A Practical Approach of Diffusion Load Balancing Algorithms

Simulated annealing-based diffusive load balancing on many-core SoC

“…In the result of DN, threads are well distributed in the sense that for every pair of two neighboring cores the load difference is only one. However, we can easily observe the global imbalance; a core at (2,1) has 3 threads while cores at (0,4) and (3,4) have no thread. On the other hand, SA-DLB can distribute the thread perfectly, which is shown in Figure 1(c).…”

“…We accordingly borrow only the concept of simulated annealing, which includes cooling schedule and probability of accepting moves, and so on. The approach in [3] also adopts the concept of simulated annealing for load balancing. However, it does not have the concept of cooling schedule, which will make it difficult to converge to an optimal state.…”

Process reassignment with reduced migration cost in grid load rebalancing

“…Each local search step (the "while" loop in lines 2-17) is refined into multiple sub-steps (the "for" loop in lines [5][6][7][8][9][10][11][12][13][14]. The different restrictions on acceptable migration cost of a process movement (thresholds on acceptable migration cost) are defined in the sub-steps.…”

“…In other dynamic data remapping works, the data and the associated workload can be divided into arbitrary small pieces and distributed among machines [17] [19] [20] [24] [14]. Since workload can be arbitrarily divided, it is easy to know how much workload should be migrated out/in the over-loaded/under-loaded machines.…”

Load Balancing: An Approach Based on Clustering in Ad Hoc Networks

“…Load Balancing algorithms are designed essentially to distribute equally the load on nodes and maximize their utilization while minimizing the total task execution time. This issue has been of considerable interest in the network research community when it comes to wired [29,30,2,26,3,17] and wireless [24,13,22,23,20] networks. It aims to guarantee that no node is underloaded or overloaded.…”