Private Cloud for High Performance Computing (HPC) Applications
Private Cloud for High Performance Computing (HPC) Applications
Matt Regan, Program Manager at Solers
Company Profile
Solers Inc. is an innovative information technology solutions provider for partners in the U.S. Government. Solers’ technical experts partner with the Department of Defense (DoD), Intelligence Community (IC) and other Federal agencies to deliver leading edge information technology solutions that enable organizations to better achieve their mission objectives and critical national security imperatives
As an employee-owned company established in 1999, Solers has grown to over 200 full-time professionals predominately degreed in Computer Science and Engineering.
The Problem
Solers deploys new “capabilities” on a regular basis. A capability is a complex application or service usually in HPC / batch computing. Capabilities include applications with multiple components, the data, and a user interaction or workflow.
Capabilities range in size and complexity but typically include:
- 32 to 64 servers
- 50 to 200 network port
- 10 to 250 TB of data
- 10 to 20 application components. The primary component is the HPC component that is distributed across most or all of the servers.
Before Nimbula Director, capabilities took about eight months to deploy.
Tasks included:
- Developing the application
- Acquiring hardware
- Installing and configuring the equipment
- Testing the application
To properly address the capability development lifecycle, each capability has its own separated isolated set of equipment for test, development, and production.
For these reasons, deployments took too long. The eight-month cycle had many redundancies. Setting up equipment and scaling it out took a lot of the time. Solers wanted to spend less time on setting up the infrastructure and more time focusing on developing the capabilities.
The Solution
Solers built a private cloud with Nimbula Director to host their HPC applications. Nimbula Director’s zero-touch installation for growing the cloud as the capabilities requirement increased eliminated much server and network installation and configuration work.
Because Solers’s HPC workload is elastic with preemption of lower priority workloads, Solers uses the Nimbula Director API for elastic provisioning and deprovisioning. Solers is able to automate the provisioning of a workload when it starts, deprovision it for a higher priority workload, reprovision it when possible, and deprovision it when completed, returning resources to the pool for new jobs.
Why Nimbula?
Reliability: The cloud should be selfhealing and manage itself. In their evaluation, Solers deliberately killed the API service and were surprised that it recovered on its own, with the cloud fully accessible and end-users still able of deploying workload
Low opex for growing the cloud: The zero-touch installation after the first three nodes is simply to plug in the servers, turn them on, and walk away. This decreased inessential labor.
Scalability: Nimbula Director is built from the ground up to scale horizontally without needing artificial segmentation of the cloud. It’s one large pool of resources to the enduser
Performance: Rapidly deploy instances in parallel. With only three servers, they deployed 20 instances in under five minutes. The distributed image storage and local caching made mass parallel launches fasted than with any other product evaluated.
True EC2-like experience: Solers needed ahigh scale elastic self-service without the limitations imposed by traditional virtualization management solutions. Elasticity is important to Solers because their capabilities involve workloads of varying priority with surprise interruptions. They need the ability to stop one thing and start another instantly without infrastructure reconfiguration
Federated deployment: Multiple sites can be managed centrally as a single administrative domain.
Security: Most of all, Solers needed a secure solution. Self-service could not come at the expense of security holes or breaches in multi-tenancy. Nimbula Director offers a secure platform and provided other security enhancements on request.
Partnership with the vendor: No cloud product did everything that Solers needed. Nimbula showed flexibility and creativity in providing required features to Solers in advance of general availability of these features.