E-Handbook: HPC services bring computational power to more organizations Article 3 of 4

pressmaster - Fotolia

Tip

Key components for negotiating an HPC colocation contract

When it comes to HPC colocation, there are specialized areas to cover in a service-level agreement. Review provider protocols for power, security and network connectivity.

When it comes to selecting a colocation provider for high-performance computing, successful contract negotiation is critical. Migrating workloads is more challenging than mainstream systems, as there are specialized hardware and management requirements.

Your HPC needs may lead to slightly higher operating costs, but don't let a facility owner try and bamboozle you with costs that are considerably higher than a standard colocation setup.

Use the provider evaluation process to negotiate an HPC colocation contract that covers hardware availability, connectivity needs, facility standards and certifications as well as termination.

Power your hardware and processing needs

You must make sure that there is not only enough power for the overall data center to cover your original hardware, but also for expected power loads into the foreseeable future.

With HPC, it's likely that your power needs are high -- not only overall, but also at the per-rack level. The contract must cover your immediate needs, and include the facility owner's future plans, both at an overall facility level and at a more discrete per-cage or per-rack level to keep pace with your expected processing and power needs.

You may struggle to predict whether your processing needs will stay the same, grow or shrink in the future. The growth of third-party cloud services means that the colocation provider could shrink its infrastructure over time. So confirm that the provider has the available infrastructure.

It's unlikely you'll move your entire HPC workload over to third-party cloud any time in the near to midterm future. If your needs do grow, confirm when you negotiate HPC colocation terms that any added hardware will be either adjacent or close to the original cage. You should not have the complexity of running two cages in two different facilities as two different HPC setups, even if they are at the same colocation site.

Establish the right connection for HPC colocation

A well-architected and operated setup is useless if the connectivity is flawed. Many HPC workloads have relatively self-contained traffic. However, HPC infrastructure is increasingly dependent on high-bandwidth and high-velocity data streams to carry out any analyses.

Make sure that the facility's connectivity is already suitable for your needs -- multi-redundant wide area network connections with enough bandwidth for the traffic the overall facility is managing, as well as for your own variable needs.

When you negotiate HPC colocation services, review and confirm the facility protocol for noisy neighbor problems, where customers should have their network traffic throttled to maintain consistent performance for all other colocation tenants. Ensure that the contract makes clear that maintaining your traffic patterns within agreed limits will not get you classified as a noisy neighbor.

Have the contract dictate that the facility owner contacts you before throttling, should you exceed traffic limits. Look for dedicated high-speed connections such as Azure ExpressRoute and AWS DirectConnect that enable peer connections between the facility and public clouds, particularly where your infrastructure has dependencies on data to or from these external offerings. You should also have the contract cover continued access.

Negotiate HPC colocation provider certifications

Facilities may need specific accreditations such as Leadership in Energy and Environmental Design or Cloud Industry Forum certification. HPC has its own needs -- particularly when it comes to security.

External factors can create a domino effect and bring an HPC colocation setup to its knees.

Double-check that facility has International Organization for Standardization codes 270001 and 20000-1, plus Statement on Standards for Attestation Engagements 18 SOC 1 Type 2, SOC 2 Type 2 and SOC 3. You may need to write these certification requirements into the contract, along with how the colocation facility provider plans to monitor and maintain such accreditation.

Depending on your organization's vertical, you may also require accreditation for the likes of HIPAA or Payment Card Industry Data Security Standard, along with provable maintenance of GDPR across the facility.

You may also consider certain security standards or specific sector accreditations. If you have these extra certifications, have the contract cover any maintenance requirements and available options if the facility owner ever loses the accreditation.

Monitor your HPC setup

External factors can create a domino effect and bring an HPC colocation setup to its knees. To prevent this, you must have a monitoring system and software for your hardware and a way to get information about happenings within the colocation facility.

Make sure that the colocation facility has its own full monitoring system and that it makes the data available to you either via a portal or, preferably, data feeds that can import directly into your monitoring software dashboards.

The contract should include information regarding ongoing access and how many out-of-cage problems you can raise to the facility's own admins and help desk. Also negotiate HPC colocation remediation terms and various service levels and confirm what financial remediation options are available should the facility owner fail to maintain the service-level agreement.

Review contract and termination protocol

A long-term contract is the best option if you can't find a short-term, rolling colocation contract. However, you should avoid signing a contract that keeps you stuck with a poor level of service, is unexpectedly too pricey or can't properly support your HPC colocation needs.

With your colocation provider, agree on terms such as how much notice is required for termination, and what qualifies as grounds for termination -- for both your organization and the provider. You must accept that the facility owner may find reasons to terminate from their side, so be sure to have an adequate alternative provider.

Dig Deeper on Data center ops, monitoring and management

SearchWindowsServer
Cloud Computing
Storage
Sustainability and ESG
Close