A bri trac tThis paper presents the bsudgeted-weighted-round-robin (BWRR) algorithm for scheduling the transmission of hardreal-time messages in a packet-switched network. The BWRR algorithm provides LI bound on the worst case delay of a message through the network. In addition, with the BWRR algorithm, the necesaaiy buffer space at each switch is predictable and can be f i e d at system Configuration time. The per switch delay and required buffer space provided by the BWRR algorithm are signi$cantly less than those provided by previous methods. The BWRR algorithm has a simple hardware implementation and admission test, and it does not require a global clock or any explicitJlow control messages. It can achieve high utilization of the network under most practical scenari0.r and reacts predictably under transient overloads.
Load balancing in a local area network attempts to improve system performance by finding the lightly loaded nodes in the system and sending them jobs that have originated in heavily loaded nodes. Load balancing algorithms vary in their complexity where complexity is measured by the amount of communication used to approximate the least loaded node. Static algorithms collect no information and make probabalistic balancing decisions, while dynamic algorithms collect varying amounts of state information to make their decisions. Eager et al. claim that the optimal level of complexity falls in the simple dynamic range, but they do not explain why more complex algorithms fail to achieve improved performance. We find that even a "perfect" balancing algorithm does not show increased performance due to a delayed information problem that increasingly affects the more complex algorithms.We obtained these results by constructing a simulation of a "perfect" balancing algorithm in which each node had immediate and free access to the states of all other nodes in the system. The surprising result was that even though each node could find the least loaded node in the system immediately, the model performed no better than less complex models. The reason why this "perfect" algorithm failed is that even if a node is able to locate the least loaded node in the system there is no guarantee that other nodes in the system are not acting on the same information and sending their jobs to the same least loaded node.The most significant parameter of the system was found to be the cost of transferring a job from one node to another. It is this cost that limits the dynamic algorithms because each node does not know how many jobs are on the way and therefore cannot send out precise information about its future state. It is this inadequate information that the more complex algorithms rely on, and it is the reason why they fail.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.