## Scheduling Strategies for Master-Slave Tasking on Heterogeneous Processor Grids (2002)

### Cached

### Download Links

- [ftp.ens-lyon.fr]
- [www.cs.ucsd.edu]
- [cseweb.ucsd.edu]
- [cseweb.ucsd.edu]
- DBLP

### Other Repositories/Bibliography

Citations: | 82 - 36 self |

### BibTeX

@MISC{Banino02schedulingstrategies,

author = {Cyril Banino and Olivier Beaumont and Arnaud Legrand and Yves Robert},

title = {Scheduling Strategies for Master-Slave Tasking on Heterogeneous Processor Grids},

year = {2002}

}

### Years of Citing Articles

### OpenURL

### Abstract

In this paper, we consider the problem of allocating a large number of independent, equal-sized tasks to a heterogeneous "grid" computing platform. We use a non-oriented graph to model a grid, where resources can have different speeds of computation and communication, as well as different overlap capabilities. We show how to determine the optimal steady-state scheduling strategy for each processor (the fraction of time spent computing and the fraction of time spent communicating with each neighbor). This result holds for a quite general framework, allowing for cycles and multiple paths in the interconnection graph, and allowing for several masters. Because

### Citations

11192 |
Computers and Intractability, A Guide to the Theory of NP-Completeness,” Bell Telephone Laboratories
- Garey, Johnson
- 1979
(Show Context)
Citation Context ...teady state optimization problems rather than standard makespan minimization problems. While minimizing the makespan, i.e., the total execution time, is a NP-hard problem in most practical situations =-=[9]-=-, [18], [7], it turns out that the optimal steady state can be characterized very efficiently, with low-degree polynomial complexity. The optimal steady state is defined as follows (see Section 2.2): ... |

8778 | Introduction to Algorithms - Cormen, Leiserson, et al. - 2009 |

653 | MPI – The Complete Reference - Snir - 1998 |

572 | Combinatorial Optimization - Polyhedra and Efficiency - Schrijver - 2003 |

242 | A compile-time scheduling heuristic for interconnection-constrained heterogeneous processor architectures - Sih, Lee - 1993 |

78 | M.: An enabling framework for master-worker applications on the computational grid - Goux, Kulkarni, et al. - 2000 |

77 | A Dynamic Matching and Scheduling Algorithm for Heterogeneous Computing Systems - Maheswaran, Siegel - 1998 |

74 | Bandwidth-Centric Allocation of Independent Tasks on Heterogeneous Platforms - Beaumont, Carter, et al. - 2002 |

72 | Efficient collective communication in distributed heterogeneous systems - Bhat, Raghavendra, et al. - 1999 |

55 | Task Scheduling Algorithms for Heterogenous Processors - Topcuoglu, Hariri, et al. - 1999 |

53 | Adaptive Scheduling for Master-Worker Applications on the Computational Grid - Heymann, Senar, et al. - 2000 |

45 | Gossiping in Minimal Time - Krumme, Cybenko, et al. - 1992 |

43 | Master/slave computing on the grid - Shao, Berman, et al. - 2000 |

41 |
Scheduling and Load Balancing
- Shirazi, Husson, et al.
- 1995
(Show Context)
Citation Context ... state optimization problems rather than standard makespan minimization problems. While minimizing the makespan, i.e., the total execution time, is a NP-hard problem in most practical situations [9], =-=[18]-=-, [7], it turns out that the optimal steady state can be characterized very efficiently, with low-degree polynomial complexity. The optimal steady state is defined as follows (see Section 2.2): For ea... |

38 | The master-slave paradigm with heterogeneous processors - BEAUMONT, LEGRAND, et al. |

33 | Adaptive Scheduling of Master/Worker Applications on Distributed Computational Resources - Shao - 2001 |

29 | Autonomous Protocols for Bandwidth-centric Scheduling of Independent-task Applications
- Kreaseck, Carter, et al.
- 2003
(Show Context)
Citation Context ...he node during the initialization period. The maximum number of such buffered tasks is at most T per node. The issue of limiting the number of buffered tasks to a more practical number is explored in =-=[13]-=-. 2.3.3 With Several Masters The extension for several masters is straightforward. Assume that there are k masters Pm1 , Pm2 , ..., Pmk , each holding (the initial data for) a large collection of task... |

28 | A Static Scheduling Heuristic for Heterogeneous Processors - Oh, Ha - 1996 |

23 |
Task Allocation on a Network of Processors
- Hsu, Lee, et al.
- 2000
(Show Context)
Citation Context ...see [21], among others. Unfortunately, all these heuristics assume no restriction on the communication resources, which renders them somewhat unrealistic to model reallife applications. Recent papers =-=[12]-=-, [19] suggest taking communication contention into account. Scheduling divisible load. Instead of scheduling several communications, one for each task, the divisible load approach consists in schedul... |

21 | Spanning graphs for optimum broadcasting and personalized communication in hypercubes - Johnsson, Ho - 1989 |

20 | Optimizing computing costs using divisible load analysis
- Sohn, Robertazzi, et al.
- 1998
(Show Context)
Citation Context ...ach consists in scheduling a single communication at the beginning of the operation. The cost of this communication is proportional to the amount of computation performed. A star graph is targeted in =-=[20]-=-, with homogeneous links and different-speed processors. The extension to heterogeneous links is dealt with in [6]. See also [14], [1] for more results on sharing bag of tasks in heterogeneous cluster... |

19 |
Optimizing computing costs using divisible load analysis
- Charcranoon, Luryi
(Show Context)
Citation Context ... is proportional to the amount of computation performed. A star graph is targeted in [20], with homogeneous links and different-speed processors. The extension to heterogeneous links is dealt with in =-=[6]-=-. See also [14], [1] for more results on sharing bag of tasks in heterogeneous clusters. Master-slave on the computational grid. Master-slave scheduling on the grid can be based on a network-flow appr... |

19 |
An Enabling Framework for
- Goux, Kulkarni, et al.
(Show Context)
Citation Context ...ull 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 described in =-=[10]-=-, [22]. 7 CONCLUSION In this paper, we have dealt with master-slave tasking on a heterogeneous platform. We have shown how to determine the best steady state scheduling strategy for a general intercon... |

17 | Scheduling multi-component applications in heterogeneous wide-area networks - Weissman - 2000 |

15 | A polynomial-time algorithm for allocating independent tasks on heterogeneous fork-graphs - Beaumont, Legrand, et al. |

11 |
Comparison of contention-aware list scheduling heuristics for cluster computing
- Sinnen, Sousa
- 2001
(Show Context)
Citation Context ...1], among others. Unfortunately, all these heuristics assume no restriction on the communication resources, which renders them somewhat unrealistic to model reallife applications. Recent papers [12], =-=[19]-=- suggest taking communication contention into account. Scheduling divisible load. Instead of scheduling several communications, one for each task, the divisible load approach consists in scheduling a ... |

10 | On multicast algorithms for heterogeneous networks of workstations - Libeskind-Hadas, Hartline, et al. |

9 | Scheduling heuristics for heterogeneous processors - Boudet, Robert - 2001 |

9 | Exploiting unused time-slots in list scheduling considering communication contention - Sinnen, Sousa - 2001 |

7 |
Adaptive Scheduling for
- Heymann, Senar, et al.
(Show Context)
Citation Context ... on sharing bag of tasks in heterogeneous clusters. Master-slave on the computational grid. Master-slave scheduling on the grid can be based on a network-flow approach [17] or on an adaptive strategy =-=[11]-=-. Note that the network-flow approach of [17] is possible only when using a full multiple-port model, where the number of simultaneous communications for a given node is not bounded. Enabling framewor... |

4 |
Scheduling Strategies for Master-Slave Tasking
- Banino, Beaumont, et al.
- 2004
(Show Context)
Citation Context ...an spanning trees, and that determining the best spanning tree is NP-hard. Nevertheless, several low-cost heuristics that achieve very good performances on a wide range of simulations are proposed in =-=[2]-=-. These positive experiments show that, in practice, it is safe to rely on spanning trees to implement master-slave tasking. ut This work can be extended in the following two directions: . On the theo... |

4 |
Optimal algorithms for the pipelined scheduling of task graphs on heterogeneous systems
- Beaumont, Legrand, et al.
- 2003
(Show Context)
Citation Context ...0] provides in time OðjEj 2 Þ a polynomial number of matchings, which are used to perform the different communications, and which provides the desired polynomial-size description of the schedule. Se=-=e [5]-=- for further details. We point out that reconstructing the final schedule is much easier for tree-shaped platforms, as shown in Section 3. Notice that it is necessary for each node to be able to hold ... |

4 | Scheduling algorithms for ecient gather operations in distributed heterogeneous systems - Hatta, Shibusawa - 2000 |

4 | Scheduling problems in a practial allocation model - Hollermann, Hsu, et al. - 1997 |

3 |
Optimal Sharing of Bags of Tasks
- Adler, Gong, et al.
- 2003
(Show Context)
Citation Context ...the amount of computation performed. A star graph is targeted in [20], with homogeneous links and different-speed processors. The extension to heterogeneous links is dealt with in [6]. See also [14], =-=[1]-=- for more results on sharing bag of tasks in heterogeneous clusters. Master-slave on the computational grid. Master-slave scheduling on the grid can be based on a network-flow approach [17] or on an a... |

3 |
On sharing bags of tasks in heterogeneous networks of workstations: greedier is not better
- Rosenberg
(Show Context)
Citation Context ...al to the amount of computation performed. A star graph is targeted in [20], with homogeneous links and different-speed processors. The extension to heterogeneous links is dealt with in [6]. See also =-=[14]-=-, [1] for more results on sharing bag of tasks in heterogeneous clusters. Master-slave on the computational grid. Master-slave scheduling on the grid can be based on a network-flow approach [17] or on... |

3 | A realistic model and an ecient heuristic for scheduling with heterogeneous processors - Beaumont, Boudet, et al. - 2002 |

2 |
Task Scheduling Algorithms for Heterogeneous
- Topcuoglu, Hariri, et al.
- 1999
(Show Context)
Citation Context ...lated papers along the following lines: Scheduling task graphs on heterogeneous platforms. Several heuristics have been introduced to schedule (acyclic) task graphs on different-speed processors, see =-=[21]-=-, among others. Unfortunately, all these heuristics assume no restriction on the communication resources, which renders them somewhat unrealistic to model reallife applications. Recent papers [12], [1... |

2 | A world wide number eld sieve factoring record: on to 512 bits - Cowie, Dodson, et al. - 1996 |

1 |
Cyril Banino is currently a PhD student in NTNU, the Norvegian University of Science and Technology. He is mainly interested in highperformance scientific computing and in combinatorial optimization. Olivier Beaumont received the PhD degree from the Unive
- Weissman
- 2000
(Show Context)
Citation Context ...ltiple-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 described in [10], =-=[22]-=-. 7 CONCLUSION In this paper, we have dealt with master-slave tasking on a heterogeneous platform. We have shown how to determine the best steady state scheduling strategy for a general interconnectio... |