects of parallel machines are exposed. Some focus on dealing with asynchrony in a shared memory context (e.g. 8, 19, 20, 26, 30, 33, 47, 54, 57] Others focus on accounting for the overheads in accessing the shared memory ( 2, 3, 24, 30, 39, 42, 50, 53] or in sending messages ([5, 9, 10, 21, 22, 37, 51, 52, 64]) Several models are primarily concerned with the memory hierarchy, especially disk I O ( 6, 58, 67] Others focus on contention at the memory location ( 26, 34] or memory module ( 7] Finally, a few models incorporate powerful aggregate communication primitives ( 14, 17] Given this plethora ....

R. Cypher and S. Konstantinidou. Bounds on the efficiency of message-passing protocols for parallel computers. In Proc. 5th ACM Symp. on Parallel Algorithms and Architectures, pages 173--181, June-July 1993.

....arrive before there corresponding receive has been posted) possibly organized according to tag and or source, so that posted receives can be matched efficiently to these queued unanticipated messages. Flow control is necessary so that the unanticipated messages do not overflow the system buffers [9]. Furthermore, special protocols may have to be implemented to handle long messages and support for complex communication patterns (such as collective communication) may be part of the communication coprocessor as well. In the shared memory paradigm, the user processes have the illusion of a ....

R. Cypher and S. Konstantinidou, "Bounds on the efficiency of message-passing protocols for parallel computers", in Proc. Fifth ACM Symp. on Parallel Algorithms and Architectures, pp. 173--181, 1993.

....arrive before their corresponding receive has been posted) possibly organized according to tag and or source, so that posted receives can be matched efficiently to these queues unanticipated messages. Flow control is necessary so that the unanticipated messages do not overflow the system buffers [9]. Furthermore, special protocols may have to be implemented to handle long messages and support for complex communication patterns (such as collective communication) may be part of the communication coprocessor as well. In the shared memory paradigm, the user processes have the illusion of a ....

R. Cypher and S. Konstantinidou, "Bounds on the efficiency of message-passing protocols for parallel computers", in Proc. Fifth ACM Symp. on Parallel Algorithms and Architectures, pp. 173--181, 1993.

....a result, it is essential to understand exactly how these properties interact with one another and with the message ordering properties. Unfortunately, the existing models of message passing systems are imprecise [2, 5, 6, 7, 13] or they model a very restricted set of send and receive primitives [1, 3, 8, 9, 10, 11, 14], or they apply only to correct programs [3, 5, 11] While it might seem sufficient to consider only correct programs, there are actually two problems with this approach. First, large and complex message passing programs are likely to contain errors during their development. As a result, the ....

....interact with one another and with the message ordering properties. Unfortunately, the existing models of message passing systems are imprecise [2, 5, 6, 7, 13] or they model a very restricted set of send and receive primitives [1, 3, 8, 9, 10, 11, 14] or they apply only to correct programs [3, 5, 11]. While it might seem sufficient to consider only correct programs, there are actually two problems with this approach. First, large and complex message passing programs are likely to contain errors during their development. As a result, the messagepassing primitives must have well defined ....

R. Cypher, M. Konstantinidou, Bounds on the Efficiency of Message-Passing Protocols for Parallel Computers, Proc. of ACM Symp. on Parallel Algorithms and Architectures, pp. 173-181, 1993.

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