| Description: |
Medium size HPC clusters play an important role in the HPC landscape in that they provide both the training environment for system scalability and a flexible production field for a large class of numerical problems. In this poster we present CRESCO4, the latest medium size HPC cluster purchased by ENEA, in operation since few months. CRESCO4 is part of a family of HPC systems, all integrated within ENEAGRID, a large infrastructure for cloud computing, which includes all the computational facilities installed at several ENEA sites in Italy. Keywords-HPC, supercomputing, Linux clusters I. ROLE OF MEDIUM SIZE HPC CLUSTERS Since a long time it has been established that performance data populating the TOP500 list tend to follow Zipf distributions [1], e.g. with k denoting the rank in the list and A and b representing time depending constants. In particular, it is known that A tends to increase exponentially in time as a reflection of the Moore's law. More subtle is the trend of the exponent b: while displaying a slight decrease over the years 1999-2007 [2], it shows a slow growth since 2009 up to date, thus signaling an increase in time of the unbalance of computing power between higher and lower ranks in the list a . More precisely, (1) holds not only on the global stage but also at a national level. If this is correct, we can speculate that both global and local HPC ecosystems have been recently evolving towards a concentration of computing power within larger and larger facilities, aimed at reaching the exa-scale computing, whereas -by contrast -intermediate machines are suffering from a less focused attention. |