This department is responsible for architecture, management and operation of compoting clusters and related services (virtualization, planning, security), their purchase and necessary renewal after end of their moral timespan.
Clusters of two kinds – SMP and HD clusters – are Center's core computing resources. Both types of clusters are composed from independent servers (nodes), the distinguishing feature being the number of cores and shared memory capacity per node.
- HD-clusters (High Density) have smaller number of cores (usually 16-20) in shared memory with capacity usually between 16 to 32 GB. HD clusters offers large number of cores for sequential applications that are not parallelized but use a large number of concurrently run independent processes.
- SMP-clusters (Symmetric MultiProcessing) have at least 64 cores (and usually more) with a shared memory of at least 256 GB. The represent an ideal substrate for applications with enormous memory requirements and/or parallelized, composed from number of processes communicating through shared memory.
Nodes of all clusters are connected with InfiniBand, a high capacity low latency network, currently able to transmit up to 40 Gbps with latency much lower than Ethernet. Nodes of some clusters (usually the HD ones) could also be equipped with proper accelerators, e.g., high end graphics cards (GPGPU). Joint access to both types of clusters with the Center allow to fulfill practically all user's requirements, with the exceptions of those in need of true supercomputing architecture or capacity. Identical internal architecture of all clusters supports very extensive load balancing according to the actual needs. In comparison to the standard cloud instances, the available InfiniBand interconnect supports even large jobs, including those working on large data sets.
All resources are fully integrated into the national e-Infrastructure (grid), they are available through the standard grid interfaces recommended by the NGI, minimally through gLite (UMD) and Globus ones. In parallel to these grid interfaces, currently combined with batch queue systems, the computing resources will be gradually accessible through cloud interfaces compatible with commercial instances like Amazon EC2 or Google cloud.
The department is also responsible for tenders to new technologies and software, including a coordination of purchases with CESNET association and IT4Innovations supercomputing center. As part of its NGI involvement it also participates in negotiations on Service Level Declarations (SLD) for research teams and their virtual organizations.