Univa Brings Cloud Automation to Slurm Users with Navops Launch 2.0
Slurm (Simple Linux Utility for Resource Management), distributed and maintained by SchedMD, is used by 60 percent of the world’s Top500 supercomputers and an estimated 45 percent of HPC cloud deployments, by Univa’s accounting.
Univa President and CEO Gary Tyreman sees a sizable opportunity in the market of academic and government users who rely on Slurm and increasingly are moving workloads to the cloud. Market research conducted by Univa showed 91 percent of respondents using, testing or interested in the cloud with one-third using it constantly or consistently.
“The advantage that we have [with this Navops Launch release] is the automation of the creation of a Slurm cluster in the cloud, the bursting of the Slurm cluster on-prem to the cloud; that capability is not straightforward today for Slurm users,” said Tyreman. While Slurm currently supports simple cloud bursting, Navops Launch provides additional features including policy-based automation, multi-cloud support, data migration, and cost control with cloud spend association.
Navops Launch also offers a unified view of cloud resource usage and application workloads and provides visibility into how Slurm is using cloud resources in real-time. The user can set policy-driven actions, such as shifting workloads to the cloud, scaling services, moving data to lower-cost storage tiers, or taking advantage of spot, preemptible or best-priced instances.
This functionality enabled Univa to spin up a one million vCPU AWS cluster last year for customer Western Digital, which reduced the time it took to run a complex multiphysics simulation of its hard drive heads from 20 days to eight hours.
Roughly 10 percent of Univa’s revenue today is from its core Grid Engine customers who are taking advantage of cloud, Tyreman told us. They do it one of three ways. The majority have traditional on-prem setups and use Navops Launch to run overflow, burst or specific workloads (that need specialized hardware for example) in the cloud. There’s a second set of customers that have power users who run their work in their own HPC cloud cluster, so they don’t have to share the company resource. And then the third set, the smallest segment, has migrated almost entirely to the cloud, and maintains only a small on-prem footprint.
Will Univa capture any new customers for Grid Engine by opening up Navops Launch to the Slurm community? It’s not the thesis of the strategy, said Tyreman. “We’re not providing Navops Launch [to Slurm users] to prove out Grid Engine’s better. However, we’re pretty firm believers in the capabilities of our product, and the scalability in the cloud that we’ve demonstrated with Western Digital and Amazon. We’re confident in the advanced enterprise capabilities. For those who are set and are happy, then they can continue on as they wish. It’s an opportunity to simplify the migration of the cloud. And then for those who are interested in taking advantage of some additional capabilities, for sure we’re not going to be shy about supporting that.”
“Univa’s support for open-source Slurm…is an important milestone in the evolution of HPC cloud-automation,” said Srini Chari, founder and managing partner of Cabot Partners. “Navops Launch is one of the few HPC management tools that is both multi-cloud and multi-scheduler.”
Navops Launch came to market in 2018, focusing on automated resource provisioning and application-aware cloud bursting. Navops Launch 2.0 adds features of cost control and budgeted cloud-spend association, as well as new features for workload and resource automation via composable automation applets.
Other new capabilities of Navops Launch 2.0 cited by Univa include:
- An enhanced WebUI that tracks cloud spending, usage, and resource consumption efficiency for multiple cost-centers, projects, departments, users, applications, and cloud resource types
- Improved automation features enabling administrators to easily automate actions based on a wide variety of real-time metrics related to applications, workload, and resources
- The ability to track usage against budgets and utilize automation to stay within hard budget restrictions
Tyreman emphasized the importance of cost control, given the potential for cloud to be anywhere from to 2x-9x more expensive than on-prem, a situation that is compounded when machines are left up and running, or are idle waiting for work or if the wrong workloads get deployed. “It’s a significant gap, and the gap that we see is people have some fairly simplistic methods of deploying HPC cloud. Our goal was to provide automation that would give people the tools to help reduce the expense, so it’s not magnitudes bigger.
“Putting in the automation helps reduce the cost, and then the cost management aspect is for those organizations that are out using the cloud; we’ve seen as many as three quarters not having a solution to cost visibility, cost management or even association. The user gets a bill from Amazon or one of the other cloud providers, but doesn’t necessarily know what ran where. So that’s the really big feature. And then when you couple that with the automation, you don’t just get the visibility, but you get the method to look at how you target the expense and reduce it by using automation and getting a little bit savvier on whether or not you needed a certain instance size. And then in the future, resizing that and again, using automation to move data, to turn off the machine, and to to ensure that things that are idle are simply just not being used.”
Univa is also at the front lines of multi-cloud implementation, which in some sense is also about cost control. The privately-held software company currently supports the three largest cloud providers–Amazon, Azure and Google—and would consider working with a fourth of fifth. “Our customers asked for a multi-cloud capability,” said Tyreman. “They don’t use multiple at the same time, but they want that path. They want to have that escape route. We don’t have anyone today that I’m aware of that’s running and leveraging the pricing ability that you could get from pushing back and forth. We haven’t seen people take advantage of that yet, but everyone talks about it.”
Univa primarily addressees the commercial market, with a reach that spans every vertical, according to Tyreman. Life sciences is notably strong, comprising 22 percent of Univa’s customer base. Financial services and manufacturing (industrial and semiconducting) are also important, along with machine learning, and increasingly transportation. Through a relationship with Boeing software, a number of airlines are running crew fleet scheduling workloads on Grid Engine and Tyreman told us that an airline industry mandate to move to the commercially supported version of Grid Engine will spell additional growth for the company. Univa also sees some business in the government sector.
Navops Launch 2.0 with Slurm support will be available this Fall, said Univa.
This article originally appeared in sister publication HPCwire.