Section: New Results
Resource Management and Scheduling
Participants : Eddy Caron, Frédéric Desprez, Gilles Fedak, Jose Luis Lucas, Christian Perez, Jonathan Rouzaud-Cornabas, Frédéric Suter.
Resource Management Architecture for Fair Scheduling of Optional Computations
Most High-Performance Computing platforms require users to submit a pre-determined number of computation requests (also called jobs). Unfortunately, this is cumbersome when some of the computations are optional, i.e., they are not critical, but their completion would improve results. For example, given a deadline, the number of requests to submit for a Monte Carlo experiment is difficult to choose. The more requests are completed, the better the results are, however, submitting too many might overload the platform. Conversely, submitting too few requests may leave resources unused and misses an opportunity to improve the results.
In cooperation with IRIT (Toulouse), we have proposed a generic client-server architecture and an implementation in Diet , a production GridRPC middleware, which auto-tunes the number of requests  . Real-life experiments show significant improvement of several metrics, such as user satisfaction, fairness and the number of completed requests. Moreover, the solution is shown to be scalable.
Advanced Promethee-based Scheduler Enriched with User-Oriented Methods
Efficiently scheduling tasks in hybrid Distributed Computing Infrastructures (DCI) is a challenging pursue because the scheduler must deal with a set of parameters that simultaneously characterize the tasks and the hosts originating from different types of infrastructure. In  , we propose a scheduling method for hybrid DCIs, based on advanced multi-criteria decision methods. The scheduling decisions are made using pairwise comparisons of the tasks for a set of criteria like expected completion time and price charged for computation. The results are obtained with an XtremWeb-like pull-based scheduler simulator using real failure traces for a combination of three types of infrastructure. We also show how such a scheduler should be configured to enhance user satisfaction regardless their profiles, while maintaining good values for makespan and cost. We validate our approach with a statistical analysis on empirical data and show that our proposed scheduling method improves performance by 12-17% compared to other scheduling methods. Experimenting on large time-series and using realistic scheduling scenarios lead us to conclude about time consistency results of the method.
Fair Resource Sharing for Dynamic Scheduling of Workflows on Heterogeneous Systems
Scheduling independent workflows on shared resources in a way that satisfy users Quality of Service is a significant challenge. In  , we described methodologies for off-line scheduling, where a schedule is generated for a set of known workflows, and on-line scheduling, where users can submit workflows at any moment in time. We consider the on-line scheduling problem in more detail and present performance comparisons of state-of-the-art algorithms for a realistic model of a heterogeneous system.
Image Transfer and Storage Cost Aware Brokering Strategies for Multiple Clouds
Nowadays, Clouds are used for hosting a large range of services. But between different Cloud Service Providers, the pricing model and the price of individual resources are very different. Furthermore hosting a service in one Cloud is the major cause of service outage. To increase resiliency and minimize the monetary cost of running a service, it becomes mandatory to span it between different Clouds. Moreover, due to dynamicity of both the service and Clouds, it could be required to migrate a service at run time. Accordingly, this ability must be integrated into the multi-Cloud resource manager, i.e. the Cloud broker. But, when migrating a VM to a new Cloud Service Provider, the VM disk image must be migrated too. Accordingly, data storage and transfer must be taken into account when choosing if and where an application will be migrated.
In  , we have extended a cost-optimization algorithm to take into account storage costs to approximate the optimal placement of a service. The data storage management consists in taking two decisions: where to upload an image, and keep it on-line during the experiment lifetime or delete it when unused. Based on our experimentations, we have shown that the storage cost of VM disk image must not be neglected as done in previous work. Moreover, we have shown that using the accurate combinations of storage policies can dramatically reduce the storage cost (from 90% to 14% of the total bill).