Benchmarks are essential in the design of modern HPC installations, as they define key aspects of system components. Beyond synthetic workloads, it is crucial to include real applications that represent user requirements into benchmark suites, to guarantee high usability and widespread adoption of a new system. Given the significant investments in leadership-class supercomputers of the exascale era, this is even more important and necessitates alignment with a vision of Open Science and reproducibility. In this work, we present the JUPITER Benchmark Suite, which incorporates 16 applications from various domains. It was designed for and used in the procurement of JUPITER, the first European exascale supercomputer. We identify requirements and challenges and outline the project and software infrastructure setup. We provide descriptions and scalability studies of selected applications and a set of key takeaways. The JUPITER Benchmark Suite is released as open source software with this work at github.com/FZJ-JSC/jubench

Minimization of the expectation EP(X) of a random variable X over a family Γ of plausible prior distributions P is addressed, when Γ is a level set of some convex integral functional. As typical cases, Γ may be an I-divergence ball or some other f-divergence ball or Bregman distance ball. Regarding localization of the infimum, we show that whether or not the minimum of Ep(X) subject to P ∈ Γ is attained, the densities of the almost minimizing distributions cluster around an explicitly specified function that may have integral less than 1 if the minimum is not attained. If Γ is an f-divergence ball of radius k, the minimum is either attained for any choice of k or it is not attained when k is less/larger than a critical value. A conjecture is formulated about extending this result beyond f-divergence balls.

The mathematical problem addressed is minimising the expectation of a random variable over a set of feasible distributions P ϵ Γ, given as a level set of a convex integral functional. As special cases, Γ may be an f-divergence or f-divergence ball or a Bregman ball around a default distribution. Our approach is motivated by geometric intuition and relies upon the theory of minimising convex integral functionals subject to moment constraints. One main result is that all “almost minimisers” P ϵ Γ belong to a small Bregman ball around a specified distribution or defective distribution P, equal to the strict minimiser if that exists but well defined also otherwise.

Version 4 of the GAP (Groups, Algorithms, Programming) system for computational discrete mathematics has a number of novel features. In this paper, we describe the type system, and the way in which it is used for method selection. This system is central to the organisation of the library which is the main part of the GAP system. Unlike simpler object-oriented systems, GAP allows method selection based on the types of all arguments and on certain aspects of the relationship between the arguments. In addition, the type of an object can change, in a controlled way, during its life. This reflects information about the object which has been computed and stored. Individual methods can be written and installed independently. Furthermore most checking of the arguments is done in a uniform way by the method selection system, making individual methods simpler and less prone to error. The methods are combined automatically to produce a powerful and usable system for interactive use or programming...