Defense: Exploring Adaptive Job Schedulers For Geographically Distributed Data Centers

Daniel Alves
Computer Engineering PhD Candidate
Location
Virtual Event
Advisor
Katia Obraczka

Join us on Zoom: https://ucsc.zoom.us/j/91645827291?pwd=ZjdZSWRlaVQ0YXhudXJlaERVbCtBZz09  / Passcode: 175746

Description: To help meet the ever increasing demand for cloud computing services world-wide, while providing resilience and adequate resource utilization, cloud service providers have opted to distribute their data centers around the world. This trend has been motivating research from the data center management research and practitioner community on new job schedulers that take into account data center geographical distribution. However, designing, testing and benchmarking new schedulers for geo-distributed data centers is complicated by the lack of a common, easily extensible experimental platform. To fill this gap, we propose GDSim, an open-source, extensible job scheduling simulation environment for geo-distributed data centers that aims at facilitating the benchmarking of existing and new geo-distributed schedulers. Using GDSim, job schedulers specifically designed for geo-distributed data centers can be tested, validated, and evaluated under a variety of data center workloads and conditions. We use GDSim to reproduce experiments and results for recently proposed geo-distributed job schedulers, as well as testing those schedulers under new conditions which can reveal trends that have not been previously uncovered. We demonstrate how GDSim can be used to design and evaluate different adaptive job schedulers, which, based on current workload and data center conditions, use heuristics to select the most appropriate scheduler.