원문정보
초록
영어
Grid Computing has enabled us to utilize the unused computing power (CPU cycles) of computers connected to networks (e.g. Internet). Nowadays, there are lots of scientific projects going on in the domain of High Energy Physics (HEP) and Grid infrastructure constitutes the core computing facility of these projects. One such project is LHC (Large Hadron Collider) deployed at CERN. These experiments produce and manage a large amount of data per day and run thousands of computing jobs to process that data. The applications for these experiments require large data transfers over the network from data sources to computing resources. It is the duty of meta-scheduler to allocate jobs to most appropriate resources, and to use network in an efficient way. In this work, a Network and Data Location Aware job scheduling has been proposed for data intensive jobs. The proposed scheduling algorithm takes into account network characteristics, disk read speed of data sources, and data locations of input files, as well as other computational factors (CPU power, memory, CPU load, e.t.c) when making scheduling decisions. This scheduling algorithm aims to minimize not only file staging (data transfer) time but also turnaround time of the jobs. There are extensions to GridWay that consider the network state when making scheduling decisions, but they have not used network information in an efficient way, and have not considered data locations of input files either. Thus, they are not efficient for data intensive jobs which require huge data movement over the network. The authors have improved the GridWay MetaScheduler with Network and Data Location Aware scheduling algorithm. The improved GridWay MetaScheduler has been tested for data intensive jobs. Results presented here shows that the data transfer time and turnaround time of jobs are reduced when network characteristics, data locations of input files, and disk read speed of storage drive at data sources are considered in the jobs scheduling process.
목차
1. Introduction
2. Related Work
2.1. Improvements to GridWay Metascheduler in Literature
2.2. Inefficiencies in Existing Improvements to GridWay
3. Network and Data Location Aware Job Scheduling in Grid
4. Experiments
5. Results
6. Conclusion and Future Work
References