Flemish tier-1 Supercomputer

Ghent University is committed to be an important partner in the Flemish Supercomputer Center and also continues to invest in supercomputing, which is fully in line with the strategic objectives of our university. Our university finances over €600,000 per year in HPC, i.e. investments in infrastructure, staff (support, maintenance and coordination) and operational expenses (cooling, power, etc.). In 2011 we also opened a new datacenter, with a state-of-the-art supercomputing room (additional investment of €3 million). The HPC room will host tier-2 equipment and the first Flemish tier-1 supercomputer.

Datacenter S10 VSC

Both KULeuven as UGent are very committed to host, in turn, a tier-1 supercomputer. In the fall of 2012, the first Flemish supercomputer will be taken into production in Gent. The system will remain operational until 2015. Tailored to the commissioning of the second tier-1 in Leuven, the UGent system will be eventually deployed in the tier-2 grid. From 2017 onwards UGent will host the third tier-1.


Flanders' first tier-1-supercomputer


After a European call for tenure Ghent University, together with the Flemish Supercomputer Centre (VSC), selected HP to construct the first Flemish supercomputer. This cluster of computers will become the largest academic computing infrastructure in Flanders. 
The High Performance Computing (HPC) system that was proposed, with a price tag of about €4.2 million, is based on the ranking at nr. 118 in the Top500 list of June 2012 also the biggest supercomputer in Belgium.

The system will be housed in the new datacenter of Ghent University at site Sterre, but can also be used by other institutions that are part of the Flemish university and college associations, and by Flemish public research institutions. It is possible that also Flemish companies will be able to use it. The system will run workloads in various fields, including space weather prediction, fluids and plasma dynamics, nano technology, material modelling, biophysics, bio-informatics and computational chemistry.

The supercomputer with Red Hat Enterprise Linux as operating system will have a peak performance of 152.3 teraflops (1012 FLoating point OPerations per Second). HP has selected the Intel Sandy Bridge microprocessor technology with a total of 8,448 compute cores and FDR Infiniband by Mellanox. HP is also responsible for the network component and cluster managment. The specialized data storage (450 terabyte, equivalent to 450.000 gigabyte) is taken care of by Data Direct Network (DDN), while Adaptive Computing delivers the scheduling software. The datacenter infrastructure is installed by APC.

The Flemish supercomputer will go into production during the second half of 2012. The funding is done by the Flemish Government, Department of Economy, Science and Innovation and the Hercules Foundation. Although the structural financing of the Flemish Supercomputer Centre is yet to be confirmed, the necessary means to cover the energy bill and pay for employment costs in 2012 have been made available by the Flemish Government.

European HPC ecosystem

Below, a representation is given of the European HPC ecosystem pyramid which was defined by the European Commission.

At the top of the pyramid there are the super-supercomputers. The first tier-0 is located in Jülich (Duitsland) and there is one installed in France near Paris. The European tier-0-research infrastructure will be further strengthened  with two machines in Munich and Stuttgart. These type of computers have a size of 2 to 5 petaflop (1015 FLoating point OPerations per Second).

The operating costs of these large supercomputers are not to be underestimated. The energy cost alone is enormous. The current average tier-0 ('petascale') have a capacity of around 2 megawatt (MW), and with the ambitious race to an exascale platform (FLOPS 1018), the consumption may reach up to 20MW in 2018.

Ecosysteem HPC

Flanders is already working on the next generation of tier-0 systems, particularly in the field of software development: in collaboration with the five Flemish universities, Imec and Intel the Flemish government has set up the Flanders Excascience Lab .

The next level, which in Europe is called Tier-1, are national and/or regional supercomputers which have sufficient computing power for the mainstream European supercomputer user. This is an important layer because a tier-1 infrastructure allows the build-up of the necessary expertise and knowledge required to develop and use a tier-0-supercomputer. These systems currently have an average size of 100 to 200 TeraFLOPS. Flanders will inaugurate their first tier-1 in 2012, which will be a first step towards Flanders being part of the European HPC ecosystem.

This bottom layer of the pyramid consists of local and smaller supercomputers which typically can be found at the different research institutions. A tier-2 today typically has a size of 10 to 30 teraflops. In recent years, the universities deployed many tier-2 clusters which are currently being linked in a tier-2 grid.

tier-2-grid Vlaanderen