Garry Shao, "Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources", Ph.D. Thesis, University of California, San Diego. May 2001.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... deal with the complexity of collective communications on heterogeneous platforms: broadcast and multicast operations are addressed in [4, 15] gather operations are studied in [11] Master slave on the computational grid Master slave scheduling on the grid can be based on a network ow approach [19, 18] or on an adaptive strategy [12] Enabling frameworks to facilitate the implementation of master slave tasking are described in [10, 22] 7 Conclusion In this paper, we have shown that master slave tasking using heterogeneous fork graphs can be solved in polynomial time, using a non trivial ....

G. Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, Dept. of Computer Science, University Of California at San Diego, 2001.

....can be introduced in a space but there they are not active objects. The JavaSpace interface does not have any operation corresponding to Linda eval primitive. 3. A Framework for Adaptive Master Worker Parallelism Master worker parallelism is a widely used form of parallel application programming [13, 2]. It is conceptually very simple and involves dividing a problem into a smaller number of independent work units which can be distributed to remote worker processes for computation in parallel. A single master process centrally controls both the distribution of work units to worker processes and ....

....the distribution of work units to worker processes and the returned of computed results back to the master process. The method of maintaining a collection of work units is referred as work queue or task farm scheduling. There are many opportunities for running distributed running applications [13]. We choose here the Jini [15] environment for the master and worker processes, and JavaSpaces for task scheduling. A typical space based compute server works as described below: A task is an entry that both describes the speci c of the task and contains methods that performs the necessary ....

G. Shao, Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources, Phd Thesis, Univ. California, San Diego, 2001.

....application. Each AppLeS template is developed to host a structurally similar class of applications. To date, we have developed two AppLeS templates APST (AppLeS Parameter Sweep Template) 14] which targets parameter sweep applications, and AMWAT (AppLeS Master Worker Application Template) [33] which targets master worker applications. In addition, we have developed a generic supercomputer AppLeS template (SA) 12] 34] for scheduling applications over space shared parallel supercomputers. AppLeS templates and generic supercomputer AppLeS constitute application execution environments ....

....memory, PVM [7] and MPI [8] Support for delivering consistent application performance is provided by both general and specialized MW scheduling functions contained in the Scheduling module of Figure 9. Additional details for each of the AMWAT modules as well as on related work can be found in [33]. Unique to the AMWAT scheduling approach is the attention given to addressing performance bottlenecks caused by the run time interaction of application demands and deliverable resource capabilities. 22 As part of the work in creating AMWAT, we developed a work ow model of MW application ....

[Article contains additional citation context not shown here]

G. Shao, Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources, Ph.D. thesis, University of California, San Diego, May 2001.

....server one at a time on the homogeneous parallel machine. NP completeness results are given in [18, 24, 9] and guaranteed approximations in [27] In constrast, our work consider the allocation of homogenous tasks on a heterogeneous system. The work most closely related to ours is Shao s thesis [33]. He suggests using the Maximum Flow algorithm on the graph (not necessarily a tree) representing heterogeneous resources to maximize steady state throughput of Master Worker applications. In contrast, our work obtains a simple, closed form formula for maximizing steady state, which in turn gives ....

....by the Network Weather Service [35] A nice feature of our approach is that each node makes its decisions autonomously. The explicit solution we derived, which maximizes the throughput of the entire grid, does not require solving a global max ow problem (as suggested, for instance, by Shao [33]. By ignoring startup costs and focusing on steady state throughput, we were able to obtain a simple, explicit solutions to our optimization problem. However, it would be interesting to invesitgate the cost of startup, through theory, simulations, and actual experiments. That said, we note that ....

Gary Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, University of California at San Diego, May 2001.

No context found.

G. Shao, "Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources," Ph.D. thesis, Univ. of California, San Diego, May 2001.

No context found.

G. Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, Dept. of Computer Science, University Of California at San Diego, 2001.

No context found.

G. Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, Dept. of Computer Science, University Of California at San Diego, 2001.

.... master slave scheduling problem is motivated by problems that are addressed by collaborative computing e orts such as SETI home [23] factoring large numbers [9] the Mersenne prime search [22] and those distributed computing problems organized by companies such as Entropia [10] Several papers [25, 24, 14, 12, 31, 3, 2] have recently revisited the master slave paradigm for processor clusters or grids, and we refer to Section 7 for comparison and discussion. This paper is a follow on of recent work by Beaumont et al. 2] who solve the master slave scheduling problem for a tree shaped heterogeneous platform. ....

....heuristics : LP Tree, Naive MST, Compute Tree, C to C Tree and BW centric Tree. broadcast and multicast operations are addressed in [5, 19] gather operations are studied in [13] Master slave on the computational grid Master slave scheduling on the grid can be based on a network ow approach [25, 24] or on an adaptive strategy [14] Note that the network ow approach of [25, 24] is possible only when using a full multiple port model, where the number of simultaneous communications for a given node is not bounded. Enabling frameworks to facilitate the implementation of master slave tasking are ....

[Article contains additional citation context not shown here]

G. Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, Dept. of Computer Science, University Of California at San Diego, 2001.

....factor. Efficiently managing the resulting computation is a difficult and challenging problem, given the heterogeneous attributes of the underlying components. This problem is well recognized [3, 18, 21, 16, 13, 26, 36, 23, 7, 2] and there is a large body of applied research, see for instance [5, 10, 1, 17, 33], providing various practical approaches toward a solution. An added complexity is that resources in these environments exhibit dynamic performance characteristics and availability, and a number of the aforementioned works have taken first steps toward addressing this issue. Our previous work [3] ....

....and our work provides a fundamental first step. Our work is related to a number of endeavors that provide usable software for deploying applications on largescale platforms [10, 1, 32, 12, 25] At the moment, most of these projects employ some flavor of centralized scheduling. The works in [33, 17, 20] allow for the well known hierarchical master worker scheme, which is strongly related to our work. However, hierarchical master worker has rarely been deployed in practice due to the lack of a decentralized, autonomous scheduling methodology. Our work provides such a methodology. 6. Conclusion ....

G. Shao. Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources. PhD thesis, UCSD, June 2001.

....a problem can be divided into a number of smaller independent work units that a master process distributes to many worker processes. Master worker computing has been successfully applied to many application domains (e.g. see [6,25,31] and has been the object of a number of research endeavors (see [29] for a comprehensive survey) The TOP C project [11] provides a simple master worker model for applications implemented in Single Program, Multiple Data (SPMD) fashion. Like MPI, it provides an API that can be used to write applications. Since the master worker model is more abstract than a pure ....

....of distributed computing applications. The goal of AMPIC is not to provide yet another set of mechanisms, but rather to provide a common and easy to use interface to existing mechanisms. AMPIC was originally developed as part of the AppLeS Master Worker Application Template (AMWAT) project [5,29] which is related to TOP C in the following way. Like TOP C, AMWAT targets master worker applications. However, AMWAT s focus in on scheduling whereas TOP C s focus is on providing a exible and powerful programming model (see Section 2) In fact, the scheduling techniques developed as part of ....

[Article contains additional citation context not shown here]

G. Shao. Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources. PhD thesis, University of California, San Diego, May 2001.

No context found.

Garry Shao, "Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources", Ph.D. Thesis, University of California, San Diego. May 2001.

No context found.

G. Shao, Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources, Ph.D. thesis, University of California, San Diego, May 2001.

No context found.

Shao, G.: Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources. PhD thesis, University of California - San Diego (2001)

No context found.

G. Shao. Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources. PhD thesis, University of California at San Diego, May 2001.

No context found.

G. Shao. Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources. PhD thesis, University of California, San Diego, 2001.

No context found.

Gary Shao. Adaptive scheduling of master/worker applications on distributed computational resources. PhD thesis, University of California at San Diego, May 2001.

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