Extreme-Scaling Applications 24/7 on JUQUEEN

The High-Q Club showcases applications which demonstrate scalability to effectively utilise all 458,752 cores of the JUQUEEN 28-rack Blue Gene/Q system at Jülich Supercomputing Centre. A broad spectrum of more than 24 application codes have qualified for membership, using up to 1.8 million processes and/or threads. Seven application code-teams used the opportunity provided by the 2015 JUQUEEN Extreme Scaling workshop to (im)prove their code scalability -- assisted by JUQUEEN and IBM technical support and experts from JSC Simulation Laboratories and Cross-sectional teams -- and within the first 24 hours of dedicated access to the entire JUQUEEN resource, all 7 participating teams had adapted their codes and datasets to exploit the massive parallelism and restricted node memory for successful full-system executions. We compare the characteristics of the workshop and High-Q member codes, considering their strong and/or weak scaling, exploitation of hardware threading, and whether/how multi-threading is employed intra-node combined with message-passing. Scalability inhibitors such as inefficient use of limited compute node memory and file I/O are identified as key governing factors of applications on JUQUEEN which are expected to impact their ability to exploit expected exascale computer systems