HPC as a Service – Responding to Opportunities
Organizations make large investments in their HPC infrastructure every few years and then try to make due until their next big window. When new problems or opportunities arise, often costs are too high or it simply takes too long for them to add the resources they need to respond.
HPC as a Service, or HPC Cloud (private or public) helps create the agility and reduce the costs of allocating temporary resources to either service a time-critical project or help maintain service guarantees without over purchasing hardware.
Adaptive Computing can enable HPC as a Service in a transparent way making remote resources appear as a natural extension of your local system. Your policies are enforced, to ensure operations are done in a cost, SLA and placement-aware manner, and unlike other solutions we model your future resource needs and dynamically right-size your allocation, whether it be to local, Open Stack, AWS or Azure resources.
The Adaptive Computing Solution:
- Simplify job submission and management with an easy-to-use portal
- Automate best-practices through custom application templates
- Transparently manage diverse workload types including HPC, HTC, SOA, and Big Data (2nd half 2016)
- Dynamically burst allocations to local, Open Stack, AWS, Azure, and other cloud resources
- Reduce expensive visualization resource costs and file transfer times with remote visualization
- Improve resource usage efficiency with easy-to-use showback and chargeback accounting
- Serve multiple tenants with more diverse application needs by dynamically load-balancing resource allocations
Viewpoint Job Submission and Management Portal:
A successful service starts with making it easy for end users to use the service and for administrators to manage the environment and the workloads. Adaptive Computing provides an easy-to-use portal that includes job submission, file management, and a visual script builder to help reduce end user errors. The portal gives administrators insight into resource utilization through its dashboard as well as management of workload and roles and permissions. Learn more about the Viewpoint Portal.
Best Practices-based Application Templates:
We also provide an application template builder so that best practices of run time, resource usage, etc. and per-application fields can be built right into the submission form, so users avoid having to revise job scripts or even having to understand them in the first place. Learn more about Application Templates.
Support for Multiple Workload Types:
The value of an infrastructure platform can be measured by how many types of application workloads it can handle (how general purpose it is) and how well those applications can perform (it’s productivity). Adaptive Computing can enable the following workloads:
- HPC workloads (both serial and parallel)
- High Throughput Computing (HTC) workloads (for hundreds of thousands to millions of short duration single node tasks)
- SOA workloads (which need API-driven submission and near-real time responsiveness)
While others typically partition their infrastructure for each workload type, Adaptive Computing can dynamically load balance resource allocation to all of these simultaneously. In the second half of 2016 we plan to support big data workloads.
Bursting Workloads in a Hybrid Environment:
Adaptive Computing’s solution can support static and manual bursting as well as dynamic, policy-based bursting, which provides real-time auto-sizing of allocations that enforce data, resource fit, cost and SLA-based compliance. Learn more about Bursting and Elastic computing.
- Automatically burst to on- or off-premise resources according to policy settings
- Target specific clouds
- Dynamically resize the resource allocation through the cloud provider when workload is greater than all available resources
- Immediately release unneeded resources to save on costs
Adaptive Computing’s solution also provides remote visualization. Instead of waiting as data is sent back and forth from a remote resource, users can access applications or entire desktops, as if from a local machine. This increases user productivity and reduces the need to buy high-end desktops, GPU’s and other expensive resources. Learn more about Remote Visualization.
The solution not only tracks usage on a user, group, project and account basis, it also allows for pre-pay and pay-as you go models. You can charge different rates for different resources, actions, or selected qualities of service. With its intelligent lien-based model as well as budget enforcement you can avoid charging race conditions and ensure usage does not go outside of agreed upon budgets. Learn more about Moab Accounting Manager.
Multi-Tenant and Platform Support:
Enable multiple tenants, each with their own isolated view of their workload and resources and with policies that meet the need of their organization. Adaptive Computing can provide either basic workload isolation or a framework to set up each tenant with their own fully dedicated HPC, HTC, or SOA stack and then dynamically load balance resource allocations based on need. Further, this can be extended in either model to include dynamically changing the OS, libraries and configurations of the resources for specific applications or preferred environment. This flexibility expands the usability of the service to include more users with more diverse application needs.
- Manage multiple tenants according to specific needs
- Set up framework to provide fully dedicated HPC, HTC, or SOA stack to each tenant
- Dynamically load-balance resource allocations based on need
- Change OS, libraries, and configurations of resources to accommodate more diverse application and environment needs
Agility and Cost Savings with a Transparent User Experience
Moab’s intelligence right-sizes your resources in real-time, optimizes application performance and helps achieve organizational goals. Whether bursting to private or public cloud, Adaptive Computing makes HPC as a Service achieve the agility and cost savings organizations are seeking.