Load balancing in dynamic networks

Lecture Notes in Computer Science, 2002

This paper proposes the Average Diffusion (ADF) method for solving the load balancing problem. It is shown that a sufficient and necessary condition for the ADF method to converge to the uniform distribution of loads is the induced network of processors to be d-regular, connected and not bipartite. Next, we proceed and apply Fourier analysis determining the convergence factor γ in terms of the diffusion parameters cij (weighted case) when the network of processors is a ring and 2D-torus. It is shown that cij = 1 2 and cij ∈ (0, 1 2 ) when the network is a ring and 2D-torus, respectively, thus solving partially the open problem which concerns the determination of the diffusion parameters cij.

2007

In this paper, first order diffusion load balancing algorithms for totally dynamic networks are investigated. Totally dynamic networks are networks in which the topology may change dynamically. Some edges or nodes can appear, disappear or move during the time. In our previous works on dynamic networks, the dynamism was limited to the edges. The main result of this study consists in proving that the load balancing algorithms reduce the unbalance on arbitrary dynamic networks. Notice that the hypotheses of our result are realistic and that for example the network does not have to be maintained connected. To study the behavior of these algorithms, we compare the load evolution by several simulations. load balancing, totally dynamic networks, iterative algorithm.

International Journal of Information Technology and Computer Science, 2014

In distributed computing system some nodes are very fast and some are slow and during the computation many fast nodes become idle or under loaded while the slow nodes become over loaded due to the uneven distribution of load in the system. In distributed system, the most common important factor is the information collection about loads on different nodes. The success of load balancing algorithm depends on how quickly the information about the load in the system is collected by a node willing to transfer or accept load. In this paper we have shown that the number of communication overheads depends on the number of overloaded nodes present in the domain of an under loaded nodes and vice-versa. We have also shown that communication overhead for load balancing is always fairly less than KN but in worst case our algorithm's complexity becomes equal to KN.

Proceedings of the sixth annual ACM symposium on Parallel algorithms and architectures - SPAA '94, 1994

Diusion is a well-known algorithm for load-balancing in which tasks move from heavily-loaded processors to lightly-loaded neighbors. This paper presents a rigorous analysis of the performance of the diusion algorithm on arbitrary networks.

Euro-Par 2010-Parallel Processing, 2010

Lecture Notes in Computer Science, 2006

In this paper, a practical approach of diffusion load balancing algorithms and its implementation are studied. Three problems are investigated. The first one is the determination of the load balancing parameters without any global knowledge. The second problem consists in estimating the cost and the benefit of a load exchange. The last one studies the convergence detection of the load balancing algorithm. For this last point we give an algorithm based on simulated annealing to reduce the convergence towards a load repartition in steps that can be done with discrete loads. Several simulations close this paper and illustrate the impact of the various methods and algorithms introduced.

Lecture Notes in Computer Science, 2004

The task of balancing dynamically generated work load occurs in a wide range of parallel and distributed applications. Diffusion based schemes, which belong to the class of nearest neighbor load balancing algorithms, are a popular way to address this problem. Originally created to equalize the amount of arbitrarily divisible load among the nodes of a static and homogeneous network, they have been generalized to heterogeneous topologies. Additionally, some simple diffusion algorithms have been adapted to work in dynamic networks as well. However, if the load is not divisible arbitrarily but consists of indivisible unit size tokens, diffusion schemes are not able to balance the load properly. In this paper we consider the problem of balancing indivisible unit size tokens on dynamic and heterogeneous systems. By modifying a randomized strategy invented for homogeneous systems, we can achieve an asymptotically minimal expected overload in l 1 , l 2 and l ∞ norm while only slightly increasing the run-time by a logarithmic factor. Our experiments show that this additional factor is usually not required in applications.

Lecture Notes in Computer Science, 2006

In this paper, we consider the application of accelerated methods in order to increase the rate of convergence of the diffusive iterative load balancing algorithms. In particular, we compare the application of Semi-Iterative, Second Degree and Variable Extrapolation techniques on the basic Diffusion method and the Extrapolated Diffusion method for torus graphs. It is shown that our methods require approximately 30% less iterations to reach the balanced state compared to the existed ones.

The Journal of Supercomputing, 2009

Load balancing a distributed/parallel system consists in allocating work (load) to its processors so that they have to process approximately the same amount of work or amounts in relation with their computation power. In this paper, we present a new distributed algorithm that implements the Most to Least Loaded (M2LL) policy. This policy aims at indicating pairs of processors, that will exchange loads, taking into account actually broken edges as well as the current load distribution in the system. The M2LL policy fixes the pairs of neighboring processors by selecting in priority the most loaded and the least loaded processor of each neighborhood. Our first and main result is that the M2LL distributed implementation terminates after at most (n/2)⋅d t iterations where n and d t are respectively the number of nodes and the degree of the system at time t. We then present a performance comparison between Generalized Adaptive Exchange (GAE) that uses M2LL and Relaxed First Order Scheme (RFOS), two load balancing algorithms for dynamic networks in which only link failures are considered. The comparison is carried out on a dedicated test bed that we have designed and implemented to this end. Our second important result is that although generating more communications, the GAE algorithm with the M2LL policy is faster than RFOS in balancing the system load. In addition, GAE M2LL is able to achieve a more stable balanced state than RFOS and scales well.

… Engi-neering Department, Arab Academy for …, 2001

The escalating complexity and mobility of today's networks has led to the increased application of mobile agent paradigm. This paradigm helps to alleviate bandwidth limitations and supports disconnected operations that are both significant problems in wireless and mobile environments. On the other hand, load balancing is one of the important problems of computer heterogeneous networks. To address this problem, many centralized approaches have been proposed in the literature but centralization has proved to raise scalability tribulations.

Lecture Notes in Computer Science, 2001

In this paper, we present a time dependent diffusion model for load balancing on synchronous networks with dynamically changing topology. This situation can arise for example from link failures. Our model is general and include Cybenko's diffusion models for fixed topology networks. We will show that under some assumptions, the time dependent load balancing algorithm converges and balances the total work load among the processors in the distributed network.

2007 46th IEEE Conference on Decision and Control, 2007

We study the distributed and decentralized load balancing problem on arbitrary connected graphs, representing an homogeneous network. The network contains several tasks, represented by possibly different integer numbers, to be processed at nodes. We propose a randomized algorithm based on gossip that achieves consensus on the load distribution within fixed bounds of the optimal one; we also show by simulations that in most cases the achieved consensus is optimal. We finally present a computationally convenient heuristic and show that it ensures the same bounds: simulation results, however, show that the heuristic performs worse.

The grid and cluster computing uses interconnected nodes to solve a problem in parallel in order to improve the response time of the system. Diffusive load balancing algorithms works well when the nodes in the system have the same processing capacity. But little attention was paid in diffusion load balancing algorithms in the literature for distributing the workload in the nodes with different processing capabilities when the load of the nodes is treated as integers. When the loads are distributed to the nodes without considering their processing capacities it would affect the response time of the system. Effective load balancing in heterogeneity can be achieved by considering the processing capacities of the nodes. This paper propose a diffusive load balancing algorithm which distributes a proportion of excessive workload of heavily loaded node to lightly loaded node by considering the nodes processing capabilities.

Load balancing is an important prerequisite to efficiently execute dynamic computations on parallel computers. In this context, this project has focussed on two topics: balancing dynamically generated work load cost efficiently in a network and partitioning graphs to equally distribute connected tasks on the processing nodes while reducing the communication overhead. We summarize new insights and results in these areas.

2009 American Control Conference, 2009

In this paper we consider the problem of load balancing over heterogeneous networks, i.e. networks whose nodes have different speeds. We assume that tasks are indivisible and with different weights. Our goal is that of minimizing the maximum execution time over nodes.