High Performance Computing Resources

High Performance Computing Resources

The current High Performance Computing Facility was released at the Italian Air Force Operational Centre for Meteorology (COMET, formerly managed by CNMCA, Italian Air Force National Centre for Meteorology and Climatology ) for operations in December 2009. The computing facility is a 194 nodes Hewlett Packard Linux Cluster arranged in three different groups; the first group consists of 128 nodes each of them with two quad-core Intel Harpertown 3.16 GHz onboard. The amount of volatile memory for a single node in this case is 8 GB. The second group consists of 64 dual-processor nodes with the 2.6 GHz AMD Opteron and in this case the amount of volatile memory is 4 GB, The third group is made up by two control nodes in High Availability configuration assisting the whole set of 192 computing nodes. The overall peak performance of the Computing Facility is 12,6 Teraflops.

 

Since November 2015 the ICT Support Squadron for Meteorology (G.S.I.M.) has been working to set up a new High Performance Computing Cluster (H.P.C.C.) named “RESIA”. RESIA H.P.C.C. will become operational in the second part of 2016. The cluster is composed by 51 DL380G9 computing nodes and 2 DL380G9 management nodes. Each computing node has an innovative hybrid CPU-GPU unit that consists of:

  • - 2 CPU Intel E5-2680v3 2.5GHz with 12 cores;
  • - 2 GPU NVIDIA K80 with 24GB of dedicated RAM;
  • - 64GB RAM.
Nevertheless, it has been planned an upgrade of the computing nodes number that will probably increase by the end of 2016. Moreover, it has also been planned a RAM upgrade: all computing nodes will be equipped with 128GB of RAM. Each management node consists of 2 CPU Intel XEON 5450 3.0GHz Quad-Core with 64GB RAM. Management nodes are in High Availability configuration and they assist the whole set of computing nodes. The overall peak performance of the RESIA H.P.C.C. is currently about 190 TFLOPS (290 TFLOPS with GPU software over boost) but this value is going to rise when the number of computing nodes will be increased and the RAM will be upgraded. With these improvements Italian Air Force has the will to make RESIA H.P.C.C. enter the Top500 list. The given performances are based only on those of the GPU part of the cluster (due to the fact that during the computation the CPU part of the computing nodes plays mainly a management role). It must be observed that Italy is one of the first countries which is taking advantage of the hybrid CPU-GPU technology for the elaboration of meteorological models. The best thing that comes from this hybrid technology is that it is possible to obtain peak performances with low energy consumption. Current activities are mainly devoted to the optimisation of the configuration of the cluster and to the initial tests of NWP models.