The HPC Continuum: From HPC as a Service to On-Premises
HPC environments require specialized expertise and can cost tens of millions of dollars in upfront infrastructure, hardware and facilities, not to mention ongoing operating expenses. Many states and local governments lack the financial resources and technical expertise to establish and manage their own HPC environments. At this early stage in the continuum, agencies need a way to tap HPC’s benefits without assuming the burdens of full ownership.
States should consider partnering with universities that offer advanced computing infrastructure and access to domain experts in modeling, simulation and algorithm development. Schools such as Texas Tech, the University of Texas and The Ohio State University provide shared HPC resources and technical expertise that state and local agencies can tap on demand. Partnerships with these institutions of higher education allow agencies to run complex analyses without the burden of maintaining their clusters, training internal staff or managing energy-intensive facilities.
As needs, capabilities and budgets continue to evolve, agencies may advance along the HPC continuum, bringing some or all of their HPC infrastructure on-premises. They might hire internal talent to manage offsite workloads or build and operate some of their clusters in-house.
Whatever path they follow, they should do so with a clear understanding of their strengths and limitations. They can combat these limitations by focusing internal resources on areas where they already have domain expertise, outsourcing the rest to partners. This approach reduces risks and costs while accelerating time-to-value.
EXPLORE: Delaware embraces data center as a service to modernize at its own pace.
Practical Considerations for Moving Along the Continuum
As agencies adopt these environments, they should embrace open standards that allow them to avoid lock-in and migrate workloads between service providers based on changing costs, needs or goals. For example, an agency might use one provider’s HPC cluster for a modeling task this month and switch to another next month if performance or pricing is better.
This approach helps organizations optimize workload placement based on several factors, including:
- Data gravity: Where data currently resides and how much it will cost to move
- Data governance and compliance: Regulatory constraints, such as laws requiring sensitive data to remain in-state
- Security models: Responsibility and control over data security in shared or remote environments
- Ingress and egress costs: The financial impact of moving data into and out of a given system
- Time-to-value: Whether a task requires high-speed processing or can tolerate slower, more economical systems
By planning for these considerations early and building flexibility into their technology stack, states can position themselves to scale their HPC use strategically. They can move along the HPC continuum at their own pace while avoiding unnecessary risk, cost or technical barriers.
Click the banner below to sign up for the StateTech newsletter for weekly updates.
