Virtual Institute — High Productivity Supercomputing


ESPT'18 Travel Support from SPEXXA

Authors of accepted papers may apply for travel support... 
more information

ESPT'18 Submission is open.

Deadline for submission of full papers is July 23,... 
more information

Website disruption

Pages (or the entire VI-HPS site) might be... 
more information


Numerous VI-HPS tutorials and workshops on four continents have already taken place and further are planned: see Training for details. These feature hands-on exercises on local HPC systems or using a provided Linux LiveISO (sometimes on DVD or USB memory stick) with a typical HPC development environment for MPI and OpenMP containing the VI-HPS tools.


Due to limited time and network bandwidth typically available during tutorials, participants who intend to install a virtual machine and download the OVA archive (or ISO image) on their notebook computers should do so in advance.

The latest OVA files are currently only available as 64-bit versions, requiring a 64-bit x86-based processor and a 64-bit OS if running a virtual machine. Your filesystem also needs to handle files larger than 4GB (e.g., ExFAT). For assistance and to report problems contact


Generally it is most convenient to boot a downloaded ISO image or OVA archive within a virtual machine (e.g., VirtualBox): in this configuration, performance is reduced and hardware counters are not accessible, however, sessions can be saved and restored. Alternatively, the ISO image can be used to create a bootable DVD or USB memory stick: natively booting from this device changes no files on the hard disk and produces better performance, including access to hardware counters. Both methods provide a safe way to experiment with the tools without installing them.

Once booted, the running system provides a typical Linux HPC environment consisting of the GNU Compiler Collection (including support for OpenMP multithreading) and OpenMPI message-passing library, along with a variety of parallel debugging, correctness checking and performance analysis tools.

Depending on available memory, it should be possible to apply the provided tools and run small-scale parallel programs (e.g., 16 MPI processes or OpenMP threads on a system with 2GB of RAM). When the available processors are over-subscribed, however, measured execution performance will not be representative of dedicated HPC compute resources. Sample measurements and analyses of example and real applications from a variety of HPC systems (many at large scale) are therefore provided for examination and investigation of actual execution performance issues.