Speaker
Mr
Mario Lassnig
(CERN & University of Innsbruck)
Description
Unrestricted user behaviour is becoming one of the most critical properties in data intensive supercomputing. While policies can help to maintain a usable environment in clearly directed cases, it is important to know how users interact with the system so that it can be adapted dynamically, automatically and timely.
We present a statistical and generative model that can replicate and simulate user behaviour in a large scale data intensive system. This model can help site administrators to anticipate future workload from users and therefore provide accurate local improvements to their storage systems. The theoretical foundation of the model is examined and validated against experimental results from the ATLAS distributed data management system DQ2.
Summary
Analysing user behaviour to improve the performance of distributed mass-storage systems.
Primary author
Mr
Mario Lassnig
(CERN & University of Innsbruck)
Co-authors
Mr
Miguel Branco
(CERN & University of Southampton)
Dr
Vincent Garonne
(CERN)
