Evaluating Moldability of LHCb Jobs for Multicore Job Submission

The LHCb experiment at CERN processes its datasets over hundred different grid sites within the Worldwide LHC Computing Grid (WLCG). All those grid sites consist of multicore CPUs nowadays. However, the number of cores per worker node will increase in the near future. Using such worker nodes more efficiently requires parallelization of software as well as modifications at the level of scheduling. This paper will evaluate a moldable job model for LHCb grid jobs where the main challenge is the definition of the best degree of parallelism. Choosing an appropriate degree of parallelism depends on the parameters, on which optimization shall be applied. Commonly used features are for example scalability, workload and turnaround time. Prediction of run time is another major problem and it will be discussed how it can be handled using historical information. Furthermore, the advantages and disadvantages of a moldable job model will be discussed as well on how it must be extended to meet the requirements of LHCb jobs.