Providing end-to-end QoS in distributed computation using non-greedy task synchronization

Abstract

End-to-end quality of service (QoS) is difficult to provide even in networks because the input at a server depends on the behavior of all the previous servers. In the case of distributed computation, this dependency becomes even more complicated. In networks, this problem is solved by introducing shapers before servers. In this paper, we propose to use a similar approach to regulate the load offered by a cyclic distributed computation on assigned resources. Delays are introduced to a computation by means of non-greedy synchronization between tasks. This allows the limiting of the minimum computation cycle time, and therefore the maximum rate at which a task requests a service from a server. Our evaluation results show that the scheme provides deadline satisfaction and lower deviation in computation cycle time, compared to the case when no delays are introduced. 1 Grid computing, end-to-end quality of service (QoS), earliest deadline first (EDF) scheduling © 2004 IEEE.