InfraGRID – UVT HPC Center – InfraGRID Cluster
InfraGRID Cluster consists of 50 compute nodes powered by IBM BladeCenter-H technology. The entire solution is built up from 4 blade center chassis, each with 14 HS21 blade server. Each blade server has a dual quad core Intel Xeon E5504 CPU (clocked at 2.00Ghz) and 10GB RAM memory. The connectivity is delivered by: (a) Infiniband for interconnect and storage, (b) fiber channel for dedicated storage and (c) Gigabit for service networking. This cluster was initially setup in 2009. In 2011 the InfraGRID Cluster received an update by adding a new IBM BladeCenter-H chassis powered with 7 dual CPU/GPU HS22 blade servers. The CPU is powered by Intel XEON technology with a clock speed at 3.46Ghz and 32GB RAM memory. The GPU cards are NVidia Tesla M2070Q (448 GPU cores and 6GB GDDR5 RAM memory). All hardware details are available in the table below.
The cluster is managed by two service nodes, one dedicated to user access (also called head node) and one dedicated exclusively for service actions and cluster management. The storage is shared using GPFS file system over Infiniband using two dedicated NSD nodes. The service management is conducted using IBM BladeCenter the advanced management module (AMM) that is built in into both blade center chassis and blade servers (IMM). This feature allows remote administration and monitoring of any of the installed hardware. Infragrid Cluster is cooled with air. The displacement of the cooling units is in-row with a well delimited hot-cold area. Currently three APC InRow cooling units are installed and working in a cluster configuration to obtain cooling unit high-availability behavior.
Administrative data
| Name | InfraGRID Cluster |
| Short Description | UVT HPC Center – InfraGRID Cluster |
| Owner | Universitatea de Vest din Timisoara (UVT) |
| Country | Romania |
Dedication to VI-SEEM
| CPU (percent) | 10% |
| Storage (percent) | 0% |
| Accelerators (percent) | 20% |
| CPU (core-hours per year) | 798,912 |
| Storage in TB | 0 |
| Accelerators (hours per year) | 5,494,272 |
Integration
| System operational since | Sep 2009 |
| Available to the project from | PM04 |
| Expected date to be phased | - |
| Interfaces | SSH |
Computational Power
| Number of servers | 57 | Interconnect type | QDR InfiniBand | Peak performance (Tflops) | 2.11 | |
| Server specification | IBM BladeCenter HS21 | Interconnect latency | 2.5 μs | Real performance (Tflops) | 3.5 | |
| CPU per server | 2 | Interconnect bandwidth | 40 Gbps | Operating system | Red Hat Enterprise Linux | |
| RAM per server | 10 GB | Local filesystem type | GPFS | Version | 6 | |
| Total number of CPU-cores | 400 | Total storage (TB) | 50 TB | Batch system/scheduler | LoadLeveler | |
| Max number of parallel processes | 800 | Peak performance CPU (Tflops) | 3.5 | Accelerators type | Nvidia Tesla M2070Q | |
| Number of cores | 448 | Servers equipped with accelerators | 7 | Peak performance accelerators (Tflops) | 3.5 | |
| Accelerators per server | 1 | |||||
| Development tools | IIntel Compilers (C/C++, FORTRAN), GNU Compilers, OpenMPI, CUDA, gdb, pgdbg | |||||
| Libraries | Intel MKL, HDF5, FFTW, NetCDF, GSL, LAPACK, Boost, BLAS | |||||
| Applications | Misc. | |||||