Amotz Bar-Noy and Shlomo Kipnis. Designing broadcasting algorithms in the postal model for message-passing systems. Mathematical Systems Theory, 27(5):431--452, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....reflect the congestion capabilities of real machines: it is either too strong (concurrent read write) or too weak (exclusive read write) most commercial and research machines are well approximated with the queueing model, however. Other related models are the BSP Model [Val90] the Postal Model [BK92] the Atomic Message Passing Model [LAB93] A completely different model for congestion that has been very popular in the past few years is the shared memory contention model [DHW93, DHW97] Unlike LogP, QRQW PRAM, and our model, among others, it does not need a notion of time but restricts ....

Amotz Bar-Noy and Shlomo Kipnis. Designing broadcasting algorithms in the postal model for message-passing systems. In Proceedings of the 4th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 13--22, San Diego, California, June 29--July 1, 1992.

....No ERBFMBICT972076 designed having in mind factors not present in any reasonable machine, such as zero communication delay or in nite bandwidth. The recent introduction of realistic parallel computer models such as the Bulk Synchronous Parallel (BSP) model of computation [21] the Postal Model [1], the Tau Model [19] and LogP [2] has addressed some of these limitations of parallel computing. The Bulk Synchronous Parallel (BSP) model of computation has been proposed as a uni ed framework for the design, analysis and programming of general purpose parallel computing systems. It allows the ....

Amotz Bar-Noy and Shlomo Kipnis. Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems. Proceedings of the ACM Symposium on Parallel Algorithms and Architectures, pages 13-22, June 1992.

....optimal parallel applications it is essential to have a model that accurately reflects the characteristics of these basic operations. In the literature several point to point communication models have been described that match the MPI abstractions up to a certain degree (e.g. the Postal Model [4], LogP [6] and LogGP [1] Although successful in many situations, these models do not incorporate all capabilities of MPI s send and receive operations. Most importantly, the effect of memory layout on communication costs is ignored completely. This is unfortunate, as Prieto et al. 9] have ....

A. Bar-Noy and S. Kipnis. Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems. Mathematical Systems Theory, 27(5):431--452, 1994.

....both because it is difficult to describe a topology by a few parameters and because congestion is usually the most important, but hardest to model, factor. In the last few years, several interesting models have been proposed in the literature. The Postal model was introduced by Bar Noy and Kipnis [2]. In this model, the system consists of N nodes, processors with local memory. Each node can simultaneously use a single input port and a single output port and the communication latency is . Namely, each node during each communication step can send a fixed size packet and at the same time receive ....

A. Bar-Noy and S. Kipnis, "Designing Broadcasting Algorithms in the Postal Model for MessagePassing Systems," in Proceedings of the ACM Symposium on Parallel Algorithms and Architectures, pp. 13--22, June 1992.

....and chaotic world of hardware [160] They can be logically placed in between abstract models, as the PRAM models, and more detailed network models that reflect the behavior of single family of interconnection networks. 4.3. 1 The Postal Model The Postal model was introduced by Bar Noy and Kipnis [9]. In this model, the system consists of N nodes, processors with local memory. Each node can simultaneously use a single input port and a single output port and the communication CHAPTER 4. ABSTRACT MACHINE MODELS 57 latency is . Namely, each node during each communication step can send a ....

Amotz Bar-Noy and Shlomo Kipnis. Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems. In Proceedings of the ACM Symposium on Parallel Algorithms and Architectures, pages 13--22, June 1992. 133 BIBLIOGRAPHY 134

....cluster computing, e.g. regular and irregular problems, a simple, rigid model could not serve our needs. 8 In general, existing parallel models that focus on message passing architecture, which include abstract architecture models (e.g. LogP [8] BSP [21] and communication models (e.g. Postal [1]) usually assume reliable network, such that they treat sending a message as a send and forget [1] event. They also assume fully connected network with the exact architecture of the underlying communication network ignored. Communication is based on point to point semantics, with the latency ....

....needs. 8 In general, existing parallel models that focus on message passing architecture, which include abstract architecture models (e.g. LogP [8] BSP [21] and communication models (e.g. Postal [1] usually assume reliable network, such that they treat sending a message as a send and forget [1] event. They also assume fully connected network with the exact architecture of the underlying communication network ignored. Communication is based on point to point semantics, with the latency between any pair of processors roughly the same time for all cases. These models provide an abstract ....

A. Bar-Noy and S. Kipnis, "Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems", in Proceedings of the ACM Symposium on Parallel Algorithms and Architectures, pp. 11-22, June 1992.

....broadcast single packets, we can: merge the message setup ( and packaging ( costs into a single preparation cost x ; ignore the marginal bandwidth limitation delay (1= of a message and focus only on the latency . b) When all x s are equal, our model coincides with the postal model [7]. Basic entities. An instance of the WBP has the form: I = n; s; f (x)g x2f1; ng ; f x g x2f1; ng ; The last time unit during which a node x participates in the broadcast (as a caller or a callee ) under broadcast algorithm B is x s idle time (x) under B. Clearly, s) s ....

....GBA solves the WBP optimally. The proof provides a series of cut and paste operations that convert any broadcast algorithm to the GBA, with no slowdown. This result, which encompasses all prior results about homogeneous clusters, shows that the GBA solves the WBP optimally within the postal model [7]. 3.4. Extending the GBA to Hyperclusters. Our efficient hypercluster broadcast algorithm selects leader nodes at each level of a hypercluster s hierarchy, which will participate in that level s WBP. The selection is via each node s level k inherent weight, which reflects the size of the level k ....

A. Bar-Noy and S. Kipnis (1994): Designing broadcasting algorithms in the postal model for message-passing systems. Math. Syst. Th. 27, 431--452.

....ways in which the trees evolve as the number of leaves increase. The analysis of L(x) is of independent interest. We illustrate this by describing an application in the derivation of exact bounds on the time needed for broadcasting in the Postal Model of computation, improving the bounds given in [2]. A more important application arises when we combine the analysis of L(x) with the first result describing the structure of trees. This will yield the major result of this paper, an exact analysis of the cost of optimal trees or Varn codes. Because we know exactly how the structure of the optimal ....

.... Gamma ffl) L(x) for all x X: This in turn implies that if n is large enough we can find x such that L(x Gamma ffl) n L(x) and therefore that x Gamma ffl H(n) x: Inverting L(x) 1 Gamma Gammac 1 c ln x o( x ) yields part 2 of the theorem. 2 3.3. 3 Applications In [2] Bar Noy and Kipnis introduce the Postal Model of message passing for distributed systems. In this model, counting time from when a sender first starts sending a message, the sender requires one unit of time before completing the work of sending and being able to do something else but the ....

[Article contains additional citation context not shown here]

A. Bar-Noy and S. Kipnis, "Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems," Mathematical Systems Theory, 27(5) (1994), pp.431--452.

....message transmission from a processor. Alexandrov and others have proposed the LogGP model [8] which incorporates long messages into the LogP model. The Gap per byte for long messages, G, is defined as the time per byte for a long message. Bar Noy and Kipnis have developed the postal model [16], a special case of LogP model, where g is one. However, they don t consider the parameters o and G. A node in LogP, LogGP, and postal models can send another message immediately g time after the previous message has been sent without waiting for the previous message to be delivered at the ....

....recurrence relations: 5.1) It can also be shown that the number of nodes, r(S) that receive the message at each step, S, follows the recurrence relations: 5. 2) These recurrence relations are a kind of generalization of the Fibonacci functions defined by Bar Noy and Kipnis for the postal model [16], and are similar to the recurrence relations of the broadcasting algorithms by Berthome and Ferreira [20] The above relations and those in [16, 20] cannot be solved for a general d. They should be computed step 0 1 2 3 4 5 6 0 7 8 9 10 11 12 13 14 15 16 1 2 17 24 25 32 33 34 35 36 37 38 39 ....

[Article contains additional citation context not shown here]

A. Bar-Noy and S. Kipnis, "Designing Broadcasting Algorithms in the Postal Model for Message-Passing Systems", 4th Annual ACM Symposium on Parallel Algorithms and Architectures, June 1992, pp. 11-22.

....size s in the LogP model to perform other computations otherwise. The sender will be able to send the next message at the time t o sg, while the receiver will be ready to receive the next message at the time t o L sg. In the particular case when g = 1; o = 0 the LogP is called the postal model [6]. Moreover, every LogP model with o = 0 is equivalent to a postal model, because one can normalize g to 1. Data Partitioning Each BVRAM register will be, in general, much longer than P . The process of splitting a large collection on several nodes is called data partitioning [27, 26, 73] or ....

Amotz Bar-Noy and Shlomo Kipnis. Designing broadcasting algorithms in the postal model for message-passing systems. In Proceedings of the 4th Annual ACM Symposium on Parallel Algorithms and Architectures, pages 13--22, 1992.

No context found.

Amotz Bar-Noy and Shlomo Kipnis. Designing broadcasting algorithms in the postal model for message-passing systems. Mathematical Systems Theory, 27(5):431--452, 1994.

No context found.

Amotz Bar-Noy and Shlomo Kipnis, "Designing broadcasting algorithms in the postal model for message passing systems", in Proc. of SPAA, 1992, pp. 13--22.

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