We recently caught up with Rob Lalonde, Univa’s VP and General Manager, Navops, and posed a few questions from our readers. From open source technology to million core clusters, from turtles to golf, here’s our conversation:
Univa recently announced the open source project ‘Tortuga’, a cluster and cloud management framework. For HPC users who are now developing a cloud strategy, is Tortuga a potential on-ramp to the cloud?
Yes absolutely, Tortuga can be an on-ramp for HPC users looking to move workloads to the cloud, but the main reason we open sourced Tortuga wasn’t strictly for HPC users. Because it’s a general purpose cluster and cloud management framework, Tortuga can be used for cloud bursting, and building cloud clusters, for all kinds of technologies. So while Univa and Navops Launch are focused on Grid Engine clusters, Tortuga can be applied to other technologies, whether that’s another HPC scheduler, general purpose IT clusters and so on. Because you can use Tortuga for bursting any type of cluster it really allows for broader usage, beyond Univa products. With Tortuga, people can take the product and use it for whatever they like, even non-HPC substrates like Kubernetes.
Can you explain more about the differences between Tortuga and Navops Launch?
Sure. Navops Launch has Tortuga at its core. Launch brings enterprise capabilities on top of that, features like a Web UI, and an automation engine that allows you to apply policy for growing and shrinking the cluster, and for managing the cluster. These automation tools are very very powerful. Launch also provides reporting and monitoring. Plus, with Navops Launch you get our enterprise support.
So back in 1492, Christopher Columbus was in the Caribbean and he saw this small island near Haiti that he thought was shaped like a turtle shell. So he named it after the Tortuga, which is Spanish for turtle. I really like turtles – they’re cool, they’re cute, and it makes for a great product logo, so we went that way.
Can you explain for us how Navops Launch makes a system administrator’s life easier, in the context of enterprise HPC in a hybrid cloud? Can you give us an example use scenario?
Navops Launch makes the administrator’s life easier because it provides automation for scaling up and scaling down cloud cluster resources. Once the administrator configures the system it’s really hands off. You let it grow and shrink based on the types of workloads that are being submitted to the cluster. The administrator doesn’t have to interact every time the user needs a specialized resource. So to give you an example, Wharton School, University of Pennsylvania, uses Navops Launch to support its researchers – everything from HPC, Natural Language Processing, Machine Learning, Big Data Analytics workloads, and so on. Once Launch is configured, a researcher sitting in the University wants to submit a workload that needs GPUs, but they don’t have GPUs available on premise. Navops Launch will automatically go – based on policy – and configure GPU nodes in the cloud, add those to the Grid Engine cluster, and then release the workload to run in the cloud. The researcher doesn’t know where he or she is submitting their work to, and the administrator doesn’t have to manually configure, spin up and then shut down cloud resources, because Launch’s automation engine – and the configuration capabilities of Grid Engine – give them that composable architecture.
This summer, Univa deployed over 1 million cores in a single Univa Grid Engine cluster in AWS, which was constructed in just 2.5 hours using Navops Launch. Is this an industry first? What can we learn from this demonstration?
Well, it’s not an industry first to deploy a one million core cluster. But what is an industry first is we did it with a single scheduler. We built this million-core cluster in two and a half hours, and ran a single Grid Engine cluster on that cloud cluster, using Amazon’s Spot Fleet technology. That’s definitely an industry first. This really demonstrates Navops Launch’s ability to scale – a million core cluster is obviously massive and rare. It also speaks to the way organizations are starting to use the cloud differently. When they needed this scale and business resiliency to solve a massive problem that could not be solved before, they just didn’t have the resources to do it. Now a customer can go out and get a million-core cluster for 8 hours, solve their problem, shut it down, and save enormous amounts of money – and more importantly, gain competitive advantage.
For HPC users who are beginning to look to the cloud as a way to augment their on-premises infrastructures, what is your #1 recommendation to them, or words of advice?
My number one piece of advice? Look for a solution that makes your job easier. You can’t be manually adding nodes to the cluster in the cloud, configuring, installing images, and dealing with multiple cloud vendors’ APIs and web interfaces. There’s too much of a learning curve in that scenario. If you can avoid vendor lock-in, and have a solution that provides not only all the integration to reduce the learning curve of going to the cloud, but also have the automation that makes your administrator’s job easier. A solution that allows you to save money by using cloud resources only when you need them, and shut them down when you don’t. A solution that’s transparent to your end users, that’s really seamless, so they don’t need to know when they are going to the cloud and when they are running on premise.
So basically, Navops Launch?
Correct. Was that a shameless product pitch?
Rob, you have been living and breathing high tech for over two decades, and countless customers around the world are grateful for your knowledge, guidance and dedication. So we want to ask, do you ever unplug? What would you do with a few days of zero computers/connectivity?
First of all, sadly it’s closer to three decades not two. I wasn’t aware that countless customers are grateful for my knowledge, but I sure appreciate hearing that! Do I unplug? Yeah… I like to get away sometimes, see new places, go to the cottage, do a bit of golfing. My golf game stinks by the way! Maybe I need to unplug more often.