CERIT-SC Centre: Looking for Synergies in Scientific Computing

David Antoš and Aleš Køenek


Institute of Computer Science, Masaryk University


Recently we presented the intent to form the CERIT-SC (CERIT---Scientific Cloud) data and computing centre at the Institute of Computer Science of Masaryk University.  Since then a project proposal was submitted, it passed evaluation, and currently the project is in its negotiation phase. Here we describe the main principles and goals of the emerging CERIT-SC centre, and we expect to start discussions on future collaboration with the conference attendees.

Considering the startup of ESFRI and CZ OP "Research and Development for Innovations" projects, deluge of experimental data is expected in near future. The unprecedented amounts of data, together with the development of the computing equipment (increasing number of CPU cores in particular), will enforce changes of both data storage and processing paradigms. Many currently deployed methods will become unusable when applied on data larger by several orders of magnitude.

New methods will have to be developed to work with the emerging equipment and with the expected data sizes efficiently. This can be achieved only by close collaboration of experts in both the application area and IT.  The main sustaining goal of the CERIT-SC centre is provision of this IT expertise, besides computing and storage resources of competitive quantity.  Following its 15 years computing centre tradition at the Institute, the CERIT-SC centre will build an expert team of researchers, administrators, and developers, including undergraduate and Ph.D. students.  The team members will, together with the centre users, work on improvement of current tools and programs, as well as development of new computational methods appropriate for emerging scientific challenges. This joint effort is expected to yield efficient and extensive use of the computing resources, and consequently novel scientific results to beachieved.

Provision of the physical resources will follow the modern "cloud" paradigms and they will be provided to the scientific community free of charge.  The practical outcome is a very flexible resource provision---the users, once registered with the centre, will not be bothered repeatedly to pass any formal process to be granted resources; those will be available on-demand (as opposed to standard supercomputing centres).  On the other hand, the centre capacity is still limited, therefore the amount of resources actually allocated to the users will be controlled, and rather complex policies will be applied.  Once the user/group achieves recognized scientific results, it will get prioritized appropriately on top of always available entry-level share. However, such priority will decay over time if it is not "refreshed" by further results.  In this way, successful communities are prioritized while new users, students, etc. are not prevented to use the resources.

With the exception of assessment of scientific results (which will follow generally accepted criteria) all the priority assignments and resulting resource allocation will be achieved by technical means, combining advanced resource scheduling, virtualization, and the cloud paradigm; no administrative process will be involved. We have several years experience with development and deployment of such system, and we will also leverage the R&D potential of the Institute and the collocated Faculty of Informatics.

The centre will offer computational clusters of two types---"High Density" (many nodes of 8--16 CPU cores and up to 32 GB RAM) and "Symmetric Multi-Processing" (fewer "fat" nodes of more than 64 CPU cores and over 128 GB RAM), and a hierarchy of storage capacities---the fastest disks to be used during computation, standard disks for active data, and Hierarchical Mass Storage for long term data archiving. Together with an appropriate network interconnect (40 Gbps Infiniband typically), these resources cover the entire spectrum of possible user requirements, with the exception of the most demanding tasks which require specialized supercomputers.

The centre initial phase is expected to be funded by the mentioned OP Research and Development for Innovations project. According to the current plans, the first SMP cluster of approx. 500 CPU cores and 250 TB of disk space will be purchased in Q3/2011. Another 500-core SMP cluster and a 500-core HD cluster will follow in 2012. Later in 2012, the HSM storage of 3 PB will be purchased. In 2013, a 2000-core HD cluster and additional 350 TB of disk space is planned. With this installation the planned sustained size of resources will be reached. Further on, the equipment will be renewed periodically, keeping it at least at this level.

The plans of the CERIT-SC centre were clarified over the last year. Despite the project funding has not been formally confirmed at the time of writing this contribution, our expectations are positive. Our goal is to prepare both users for new infrastructure possibilities and the infrastructure for users' needs. Close cooperation of the Centre and the users, from mere ad-hoc consultations to common projects, will be beneficial for both sides. As such, it is one of the cornerstones of the Centre long-term strategy.