Primeur Magazine:What kind of application is that?
Rob Lalonde:The application that the cluster is targeted at, is an EDA-style HPC application. Univa spun it up with its Navops launch product which is a provisioning solution and Cloud adapter solution. It essentially allows you to build either dedicated clusters - this was a dedicated one-million-core cluster in this case - but you can also do HPC hybrid Clouds where you are extending your on-premise cluster to the Cloud. Univa does a lot of that with its customers, using Navops Cloud.
Primeur Magazine:You spun up one million cores. How long does this take?
Rob Lalonde:One million cores: this is about 55,000 servers or instances, I should say. This takes two and a half hours, surprisingly fast. Can you imagine building a one-million-core cluster on-premise? How long that would take? Univa could put it up, using Spot Fleet which is a new technology from Amazon. They are extending their fleet capability and their spark capability to do a very large deployment. Univa took that to the limit and pretty much swamped one of their availability zones in the weekend of June 24.
Primeur Magazine:Are there other important clusters or important customers that you had recently?
Rob Lalonde:Univa is doing a lot of work in the Cloud right now. Over the last months or two, the company has announced a couple of case studies where it is doing hybrid Cloud. One is with Mellanox, a technology out of Israel and a large hybrid cluster that Univa is doing Cloud bursting to or to the Azure Cloud for the Navops launch product. The other one is Wharton School of Business where the company is also doing Cloud bursting. This one is to the Amazon Cloud. Univa supports all the major Cloud providers and they have got a lot of cool use cases lately that the company has been announcing.
Primeur Magazine:One of the nice things is that you also do, like you say, the hybrid part like Cloud bursting.
Rob Lalonde:Univa just did a recent study. I think that 60% of its customers right now - or at least of the customers the company surveyed - are in the process of going to the Cloud or are planning their Cloud process and of those, 70% are going to do hybrid rather than dedicated Cloud. There is a lot of interest in dedicated as well as in hybrid Cloud. There is a lot of investment already in those on-premise infrastructures. The company doesn't want to throw that away and just switch to Cloud. They want to burst that and take the peaks of those clusters. When they don't have enough resources available on premise, they will burst to the Cloud or maybe they want specialized resources like GPUs or really big memory machines or things like that. Then, they can just go to the Cloud when they need those and use those for a short time and shut them down.
Navops launch has the ability to grow and shrink the cluster based on policy. If people need specialized workloads, they can use those, shut them down when they don't need them and therefore save themselves a lot of money.
Primeur Magazine:Is it mainly for firing up virtual machines or are you also doing bare metal?
Rob Lalonde:Univa also supports bare metal. It supports OpenStack and bare metal clusters. If that is of interest, that is something the company can do. Right now, Univa is seeing mostly interest in public Cloud but it is seeing people do it using on-premise infrastructure as well.
Primeur Magazine:Can you tell a little bit more about the new technology?
Rob Lalonde:Univa has been in the Grid Engine business doing workload management for quite some time. The company has moved that technology along pretty rapidly. It has done a lot of modernization around that technology so it is seeing a lot of container usage, both Docker and Singularity, within its customer base. Univa is seeing a lot of advances around that and the company is certainly a leader in container capabilities. Univa has done a lot of work with GPUs with its solution. A lot of its customers or almost all of them are doing some NVIDIA computing and some part of them reserve their cluster for machine learning or other types of applications. The company is seeing a lot of activity around that.
There is the company's Navops launch product for doing the Cloud bursting and all of its customers now seem to be talking about Cloud and heading in that direction or even investing in that. Univa is keeping itself really busy with that technology. Univa has announced support for new public Cloud providers. It supports Oracle now, Amazon Microsoft and Google for a long time. Univa supports all the major Cloud providers. It has been adding new policy capabilities to that so it can do things like help manage the data or at least interact with a data manager or a third party data management provider so that the data gets into the Cloud before the workload arrives to run. It is really important how you synchronize that data. There are lots of really interesting capabilities. Univa is really at the forefront with what it is doing with the Navops launch product.
Primeur Magazine:You mentioned containers. Is that something that is now taking off?
Rob Lalonde:Absolutely. A few years ago, people were just talking about it but it seems like everybody is doing it now in some form. They have not moved all their workload to containers because that takes quite an effort but either because they want that portability, maybe they are moving it to the Cloud, maybe they just want to guarantee that what is in the container, is going to run the same way every time and get rid of software dependencies, those types of things. There are lots of reasons why people are doing containers. It is becoming very popular and it is moving kind of in concert with that growth and that migration to the Cloud that you are seeing. There are a lot of parallels between that and containers because they work well together.
Primeur Magazine:Thank you very much for this interview.