High-Performance Computing for a Stronger Future

We routinely review provision of HPC services for customers and provide advice on how best to expand existing HPC services. It’s worth noting that deciding how best to deliver greater capacity without service disruption means balancing functional and non-functional requirements. For any system we help procure for our customers, we consider scalability and expansion options to help an organisation plan for capacity growth.

This process is the culmination of a lot of preparation and analysis of requirements which allows an organisation to understand the value of HPC as well as the costs. Knowing which strategy and the scale of the investment over a realistic period is vital. HPC system lifetimes are measured on the order of five years and investments in datacentres, mechanical and engineering elements are of course managed over considerably longer cycles.

This is highly dependent on the size and scale of the solution. In terms of actual build time this can typically last anywhere between a week to a month. However, depending on the extent of system testing, the full shipping, building and testing process can take up to 6 months or in some cases more.

We have established processes for managing software updates and patches in HPC clusters. Whether it is OS or applications updates, our approach includes testing updates in a sandbox environment before it is deployment in a production environment. Additionally we can work alongside the clients internal IT team to assist with vulnerability scanning, patching and remediation.

Comparing on-premises and cloud costs is not a straight forward task. Why, because you need to take into account a wide range of costs. These are not just the hardware purchase cost, of any resources eg cooling, people, and power. This task is highly dependent on each organisation and requires a full TCO model to be developed. Cloud can be a cost saving for most organisations if utilised in the proper manner eg by making the most of spot nodes and their reduced pricing.

At first glance outsourcing HPC support can seem expensive though in reality it is not. An internal HPC support team may consist of two to three FTE HPC specialists to allow for training, holidays, sickness, staff turnover etc. ROMS provides a team of HPC specialists that removes the expensive costs of an HPC support team from the customer. HPC administration requires experience and expertise. Outsourcing HPC support means your organisation and users can focus on productivity.

HPC deployments are complex and require project management through the entire procurement and deployment cycle. The project management team must have the experience and expertise of end to end HPC deployments, including the hosting infrastructure. Initially a business case must be approved followed by a rigorous requirements gathering that feeds into the RFP. The hosting infrastructure and the HPC solutions must be coordinated to ensure the infrastructure meets the requirements of the HPC. As an example consider what the cooling solution will be for the HPC as this will impact the hosting facility’s ability to provision the correct solution.

TCO evaluation assists in identifying possible dangers and uncertainties linked with HPC and AI projects such as, unexpected expenditures, and financial limitations. By being proactive about these risks, companies can put in place strategies to reduce their impact, leading to better project outcomes and financial security.