The VSC does not only rely on the Tier-1 supercomputer to respond to the need for computing capacity. The HPC clusters of the University of Antwerp, VUB, Ghent University and KU Leuven constitute the VSC Tier-2 infrastructure, with a total computing capacity of 416.2 TFlops. Hasselt University invests in the HPC cluster of Leuven. Each cluster has its own specificity and is managed by the university’s dedicated HPC/ICT team. The clusters are interconnected with a 10 Gbps BELNET network, ensuring maximal cross-site access to the different cluster architectures. For instance, a VSC user from Antwerp can easily log in to the infrastructure at Leuven.
- The Tier-2 of the University of Antwerp consists of a cluster with 168 nodes, accounting for 3.360 cores (336 processors) and 75 TFlops. Storage capacity is 100 TB. By the spring of 2017 a new cluster will gradually becoming available, containing 152 regular compute nodes and some facilities for visualisation and to test GPU-computing and Xeon Phi computing.
- The Tier-2 of VUB (Hydra) consists of 3 clusters of successive generations of processors with a peak capacity of 75 TFlops (estimated). The total storage capacity is 446 TB. It has a relatively large memory per computing node and is therefore best fit for computing jobs that require a lot of memory per node or per core. This configuration is complemented by a High Troughput Computing (HTC) grid infrastructure.
- The Tier-2 of Ghent University (Stevin) represents a capacity of 226 TFlops (11.328 cores over 568 nodes) and a storage capacity of 1,430 TB. It is composed of several clusters, 1 of which is intended for single-node computing jobs and 4 for multi-node jobs. One cluster has been optimized for memory-intensive computing jobs and BigData problems.
- The joint KU Leuven/UHasselt Tier-2 housed by KU Leuven focuses on small capability computing and tasks requiring a fairly high disk bandwidth. The infrastructure consists of a thin node cluster with 7.616 cores and a total capacity of 230 TFlops and a new thin node cluster with 4176 cores and 307 TFlops but also contianing 80 P100 GPUs. This is complemented with a shared memory system with 14 TB of RAM and 640 cores yields an additional 12 TFlops. A total storage of 1.2 PB provides the necessary I/O capacity. Furthermore, there are a number of nodes with accellerators (including the GPU/Xeon Phi cluster purchased as an experimental tier-1 setup) and 2 visualization nodes.