A bag-of-tasks scheduler for cloud spot market instances

From Master Projects
Jump to: navigation, search

About A bag-of-tasks scheduler for cloud spot market instances


In previous work, we have developed BaTS, a cost-aware scheduler for large-scale bag-of-tasks applications on clouds. Currently, BaTS estimates both cost and completion time of a bag of tasks by running a small sample of tasks, and performing some statistical analysis and prediction. A user can select either a cheaper solution or a faster one. At runtime, BaTS enforces the selected choice.

Right now, BaTS assumes fixed proces for each type of cloud machines. More recently, commercial cloud offerings, such as Amazon’s EC2, allow users to bid for unused compute resources, changing the traditional flat cost model. This project shall start with a study of these so-called spot market instances and the way they are assigned and removed from a user. The conclusions of this study shall be used for developing a new version of BaTS that can make efficient use of spot instances, either to lower the costs, or to speed up the execution of a user's bag of tasks.

Previous work

Bag-of-Tasks Scheduling under Budget Constraints - Ana-Maria Oprescu, Thilo Kielmann http://dl.acm.org/citation.cfm?id=1931807