February 6, 2015 | Written by: Fabio Furlani
Share this post:
High-performance computing (HPC) has always been seen as a dedicated, on-premises computing environment. Now with a flexible and powerful cloud infrastructure, IBM SoftLayer can change the game and make HPC in cloud a valuable option. In the last few weeks, I discussed this topic with multiple customers, so I wanted to share the five main advantages of running HPC workloads in SoftLayer.
Operational expenditure (OPEX) is the focus as an alternative to capital expenditure (CAPEX)
Major upfront capital investment for equipment can be lowered as an on-demand, pay-per-use HPC infrastructure is dynamically provisioned and de-provisioned. I normally propose HPC in SoftLayer to burst computing capacity for temporary workload peaks or to distribute compute resources in different data centers to reach a geographically dispersed group of users. When you focus on OPEX as a company, costs are more under control. If you use HPC in SoftLayer, you can access a private network at no cost. Together, these options are interesting to most customers.
Always enough but not too much
Using a flexible cloud infrastructure, over provisioning can be avoided and you can design a correctly sized and optimized infrastructure based on the workload. Customers can start with a minimum of provisioned resources and allocate new servers and storage on demand or on a scheduled plan that identifies different requirements for specific project phases.
Flexible and scalable
SoftLayer infrastructure and management systems allow you to vary capacity up or down in minutes in the case of virtual servers, or in a few hours for bare metal dedicated resources. This flexibility, linked to the automation you can reach using application programming interfaces (APIs), is the foundation of a really scalable infrastructure. New nodes in a SoftLayer data center can easily be provisioned and linked to the cluster manager and used to run computing tasks. Resources can be considered unlimited for most, and customers love the fact that they cannot run out of computing power!
Manage shorter technology refresh cycles
The average capitalization period of HPC clusters is normally 36 to 48 months. But technology runs much faster, so the normal refresh cycle should be 24 months or less. Customers are normally forced to use old technology only because it is still in their capital assets. Using HPC in cloud allows them to easily shift workloads to new technology when available and always be up to date with the most current platform for business owners.
Shortening deployment time of new projects is always a point of attention for all kinds of businesses. When a new project requires computing resources in a different geography, most of the time with stringent privacy and access laws, all customers have to consider deploying capacity in service and HPC is not different. It can help reduce deployment time and encourage improvements and innovation with no secondary effects (for instance, a possible wrong decision can be changed with a minimum associated cost).
If you need to know more, there is a wealth of information available here. I‘d like to hear about your considerations for HPC in cloud. Please leave a comment below to get the conversation started or contact me on Twitter @ffurlo.