A comparison of preemptive and non-preemptive load distributing

Abstract

Numerous load distributing algorithms have been proposed over the past several years, with widely varying characteristics. While some of these algorithms rely solely on non-preemptive process placement, others make use of preemptive process migration. Because the state of a process becomes considerably more complex after it begins execution, the mechanism necessary for migration is correspondingly more complex than that for placement, and may incur significantly greater resource overhead. In light of this complexity, as well as the consequent implementation expense, we consider whether the addition of a migration facility to a distributed scheduler already capable of placement can significantly improve performance. We examine performance over a broad range of workload characteristics and file system stNctures. We Jind that, while placement alone is capable of large improvement in performance, the addition of migration can achieve considerable additional improvement.

Evaluation of loadsharing algorithms for heterogeneous distributed systems

“…The cost of process migration once a job has started execution can be excessive and is an operation that is difficult to implement on many current systems. So the algorithms studied are sender-initiated [9], where any loadsharing is implemented on the initial arrival of a job to the system. By convention load sharing algorithms are described by dividing them into separate policies, as first introduced in [2] .…”

Single system image: A survey

“…There are two broad approaches to the provision of unified process spaces: placement and migration [84]. Placement, also referred to as remote invocation, is a non-preemptive action that transfers the information required to start the process to another node before execution commences.…”

A Framework for workload allocation in distributed transaction processing systems

“…The distinction between migration (preemptive) and placement (nonpreemptive) strategies has been made in related work on load balancing [32]. Transaction migration E. Rahm seems difficult to realize since locks, data base pages, or terminal control blocks associated with the transaction would have to be moved, too.…”

“…The main corrective action is an adaptation of the workload allocation policy by changing the algorithms and/or adjusting control parameters used to implement the routing policy (e.g., routing table) [32,34]. With data sharing and primary copy locking, the data allocation can also be adapted automatically.…”