Efficient unpacking of required software from CERNVM-FS

In recent times a tool for efficient unpacking of software work-flows from CernVM File System(CVMFS) into standalone images has become necessary.There are two types of use cases for such images: On the one hand they can be used to deliverHEP software to compute nodes which do not support the traditional CVMFS delivery architecture(in particular HPC compute nodes). On the other hand the images can be used for benchmarksin which the network connectivity should not influence the benchmark results.We present a new software utility which allows CVMFS-sourced image creation and synchroniza-tion. The resulting standalone images can run HEP applications independently from an internetconnection. We further provide a methodology for automatically extracting the resource require-ments from a given software work-flow through file access tracing.In first experimental evaluations the new tool outperformed mechanisms traditionally used for thistask like rsync and provides finer grained unpack functionalities than uncvmfs.

Report

CERN Openlab Report