systemoverview
The MARBITS bioinformatics platform is built around a High Performance Computing (HPC) cluster with 22 compute nodes, one master node and one login node, mixing IBM, Fujitsu and Supermicro hardware. It provides 396 computing cores (792 with hyperthreading) and 4.6 T of RAM. Storage is provided by a shared lustre filesystem with more than 450 T of disk. It is physically located in the CPD room of the Centre Mediterrani d’Investigacions Marines i Ambientals (CMiMA), close by to the Somorrostro beach in Barcelona.
The master/login node is called marbits. It’s Fujitsu Primergy RX2530 M2 server with 2 x Intel Xeon E5-2603v4, 2 x 240 GB SSD disks and 64GB RAM. It runs CentOS 7.5, a RedHat-based linux distribution, with OpenHPC on top and slurm as job scheduler. The master node can only be accessed within the CMiMA local network.
The 22 computing nodes are divided in 3 classes according to their configuration and brand:
- 8 Supermicro nodes deployed in 2 Twin systems (2x SYS2028TP-HTR) of 4 nodes each. Each node has 2 Intel Xeon processors (E5-2650v4) with 12 cores each to sum up to a total of 192 physical cores (no hyperthreading) and 2048 GB of RAM.
- 2 Supermicro nodes deployed in 1 Twin systems (2x SYS2028TP-HTR+) of 4 nodes. (there are 2 barebones nodes that can be set up with memory and 2x CPUs in the future). Each node has 2 Intel Xeon processors (E5-2650v4) with 12 cores each to sum up to a total of 48 physical cores (no hyperthreading) and 772 GB of RAM each (1.54 T in total). They are used for memory-hungry processes such as metagenome assemblies.
- 12 IBM dx360 M3 servers with 2 x 6-core Intel X5650 2.66GHz processors, 80G of RAM and 1 x 250G 7.2K RPM hard drive each. These class adds up 144 more physical cores and 960 GB of RAM.
Marbits cluster has a dedicated high-speed network (10 Gbps; Netgear XS748T switch) for data transfer to the lustre filesystem and a 1 Gbps HPC network (SMC8150L2 switch), plus a private management network (1 Gbps SMC8150L2 switch).