9-12 May 2017
Europe/Rome timezone

The OCCAM Open Computing Cluster: flexible provisioning of HPC resources

Not scheduled


Stefano Bagnasco (INFN)


The Open Computing Cluster for Advanced data Manipulation (OCCAM) is a HPC facility designed and operated by a collaboration between Università degli Studi di Torino and the Torino branch of the Istituto Nazionale di Fisica Nucleare. It is aimed at providing a flexible and multi-purpose infrastructure to cater to a broad range of different scientific computing needs, as well as a platform for R&D activities on computational technologies themselves. The system includes different types of computing nodes (standard nodes, large 4-way nodes, and GPU nodes), a scratch storage intended for intensive random access, powered by a parallel filesystem with high I/O performance, and a larger partition for data staging, requiring less intensive access. Two network layers connect each component: a 10Gb/s Ethernet and an InfiniBand FDR with a fat tree topology. The broad and variegated range of use cases challenges for great flexibility in providing resources to the final users. We choose to borrow some ideas from Cloud Computing technologies introducing the concept of Computing Applications, each defined by its runtime environment, resource requirements and an execution model. By dynamically partitioning the system, we deploy consistent and sandboxed sub-clusters tailored to satisfy a Computing Application’s requirements. The long-term objective is to provide a full PaaS-like environment to enable skilled users to build their own applications. The pivotal technologies for the middleware architecture are Linux containers, currently managed with Docker, and several building blocks and orchestration tools taken from products developed by the INDIGO-DataCloud project. As an example, one of our first provisioning models is based on the Batch System as a Service INDIGO product, leveraging HTCondor.

Primary authors

Prof. Marco Aldinucci (Computer Science Department, University of Torino) Matteo Concas (INFN Torino) Paolo Pasteris (Dept. of Computer Science, University of Torino) Dr Sara Vallero (INFN Torino) Sergio Rabellino (Dept. of Computer Science, University of Torino) Stefano Bagnasco (INFN) Dr Stefano Lusso (INFN Torino)

Presentation Materials