Efficient performance of the Met Office Unified Model v8.2 on Intel Xeon partially used nodes

摘要

The atmospheric Unified Model (UM) developed at the UK Met Office is used forweather and climate prediction by forecast teams at a number of internationalmeteorological centres and research institutes on a wide variety of hardwareand software environments. Over its 25 year history the UM sources have beenoptimised for better application performance on a number of High PerformanceComputing (HPC) systems including NEC SX vector architecture systems andrecently the IBM Power6/Power7 platforms. Understanding the influence of thecompiler flags, Message Passing Interface (MPI) libraries and runconfigurations is crucial to achieving the shortest elapsed times for a UMapplication on any particular HPC system. These aspects are very importantfor applications that must run within operational time frames. Driving thecurrent study is the HPC industry trend since 1980 for processor arithmeticperformance to increase at a faster rate than memory bandwidth. This gap hasbeen growing especially fast for multicore processors in the past 10 yearsand it can have significant implication for the performance and performancescaling of memory bandwidth intensive applications, such as the UM. Analysisof partially used nodes on Intel Xeon clusters is provided in this paper forshort- and medium-range weather forecasting systems using global andlimited-area configurations. It is shown that on the Intel Xeon-basedclusters the fastest elapsed times and the most efficient system usage can beachieved using partially committed nodes.