EigenForge only selects systems and support from 'Tier-1' vendors. We do not buy the individual components and assemble them ourselves. Instead we prefer to rely on high quality factory integrated and tested machines from the hardware vendors, and avail of their support.

With regard to the CPU processor EigenForge is irreligious as to the processor architecture; rather we will discuss the workflow with the client and make a recommendation in the capacity of a trusted advisor. In addition to the general processor architecture EigenForge is also interested in 'accelerators': FPGA, GPGPU. Whilst this technology is not mainstream it can offer tremendous code execution time improvements for certain codes.

However, some users may not require more than 1 processor for a simulation but they may require a very large volume of memory (100's of gigabytes for example). EigenForge can build this type of provision into the Service Level Agreement for our clients and provide systems with several terrabytes of addressable memory in a system image.

To maintain greater uptime plus bandwidth performance the EigenForge systems separate the systems management and user jobs roles onto their own separate networks. In addition, the user network can be physically delegated further to provide even greater bandwidth and performance: clients with specialist data or parallel communications requirements can also use a dedicated high performance low latency network for these purposes.

Linux is the most popular Operating System (OS) for supercomputing facilities and is accordingly EigenForge's primary OS. However we are also looking at the HPC Server OS from Microsoft with interest (please refer below). EigenForge can also arrange user training on any aspect of HPC, in particular operating systems, MPI, methods in code optimisation and tuning, and advanced batch scheduling techniques for example.

        Usage


Typically every client will have their own 'login-node'. The login node is your access to the supercomputing facilities and where users can access their data, perform analysis, develop and compile code. Users submit and monitor their jobs on the main systems via the batch scheduling system while logged into their assigned login-node/clone.

Clients with large number of users can have several (e.g. one per department). The login node(s) will have an identical CPU architecture and software stack as that on the supercomputer. However individual login nodes may have extra memory, graphics capability for example over and above the nodes on the supercomputer. If these are insufficient in their compute resource EigenForge can also offer a 'clone': a small replica of the main system for login and code development purposes.

The majority of computer users are more familiar with one-user desktop Windows environments; as opposed to a shared Linux supercomputing facility. These users and organisations may recognise a need to make the transition to supercomputing but are unable to do so (software site licensing agreements, training, human resource etc). The advent of HPC Server from Microsoft now makes the progression to high performance computing for these users and organisations transparent and manageable. Please contact us if you would like to learn more.