Analyzing QoS Approach for Parallel Soft Real-Time

Abstract

Parallel transactional systems are systems that execute transaction tasks in multiple servers, with load-balancing. We consider time constraints on those tasks and the fact that the tasks run in milliseconds means that fast admission control should be used. We propose such an admission control approach. Although there are scheduler designs building and maintaining complete deadline feasible schedules for jobs, with typical costs of O(n2) on the number of jobs, we are looking for simple and fast admission test add-on solutions to LWR that can be used with tasks running in milliseconds (e.g. transactions) as well as with lengthier jobs. We give the strictest possible bounds that guarantee deadlines, the deadline-feasibility algorithm with complexity O(n/number of servers), a system model and the approach itself. We show, by means of simulation comparison to typical alternatives, that the proposed approach is a good solution.