D.H.J. Epema. An analysis of decay-usage scheduling in multiprocessors. In ACM Sigmetrics '95 Conference on Measurement and Modeling of Computer Systems, pages 74--85, 1995.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....reduce the need for load balancing and the effectiveness of migration (both non preemptive and preemptive) In reality, the local scheduling used in practice lies somewhere between processorsharing and feedback. Decay usage scheduling as used in UNIX has some characteristics 116 of both policies [24]. Young jobs do have some precedence, but old jobs that perform interaction or other I O are given higher priority, which allows them to interfere with short jobs. The more recent lottery scheduling [83] behaves more like processor sharing. To understand the effect of local scheduling on load ....

D.H.J. Epema. An analysis of decay-usage scheduling in multiprocessors. In ACM Sigmetrics '95 Conference on Measurement and Modeling of Computer Systems, pages 74--85, 1995.

....share group. However, Hellerstein s model considers only compute bound processes, not considering interactive or I O related activity, and is not directly applicable to real world conditions. Epema extended Hellerstein s analysis technique to decay usage scheduling in multiprocessor systems [5], showing that (i) the relationship between the shares achieved and the number of jobs in the classes is subtle, and (ii) among the schedulers studied, 4.3BSD provides the highest level of control over the share ratios. All of the work reviewed above views fair process scheduling from the user s ....

....and results. 3. 2 Workload To understand the fairness characteristics of decay usage process schedulers, we must consider realistic workloads containing a mix of the various kinds of processes found on a real system, unlike the simplistic workload assumptions made by Hellerstein [10] and Epema [5]. We base our workload selection on data we gathered through system activity monitoring experiments, extending on the studies of Shet [16] Shet monitored the load on a system continuously, classified the workload in terms of a few parameters, and predicted the response time of a process by ....

D. H. J. Epema. An Analysis of Decay Usage Scheduling in Multiprocessors. In Proceedings of the ACM SIGMETRICS Conference in Measurement and Modeling of Computer Systems, pages 74--85, May 1995.

....in the same job. This will degrade the service to all jobs, including the one that created the spurious threads. 26 There has also been work on how to set the priorities in the conventional Unix scheduler, which uses decay usage scheduling, to achieve the desired division of resources [259, 189]. 88 strained by architectural features or by lack of hardware support. Interaction with memory management Memory management is an important service provided by uniprocessor operating systems. The essence of this service is to strike a balance between the memory requirements of different ....

D. H. J. Epema, "An analysis of decay-usage scheduling in multiprocessors". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 74--85, May 1995.

....priorities are used, the low priority jobs wait less than in EASY. The weighed importance given to different considerations (for example, utilization versus priorities) can be set by the administrators. The resulting mathematical model seems simpler than other popular priority schedulers [2, 18, 9]. The following sections include a description of EASY, conservative backfilling, and finally the priority scheduler. Afterwards, performance results from simulations based on real production workloads are presented and discussed. 2 Requirements 2.1 Formalities A job that a user submits will ....

D. H. J. Epema, "An analysis of decay-usage scheduling in multiprocessors". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 74--85, May 1995.

....processes (slowdown greater than 5) by 30 60 . An issue that remains unresolved is whether feedback scheduling is as effective in real systems as it was in our simulations. For example, decay usage scheduling as used in UNIX has some characteristics of both round robin and feedback policies [Epema 1995]. Young jobs do have some precedence, but old jobs that perform interaction or other I O are given higher priority, which allows them to interfere with short jobs. In our experiments on a SPARC workstation running SunOS, we found that a long running job that performs periodic I O can obtain more ....

Epema, D. 1995. An analysis of decay-usage scheduling in multiprocessors. In ACM Sigmetrics Conference on Measurement and Modeling of Computer Systems (1995), pp. 74--85.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC