Cycle Computing’s Software Ramps Global 30,000-Core Cluster in the Cloud
September 19, 2011
Cycle Computing recently provisioned a 30,000-core cluster in the cloud, the first of its kind in the industry and the company’s second publicly announced massive cluster run this year. This milestone for scientific computing follows Cycle’s earlier scaling of a 10,000 core cluster in April. Since 2005, Cycle has helped its clients maximize the world’s compute resources through its reliable, secure and elastic HPC solutions, both internally and in the cloud…The global 30,000-core cluster was run with CycleCloud, Cycle’s flagship HPC in the cloud service. Automating the process of provisioning resources and replicating data across two continents, CycleCloud performed hundreds of thousands of computational tests with the run time per job averaging 37 minutes and the total work completed nearing 100,000 hours. The Global 500 pharmaceutical company’s researchers completed nearly 11 years of molecular dynamics computing in a few hours, at a cost of $1,279 per hour at peak, and it required no upfront capital from the client.
Cycle Computing’s flagship cluster and performance analytics software product, called CycleServer, was used to track utilization, diagnose performance and manage the progress of the scientific workflow. In addition to the CycleServer software, Cycle engineers leveraged open source projects including Condor, Linux, and Opscode’s Chef cloud infrastructure automation system.
Cycle Computing also announced today a new CycleServer plug-in for Chef monitoring and analytics, called Grill, providing visualization into what’s cooking infrastructure-wise for this 30,000-core Chef environment. Opcode’s Chef software enabled Cycle to consistently configure over 3,800 servers for this cluster, shaving off days of preparation and operational overhead. Grill enables CycleServer’s visualization and analytics-based alert technology to now support data about Chef installations.
"Cycle is unique amongst our Chef users in leveraging Chef’s ability to configure this scale of infrastructure to solve scientific problems," says Christopher Brown, Chief Technology Officer of Opscode. "We think CycleCloud is emblematic of the kind of agility Chef provides users, and are excited to see the Grill plug-in supporting Chef in Cycle’s suite of HPC analytics tools."
Leveraging CycleCloud software and Cycle’s HPC proficiency delivered these stats:
— Infrastructure: 3,809 instances with 8-core/7-GB RAM
— Global-scale: Multi-datacenter clusters with simple user interfaces
— Cluster Size: 30,472 cores, 26.7TB RAM, 2 PB of disk space total
— Security: Engineered with HTTPS, SSH & 256-bit AES encryption
The end-user experience for using CycleCloud is:
— User Effort: One-click global cluster at massive scale
— Start-up Time: Thousands of cores in minutes
— Up-front Capital Investment/Licensing Fees: $0
— Total CycleCloud and Infrastructure Cost: $1,279/hour
"CycleCloud is dedicated to changing the way the industry looks at scientific computing," said Jason Stowe, founder and CEO, Cycle Computing. "With this 30,000-core cluster under our belts, we have found our work is not only repeatable, but customizable to meet the timeframe and budgetary needs. We will continue to write software that creates secure, mega-elastic and fully-supported cloud clusters, and strive to empower researchers to answer questions that were unanswerable before CycleCloud."
To learn more about the development of the 30,000-core cluster and Cycle’s projects leading up to this accomplishment, please visit the Cycle Computing blog: Compute Cycles ( http://blog.cyclecomputing.com/ ).