Data-intensive computing has a long history in HEP experiments. Moreover, compute-intensive HEP applications are now producing datastreams on HPC platforms that rival — and in principle can easily exceed — the data volumes produced by experiments. Analysis of the complex datasets from both of these sources — particularly so for the experiments — requires a large and complex software stack. As HPC systems are evolving in more data-intensive directions, support for these software stacks and other modes of operation are being actively envisaged.
Linux containers package an application with its entire software stack and user environment. This is a particularly useful mode of operation for HEP experiments in all of the three frontiers. NERSC has deployed Shifter for scalable running of containers in HPC systems. The applications for several HEP experiments software have now been run at scale using Shifter, as reported at CHEP 2016. This has shown benefits not only for portability, but also in performance for very small I/O transactions and software load times. HEP-CCE activity is now supporting the application of this technology for smaller or future HEP experiments such as DESI, LZ, DESC/LSST: providing HEP-specific documentation and examples as well as a HEP-CCE docker hub.
The HEP-CCE is also partnering with ASCR via the EASy (Extreme Analytics Systems) project to deploy containers for HEP applications on multiple systems, including “bare metal” and OpenStack clusters as well as HPC platforms.