3 June 2019

Simulation facilities

CAI Team (Research Support Cluster)It is composed of 3 chassis class-C (c7000) of HP,each of which harbor 16 nodes bl460c gen8, withwhich we obtain a total of 48 nodes. Each chassissupports those 16 nodes with only 3 power supplies,although by edundancy have 6 units now.

It is also cooled by 10 PARSEC fans. Each node has 2 processors Intel Xeon E5 2670 at 2.6 GHz. Each processor is octo core, it means, eight core, so we have 48 × 2 × 8 = 768 calculation cores. The efficiency is 8 floating-point operations per core and cycle (8 flops), so that each node yields 2.6 GHz × 8 flops = 20.8 Gflops. For the 768 cores we have 20.8 Gflops × 768 that approaches 16 Tflops, that is, 16 trillion floating point operations  per second. Each node has 128 GB of RAM, so the cluster as a whole includes more than 6 TB of main memory. Local storage is done on a SATA  disk drive of 500 GB  capacity and 10 krpm.

However, global storage, node images, scratch spacing and user accounts are implemented by means of a HP (Storage Area Network) HP P2000.

This device has double  AID contoladora, and 60 TB gross capacity. The fiber connection to the nodes reaches 8 Gbps.

In relation to the network, three differentiated  networks have been designed. The calculation network, inaccessible from the outside, has 10 Gigabit Ethernet architecture of ultra low latency, and allows the exchange of information between the nodes when computing. The main backplane reaches 160 Gibps, connecting each chassis to 40 Gbps. The other networks of consoles and internal management allow the management of the nodes and their cloning and configuration.

Supercomputer cluster