Cluster

The Roma 3 GRID Laboratory was established in 2008 thanks to a project that involved as main founders the Atlas and the ARGO-YBJ experiments with the financial support of INFN and Department of Physics of the Roma Tre University. After the first year of usage, the interest from other groups within our structure has grown and therefore the sharing of resources to all groups belonging to INFN and the University that require more computing power was extended.
The site is a TIER 3 for the ATLAS experiment and has been a TIER 2 for the ARGO experiment, it also used for several projects related to development activities (EG:CHAIN-REDS, agINFRA, EUMEDGRID, EUChinaGrid, DCH-RP, EGI-InSPIRE.) and a comunity of local users use the resources “locally”.
The Data Center actually consists of five racks (42 U) cooled by an air conditioning system with direct expansion of gas-based air flow inside the rack. This architecture is therefore “heat-isolated” from the room. Inside the room there is also a UPS that prevents the sudden shutdown of equipment in the event of a blackout. All the machines are connected within the rack through a 1 Gigabit and 10 Gigabit Ethernet network.
Concerning the computing power, the site is equipped with about 50 server (mainly based on Blade tecnology) with a total amount of cores available (or VCPU) of about 1500 interconnected with Infiniband (DDR 20Gbps e QDR 40Gbps). The site has also 12 Graphical Processor Unit (GPU) system NVIDIA Tesla 2070 and K80.
All the users data and experimental data is stored in one of the three storage system present in the cluster for a total amount of about 700TB interconnected with 6 machines (DS) using 4 Gbps / 8 /16Gbps Fiber Channel connections. In order to share the data between all the cluster in each DS is installed GPFS (General Parallel File System) server provided from IBM.
All the services needed to manage the computing site are supported by: two Computing Element (CE), three User Interface (UI) units, eight machines for storage management, one Grid Storage Element (StoRM) and 2 machines to monitor all the systems.
The CE is a queue manager with batch queuing system based on PBS/MAUI and, in order to offer the resources in the grid, it is also configured as a CreamCE. The CE is configured to accept jobs of users belonging to several Virtual Organization (VO).

Currently the staff is studying and testing the latest cloud technologies in order to evaluate a possible migration to these technologies