Elastic Grid Resources using Cloud Technologies

Abstract

A Large Ion Collider Experiment (ALICE) is one of four experiments at the Large Hadron Collider (LHC) at CERN. The detectors in the ALICE experiment produce data at a rate of 4 GB/s after being filtered and compressed online. The data are stored and processed in a Grid system. A Grid system allows for sharing globally distributed computing resources crossing administrative domains. The ALICE collaboration have created its own Grid middleware called Alice Environment (AliEn) to facilitate the processing and storage. This project will examine a possible way of better utilizing AliEn computing resources by using Cloud techniques, more specifically OpenStack[6] together with the virtual appliance CernVM. Cloud techniques allow for adding and removing virtual computing resources through an API, providing elasticity in a computing center. This technique gives the possibility of removing the need for physical dedicated AliEn computer resources, and instead make them disposable; the virtual computing resources should only exist while needed. This report will begin with a short general introduction and history of the technologies used in this thesis, followed by an introduction to Grid technology and AliEn. An introduction to Cloud technologies, OpenStack, and Virtual machines will then follow. After introducing the main concepts and tools, a description of a testbed and its setup process will be given, followed by an implementation of a prototype. Lastly, a short performance test, evaluation of the prototype and conclusions will follow. Results show that implementing an elastic AliEn site using Cloud techniques is indeed feasible. The solution give an overhead of ~2:30 minutes per AliEn job agent, which is short compared to the lifespan of AliEn job agents, which normally is of 48 hours. Additionally, some possible ways of further reducing the overhead will be described in this report.