Rescale and OSC Offer Supercomputing-as-a-Service: Interview with Infrastructure Partner Ohio Supercomputer Center

Rescale has partnered with Ohio Supercomputer Center, a national center that drives research and development in computational science and the applications of supercomputing, since 2015. Under the partnership, Rescale allows joint customers to run over 200 pre-installed, pre-tuned HPC applications on OSC’s specialized HPC infrastructure with zero setup or installation. Keep reading for a candid discussion with Alan Chalker, OSC’s director of strategic programs, about the ways Rescale and OSC together enable specialized supercomputing on-demand.

Owens Cluster image

Rescale: Can you start off by introducing yourself and describing your role at Ohio Supercomputer Center?

Alan Chalker: I’m Alan Chalker. I am director of strategic programs and the director of AweSim, which is our industrial engagement program. I’m responsible for all of our non-academic engagements. That includes engagement with the federal government, state government, nonprofit hospitals, and things like that.

Rescale:  What should we know about OSC?

Alan Chalker: OSC was created in 1987 by the state of Ohio. The bill that created OSC, House Bill 171 signed on June 30, 1987, recognized OSC as an economic development asset for the state and therefore required engagement with both industry and academia. That’s what sets us apart from many of the other centers—since day one, we have had a charter to serve both academics, who are the primary clients, but also industry. Because we’re viewed by the state economic development entity as an attraction agent, we can serve companies outside of Ohio. By allowing companies outside of Ohio to use our center and see the cool things we’re doing here, we raise awareness of Ohio as a good place to work and do business. Fundamentally, our goal is to provide infrastructure and computational services. We are not just providing a computer and saying “Have at it.” We are very different than AWS because we provide a wide variety of products and services, including expertise, software, hardware, and software development tools.

The current three systems we have are Oakley, named after Annie Oakley; Ruby, after Ruby Dee; and Owens, after Jesse Owens—all named after famous Ohioans. Owens is an $8 million system we acquired late last year; it has 23,000 cores in it. It’s the 202nd most powerful non-distributed computer in the world according to the TOP500 list. It has the latest generation of Intel processors, which are called Broadwell, and very high-speed InfiniBand interconnect. We’re providing leading-edge capability that you can’t get anywhere else.

Rescale: Do you have any bare metal servers?

Alan Chalker: They are all bare metal. That’s one of the big differences between what we do versus public cloud providers. Our clients have full bare metal access to nodes—nothing is virtualized. We’re not using containers, we’re not doing VMs. They are logged on to a bare metal node and have it for their dedicated use.

Rescale: How is the performance of bare metal differentiated?

Alan Chalker:  When you get a virtualized machine, there’s overhead. You don’t know exactly how many processors you’re using, how much bandwidth you have, things like that. When you have bare metal machines, you know exactly what you’re getting in terms of the number of cores, Intel’s technology generation, megahertz, and bandwidth.

I want to be explicitly clear that while we can handle almost any kind of computation on our systems, the converse is not true. If you have a really large-scale problem, you can’t accomplish it at a reasonable cost with a general cloud provider because those systems don’t scale the same way. We are ideally suited for a certain scale, speed, and class of problem. We can handle many other problems, but our forte is in large-scale problems.

Rescale:  Do you have any GPUs in your hardware systems?

Alan Chalker: Oh yes. Owens, for example, has 160 NVIDIA Tesla P100s with Pascal GPU architecture, which are the latest and greatest new generation of GPUs. Those 160 Tesla P100s have about the same theoretical computational performance—about 700 teraflops—as the rest of the Owens system, the remaining 23,000 cores across 800 nodes.

We also have some other specialized nodes. For example, we have data analytics nodes with very large RAM—1.5 terabytes of RAM per node. Typically, you need a lot of RAM available for big data problems.

Rescale:  What types of customers get the most out of your infrastructure?

Alan Chalker: It varies. We have clients that are big-name Fortune 100 companies like Honda and Procter & Gamble; we also have clients that are small startups. They span the gamut of skill sets and domains too. At the end of the day, people are going to benefit from our system if either: a) they need to have access to the latest, greatest technology because they need answers at a large scale extremely quickly, or b) they need a package deal that includes pre-installed software and advanced technical support to use the software and debug their codes.

Rescale: Can you tell us about your partnership with Rescale?

Alan Chalker:  What we bring to Rescale is capability at a very cost-effective rate. We charge a rate that is significantly lower than pretty much anybody else out there. What you bring to us is twofold. First, you guys are clients. More importantly, we are able to leverage the fact that you’re interfacing with lots of other businesses. You are basically a sales channel for us, so we don’t have to worry about reaching out to companies that you’re already working with. It’s a win-win for both sides.

As a side note, I want to say that I am thoroughly impressed and amazed at all Rescale has done over the past few years—how much it’s growing by leaps and bounds. It’s a beautiful and wonderful thing.

Rescale: Thank you for saying that. Can you give us an example of a shared customer between OSC and Rescale and how the customer leveraged our joint solution?

Alan Chalker: A good example is a thermal heating systems company that approached us and was interested in using ANSYS elastic units, or ANSYS’ pay-per-use licenses, on our systems. We don’t support the elastic licenses, but Rescale does  so we sent them to you. They used their elastic units on the Rescale platform, but they’re still using our systems behind the scenes, so everyone wins. That’s a great example of a customer that wanted to use the specialized high-performance computing that we offer and a software solution that you offer. Together we were able to deliver a solution.

Rescale:  Beyond acting as a sales channel, is there anything else that the Rescale partnership enables you to do that you couldn’t do before?

Alan Chalker: Thanks to Rescale, we now have software available on our systems that we wouldn’t have installed on our own because we either don’t have the expertise, client demand, or the staff time to do it. So that’s definitely something that you bring to the table that we didn’t have before, and we appreciate it.

This article was written by Rescale.