Citation Context

...eduler to perform better packing. Moreover, a recent study indicates that users are actually quite confident of their estimates, and most probably would not be able to provide much better information =-=[23]-=-. Estimates also embody a characteristic that is particularly harmful for backfilling: they are inherently modal, as users tend to choose “round” estimates (e.g. one hour) resulting in 90% of the jobs...

Citation Context

... of hardware configurations. For example, in HPC there is considerable investigation into the relationship between requested time and execution time, a fertile ground for optimization [23] and debate =-=[19]-=-. In cloud computing, services such as web search and email do not have a requested time or deadline. Moreover, in cloud backends, the use of admission control and capacity planning ensure that jobs h...

Citation Context

.... The span of the workload is about 12 days. The actual computational cost of each job is set to U(30 s, 10 h, 1 s). The estimated computational cost of the job may be different from the actual value =-=[14]-=-.To simulate the accuracy of the estimated job computational cost, we use the following method that is derived from the model in [15] with modifications to conform to the fact that users often overest...

Citation Context

... a market implies that users must be able to estimate the application demands in order to submit their job. Experience with supercomputer users suggests that users have a hard time trying to do this [=-=Lee et al., 2004-=-]. To circumvent deployment difficulties imposed by the grid economy, OurGridstakes the peer-to-peer systems approach. Most current peer-to-peer systems share files and storage rather than computing p...

Citation Context

...assure good scheduling; information about tasks sizes is also needed, and such information is often difficult to obtain. Asking the user about task execution time produces notoriously bad information =-=[16]-=-. Sometimes the only way to obtain it is to submit the task on a given resource. A task can be submitted in all resources and its execution time can then predict the execution time of the remaining ta...

Citation Context

... However, user estimates are often notoriously off from the actual amount of work. This has been evident in both volunteer computing systems such as BOINC and also with the usage of traditional MPP’s =-=[13]-=-. As such, the scheduler uses a duration correction factor to offset any error in estimation based on the history of executed jobs. In this section, we measure the impact of using DCF for a range of e...

Citation Context

...m is not aware of dependencies between those jobs. Although some users may have knowledge about the processing time of their jobs, some studies show that user estimates are unreliable, see Lee et al. =-=[6]-=-. These observations indicate that Garey and Graham’s model is still a valid basic abstraction of a parallel computer and its applications. Our Grid model simply extends this model by assuming that th...

Citation Context

...tes in impacting system performance than job allocation is. C. Efforts on Improving User Estimations Numerous efforts have been devoted to improving the accuracy of user runtime estimates. Lee et al. =-=[19]-=- tried to improve user estimation by removing the threat of job killing at walltime expiration and providing tangible reward for accurate estimates. However, the experiments show their method leads to...

Citation Context

...ber of hosts. A user estimation for task executing times may be used, but most of user estimations are incorrect [10] and the less knowledgeable the user is, the more prone to error his estimates are =-=[8]-=-. If the provided estimates are below the real time, a runtime adjustment should be made, if the estimation are high more hosts than the necessary are allocated. Furthermore, the automation of task ru...

Citation Context

...using non-perfectly accurate execution times (e.g., using inaccurate userprovided estimates, multiplying the perfectly accurate estimate by a certain factor). We refer the reader to the discussion in =-=[37]-=- for more details. At any rate, in these studies the potential advantage of using inaccurate estimates is shown to be relatively small, while our results show that our approach outperforms EASY by ord...