Towards job accounting in existing resource schedulers : Weaknesses and improvements

The vision of having access to tremendous amounts of computation and storage resources on demand, together with access to special devices, similar to the availability of today's power grids has been formulated by Ian Foster and Carl Kesselman in [1] in 1997 and since then has been known by the term Grid computing. As this vision slowly became reality and we're now at the verge to having Grids production ready not only for scientific communities but also for industrial partners security, accounting and billing are now major concerns that need to be reflected and further improved. This paper analyzes two of the major local resource managers, Condor [2] and Torque[3], that are being used as local resource managers in the major grid middlewares Globus [4,5,6,7,8] as well as in the gLite and LOG [9,10] software stack with respect of being able to track malicious jobs and enforce a site policy. As weaknesses have been found we also present an approach that is capable of truly tracking any kind of job.