Startup Brings HPC Containers to the Enterprise
The driving force behind Singularity, the popular HPC container technology, is bringing the open source platform to the enterprise with the launch of a new venture, Sylabs Inc., which emerged this week from stealth mode.
Sylabs CEO Gregory Kurtzer, who founded the Singularity project along with other open source efforts, said his startup would bring the horsepower of Singularity containers to a broader set of users. Kurtzer said the launch of Sylabs coincides with greater enterprise reliance on high-end computing. "There's a shift happening," he said.
As the enterprise container ecosystem continues to expand, most of that infrastructure is designed to deliver micro-services. The startup's goal is to deliver "enterprise performance computing," or EPC, moving beyond services to handle more demanding artificial intelligence, machine and deep learning as well as advanced analytics workloads.
"The need to properly containerize and support those workflows has grown substantially," Kurtzer said.
Given the rise of application container technology, the rapid adoption of Singularity by scientific users and the transition to data-driven workloads, the "next logical step is how to enable enterprises doing HPC-like jobs," Kurtzer added in an interview.
Along with sheer processing power, the startup hopes to differentiate itself from the vibrant container ecosystem by offering native support for GPUs, thereby allowing users to go beyond micro-services deployments to create what Kurtzer called "build once, run anywhere" applications. Those applications could then move among HPC, enterprise and cloud resources, he explained.
The startup also notes that Singularity natively supports Infinband and Intel's Xeon Phi "manycore" processors. These and other high performance components are required to accelerate data-driven computing workloads, Kurtzer noted.
Sylabs' EPC also would offer compatibility with Docker Hub, the container leader's image repository.
Container security was an early issue for pioneers like Docker. Many of those container isolation challenges have been resolved as more cloud-native applications emerge. Among the security advantages of Singularity, Kurtzer noted, is the ability to run containers in a "trusted environment" without the permissions required by most micro-services approaches.
Singularity containers also use a single file format that encapsulates the runtime environment. The advantages of that approach include compliance with security controls and the ability to validate a trusted environment by cryptographically signing a runtime image.
Along with more performance, enterprises "want things that are simple and easy to use," Kurtzer said.
Meanwhile, Microsoft (NASDAQ: MSFT) is among the first public cloud vendors to announce support for Singularity via Azure Batch, its job scheduling service that manages a pool of virtual machines, then runs scheduled HPC batch jobs on those nodes.
Singularity's ability to run on VMs and bare metal is another way Sylabs hopes to differentiate its EPC approach from micro-services. Meanwhile, Kurtzer said the startup is working to support popular container orchestration platforms like Kubernetes and Mesos. It is also working with standards groups such as the Open Container Initiative.
Founded in 2015, Singularity is now in its 13th release (version 2.4.2). Kurtzer said Singularity is daily running more than 1 million containers via the Open Science Grid, the consortium that provides distributed computing resources for scientific research.
Sylab officially emerges from stealth mode on Thursday (Feb. 8). Based in Albany, Calif., just north of Lawrence Berkeley National Laboratory where Kurtzer worked for nearly 20 years, the startup so far has a dozen employees. It is currently working with unidentified companies in beta testing.
A commercial version, Singularity Pro, is in development and will be offered via a subscription license, Kurtzer said.