## Theoretical Analysis of the Heterogeneous Dynamic Load Balancing Problem Using a Hydro-Dynamic Approach (1996)

Venue: | Journal of Parallel and Distributed Computing archive Volume 43 , Issue |

Citations: | 2 - 0 self |

### BibTeX

@ARTICLE{Hui96theoreticalanalysis,

author = {Chi-chung Hui and Samuel T. Chanson},

title = {Theoretical Analysis of the Heterogeneous Dynamic Load Balancing Problem Using a Hydro-Dynamic Approach},

journal = {Journal of Parallel and Distributed Computing archive Volume 43 , Issue},

year = {1996},

volume = {2},

pages = {146--0743}

}

### OpenURL

### Abstract

This paper presents a hydro-dynamic framework to solving the dynamic load balancing problem on a network of heterogeneous computers. In this approach, each processor is viewed as a liquid cylinder where the cross-sectional area corresponds to the capacity of the processor, the communication links are modeled as liquid channels between the cylinders, the workload is represented as liquid, and the load balancing algorithm describes the flow of the liquid. It is proved that all algorithms under this framework converge geometrically to the state of equilibrium, in which the heights of the liquid columns are the same in all the cylinders. In this way, each processor obtains an amount of workload proportional to its capacity. The parameters that affect the convergence rate of the algorithms are also identified and discussed. 1 Introduction It is useful to explore remote computing power in local area networks (LANs) as processors get more and more powerful and the availability of high spee...

### Citations

674 |
Parallel and distributed computation: Numerical methods. Englewood Cli⁄s
- Bertsekas, Tsitsiklis
- 1989
(Show Context)
Citation Context ...s mesh, torus and n-D hypercube, optimal parameters that maximize the convergence rate have been derived [19]. The convergence of the asynchronous version of the diffusion method has also been proved =-=[20, 21]-=-. A processor in the dimension exchange method balances the workload with its neighbors one at a time. It has been proved that on a hypercube, the entire system is balanced when every processor has ex... |

348 |
Load balancing for distributed memory multiprocessors
- Cybenko
- 1989
(Show Context)
Citation Context ...orithms [8, 9, 10], whereas more recent work has concentrated on proving the convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method =-=[11]-=-, the dimension exchange method [11, 12] and the gradient based method [13, 14]. Comparisons of the different algorithms have also been reported [15, 4, 6, 16]. In the diffusion method, each processor... |

267 | A Taxonomy of Scheduling in General-Purpose Distributed Computing Systems
- Casavant, Kuhl
- 1988
(Show Context)
Citation Context ...is possible to improve the overall system throughput by balancing the workload among the workstations. Because of this potential gain, load balancing has been investigated intensively in recent years =-=[3, 4, 5, 6, 7]-=-. Load balancing techniques can be classified as either static or dynamic. Static load balancing requires complete global information on the computing system and workload characteristics. On the other... |

159 |
Strategies for dynamic load balancing on highly parallel computers
- Willebeek-LeMair, Reeves
- 1993
(Show Context)
Citation Context ...is possible to improve the overall system throughput by balancing the workload among the workstations. Because of this potential gain, load balancing has been investigated intensively in recent years =-=[3, 4, 5, 6, 7]-=-. Load balancing techniques can be classified as either static or dynamic. Static load balancing requires complete global information on the computing system and workload characteristics. On the other... |

132 |
The gradient model load balancing method
- Lin, Keller
- 1987
(Show Context)
Citation Context ...e convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimension exchange method [11, 12] and the gradient based method =-=[13, 14]-=-. Comparisons of the different algorithms have also been reported [15, 4, 6, 16]. In the diffusion method, each processor simultaneously sends workload to its neighbors with lighter workload and recei... |

44 | The generalized dimension exchange method for load balancing in k-ary n-cubes and variants
- Xu, Lau
- 1995
(Show Context)
Citation Context ...thod can also be applied to arbitrary graphs [12]. More recently, this method has been generalized and optimal parameters derived to maximize the convergence rate on n-D mesh, torus and k-ary n-cubes =-=[23, 24]-=-. The processors in the gradient based method maintain gradient maps which describe the workload variations in the system. Tasks are moved toward the processors with the steepest gradient. In the grad... |

42 |
Comparing the performance of two dynamic load distribution methods
- Kalé
- 1988
(Show Context)
Citation Context ... the major approaches include the diffusion method [11], the dimension exchange method [11, 12] and the gradient based method [13, 14]. Comparisons of the different algorithms have also been reported =-=[15, 4, 6, 16]-=-. In the diffusion method, each processor simultaneously sends workload to its neighbors with lighter workload and receives workload from its neighbors with heavier workload. Under the synchronous ass... |

42 | Analysis of the generalized dimension exchange method for dynamic load balancing
- Xu, Lau
- 1992
(Show Context)
Citation Context ...thod can also be applied to arbitrary graphs [12]. More recently, this method has been generalized and optimal parameters derived to maximize the convergence rate on n-D mesh, torus and k-ary n-cubes =-=[23, 24]-=-. The processors in the gradient based method maintain gradient maps which describe the workload variations in the system. Tasks are moved toward the processors with the steepest gradient. In the grad... |

41 | Load balancing in large networks: A comparative study
- Liiling, Monien, et al.
- 1991
(Show Context)
Citation Context ...is possible to improve the overall system throughput by balancing the workload among the workstations. Because of this potential gain, load balancing has been investigated intensively in recent years =-=[3, 4, 5, 6, 7]-=-. Load balancing techniques can be classified as either static or dynamic. Static load balancing requires complete global information on the computing system and workload characteristics. On the other... |

38 |
A stable distributed scheduling algorithm
- Bryant, Finkle
- 1981
(Show Context)
Citation Context ...t affect the convergence rate are studied in section 7. Finally, section 8 concludes the paper. 2 Related Work Early work on the nearest-neighbor approach investigated the stability of the algorithms =-=[8, 9, 10]-=-, whereas more recent work has concentrated on proving the convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimensio... |

34 |
Analysis of a graph coloring based distributed load balancing algorithm
- Hosseini, Litow, et al.
- 1990
(Show Context)
Citation Context ...ent work has concentrated on proving the convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimension exchange method =-=[11, 12]-=- and the gradient based method [13, 14]. Comparisons of the different algorithms have also been reported [15, 4, 6, 16]. In the diffusion method, each processor simultaneously sends workload to its ne... |

32 |
Stability and Distributed Scheduling Algorithms
- Stankovic
- 1985
(Show Context)
Citation Context ...t affect the convergence rate are studied in section 7. Finally, section 8 concludes the paper. 2 Related Work Early work on the nearest-neighbor approach investigated the stability of the algorithms =-=[8, 9, 10]-=-, whereas more recent work has concentrated on proving the convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimensio... |

28 |
A partially asynchronous and iterative algorithm for distributed load balancing
- Song
- 1994
(Show Context)
Citation Context ...s mesh, torus and n-D hypercube, optimal parameters that maximize the convergence rate have been derived [19]. The convergence of the asynchronous version of the diffusion method has also been proved =-=[20, 21]-=-. A processor in the dimension exchange method balances the workload with its neighbors one at a time. It has been proved that on a hypercube, the entire system is balanced when every processor has ex... |

21 | Iterative dynamic load balancing in multicomputers
- Xu, Lau
- 1994
(Show Context)
Citation Context |

19 |
A dynamic scheduling strategy for the chare-kernel system
- Shu, Kale
- 1989
(Show Context)
Citation Context ...e convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimension exchange method [11, 12] and the gradient based method =-=[13, 14]-=-. Comparisons of the different algorithms have also been reported [15, 4, 6, 16]. In the diffusion method, each processor simultaneously sends workload to its neighbors with lighter workload and recei... |

19 | Nearest-neighbor algorithms for loadbalancing in parallel computers. Concurrency: Pract
- Xu, Lau, et al.
- 1995
(Show Context)
Citation Context ... the major approaches include the diffusion method [11], the dimension exchange method [11, 12] and the gradient based method [13, 14]. Comparisons of the different algorithms have also been reported =-=[15, 4, 6, 16]-=-. In the diffusion method, each processor simultaneously sends workload to its neighbors with lighter workload and receives workload from its neighbors with heavier workload. Under the synchronous ass... |

13 | Optimal parameters for load balancing with the diffusion method in mesh networks
- Xu, Lau
- 1994
(Show Context)
Citation Context ... variance of the unbalanced workload is bounded [11, 18]. For regular network topologies such as mesh, torus and n-D hypercube, optimal parameters that maximize the convergence rate have been derived =-=[19]-=-. The convergence of the asynchronous version of the diffusion method has also been proved [20, 21]. A processor in the dimension exchange method balances the workload with its neighbors one at a time... |

12 |
Effects of response and stability on scheduling in distributed computing systems
- Casavant, Kuhl
- 1988
(Show Context)
Citation Context ...t affect the convergence rate are studied in section 7. Finally, section 8 concludes the paper. 2 Related Work Early work on the nearest-neighbor approach investigated the stability of the algorithms =-=[8, 9, 10]-=-, whereas more recent work has concentrated on proving the convergence property and the convergence rate of the algorithms. Some of the major approaches include the diffusion method [11], the dimensio... |

11 |
Load balancing on generalized hypercube and mesh multiprocessors with
- Qian, Yang
- 1991
(Show Context)
Citation Context ...d no existing workload is completed during execution of the algorithm [17, 11]. Without the quiescent assumption, it is still possible to prove that the variance of the unbalanced workload is bounded =-=[11, 18]-=-. For regular network topologies such as mesh, torus and n-D hypercube, optimal parameters that maximize the convergence rate have been derived [19]. The convergence of the asynchronous version of the... |

9 |
Load balancing and Poisson equation in a graph", Concurrency: Pract
- Boillat
- 1990
(Show Context)
Citation Context ...verge in polynomial time for any initial workload distribution given the quiescent assumption that no new workload is generated and no existing workload is completed during execution of the algorithm =-=[17, 11]-=-. Without the quiescent assumption, it is still possible to prove that the variance of the unbalanced workload is bounded [11, 18]. For regular network topologies such as mesh, torus and n-D hypercube... |

7 |
Load balancing on the hypercube and related networks
- J'aJ'a, Ryu
- 1990
(Show Context)
Citation Context ...exchanged workload with all its neighbors once [11]. The performance of the dimension exchange method on hypercube, the shuffle-exchange, the cube-connected cycles and the butterfly has been compared =-=[22]-=-. By applying the edge coloring technique to map the edges into dimensions, the dimension exchange method can also be applied to arbitrary graphs [12]. More recently, this method has been generalized ... |

2 |
The Stealth distributed schedular
- Krueger, Chawla
- 1991
(Show Context)
Citation Context ...works such as fast Ethernet, FDDI and ATM has reduced the cost of interprocessor communication. In the LAN environment, a significant portion of the workstations is left idle even during busy periods =-=[1, 2]-=-. The aggregate free CPU power in a large scale LAN may be comparable to that of a supercomputer, and it is possible to improve the overall system throughput by balancing the workload among the workst... |

2 |
A comparison of workload models of the capacity available for sharing among privately owned workstations
- Mutka
- 1991
(Show Context)
Citation Context ...works such as fast Ethernet, FDDI and ATM has reduced the cost of interprocessor communication. In the LAN environment, a significant portion of the workstations is left idle even during busy periods =-=[1, 2]-=-. The aggregate free CPU power in a large scale LAN may be comparable to that of a supercomputer, and it is possible to improve the overall system throughput by balancing the workload among the workst... |

2 |
Taxonomy for assignment in parallel processor systems
- Manoharan
- 1991
(Show Context)
Citation Context |