News Interview: Altair's Bill Nitzberg 
 

Hardware or Cloudware?

Altair makes software for local high-performance computing systems and also provides HPC services through the cloud. We asked Bill Nitzberg, CTO of Altair's PBS Works division, about the changing market and the relative benefits of cloud versus local HPC. By Joe Casad

ADMIN magazine: Altair's flagship products are its software based-simulation tools for engineering and design. In the past, you were a conventional software company, selling software you licensed to other users to run on local systems, which you still do of course, but what are you doing now to meet the challenge of the cloud?

BN: Ah the cloud … I'm a little bit of a cynic. Let me first tell you my view of what cloud is and then where we play in cloud. In some ways, cloud computing is really new, and in some ways it is not. Since the 1960s people wanted to connect computers together to seamlessly create the illusion that there was just one computer. And that went through multiple iterations – in the '70s with distributed computing, in the '80s with network computing, in the '90s with network Sparc stations and cluster computing, and then in the 2000s with grid computing – and now we have cloud computing. The big picture of cloud computing, and what everybody gets super-excited about, is much more centered around traditional business computing: data center stuff, email, web, backoffice applications. The real impact and huge opportunities are in the data center side of cloud. Typical data centers in some random company are going to be using their couple of email machines maybe 20 percent of the time – so 20 percent utilization – leaving a huge amount of waste. Consolidating that is a huge opportunity.

The HPC space is a little different, and Altair plays in the HPC space. High-performance computing is about doing really hard problems and using tons and tons of computing to get the work done. Traditionally, anyone who is doing HPC has been using their systems at a 70 to 80 to 90 percent utilization or more. So the benefits of consolidation aren't as significant. But another side of the cloud is this beautiful, abstract interface between users and producers that is implemented as portals. So when you think of cloud on the business side, you don't really care what is behind the interface when you log in. It could be a whole bunch of people, or it could be a whole bunch of machines. You don't have to know. And that actually carries over from the data center market to the HPC market.

AM: Isn't it possible that it would take a very large organization to keep an in-house HPC system running 24/7 with new calculations at 90 to 95 percent usage, and that maybe cloud creates an entry into the market for smaller organizations that don't have enough work to keep an in-house HPC system busy?

BN: That's absolutely true. If you're a big company, you can build an infrastructure in-house that is cheaper than what is provided by Google or Amazon for high-performance computing – it seems, when I talk to people, at about half the cost. If you are a small player, and only once a month out of the year – or, say, two months out of the year – you redesign some part, and you only need to use HPC computing for two months out of the year, then actually using HPC cloud computing is a huge advantage over trying to buy and manage your own.

AM: As for the interface as a critical component, Altair has put out a couple of products recently: Compute Manager and PBS Desktop. How do these tools fit in with this concept of an interface to the cloud?

BN: Compute Manager really is our flagship portal – our third-generation portal – and we've learned a lot. We think Compute Manager does some really unique things. For engineers, it presents a workflow in relation to the kinds of things you need to do. I want to submit my job, I have some jobs running, and I want to actually monitor them, and I don't just mean see which ones are running and which ones are not, but I mean being able to click through all the way to the data that is running on the device, and say, "Hey, that simulation is not converging. Let me change some parameters and resubmit it," right in the middle of a run. So it is really tuned to give you a view of your simulation without exposing all the computer science details and making you learn the intricacies of Linux and Windows. That's from the user side. From the administration side, we factored out a bunch of the administrative parts in a very denotational way, basically described in an XML file, in a very portable way, so that different organizations can actually share what we call application definitions and then only have to configure the very site-specific stuff for a particular set of applications.

AM: Compute Manager is a software product you would sell to someone to run on their own HPC system. Is it also running in your own cloud system?

BN: Absolutely, Altair's HyperWorks on Demand is built on Altair's cloud stack – so Compute Manager, PBS Professional, and of course HyperWorks.

AM: And the PBS Desktop tool?

BN: When we were building portals, we couldn't decide whether the best interface we could present to our customers was a web interface, which has a large set of advantages, or an app interface more in the style of an iPhone or iPad app. So we built Compute Manager as a web interface, and at the same time, we built PBS Desktop, which is a little bit stripped down but also a little easier to use, that runs as a client application on Windows and Linux. And so, for example, one of the things that is slightly hard to do with a web application is to drag and drop a file on your desktop, but this is really easy to do with a client application. So if you're sitting in front our your Windows or Linux desktop, you can grab an input file, and drag-and-drop it on PBS Desktop; PBS Desktop will automatically recognize what application it is, fill in all the parameters, and you can just press Go. Then you can close your laptop, and open it up again later check on things. When the run is done, PBS Desktop will automatically copy all the files back for you.

AM: What would you tell people if they are thinking about getting into HPC now, and they are trying to decide: Do I want my own HPC system, or do I want to work in some sort of cloud-based environment.

BN: In the HPC world, each customer is really unique, so it is slightly hard to do a rule of thumb. One problem is, if you are just entering the HPC world now, if you don't have a lot of experience with the applications you are trying to use, that's actually more important than how you actually execute them, because if you're trying to design a new widget, you really need to know about the widget design software. Then, if you're past that threshold, you want to know: Should I use a cloud or should I buy my computers? The only rule of thumb is, you almost have to be doing it for a while to know if you're going to fit into one camp or another. If your need is really bursty, that's a better fit for a cloud. Do I want to go buy half a rack of machines and put it in my data center, and that's going to cost me, say $100,000 to $250,000? If you can actually keep it busy, you are going to save a lot of money by doing that. If you can't keep it busy, you're going to waste a lot of money.

AM: So you'd want to start by operating in some kind of on-demand basis to get some experience using the tools before you make a final decision?

BN: If somebody came to me and said, "We're running the Altair tool suite, and we're seeking to buy a cluster. Should we do that?" I would definitely say, "Come to HyperWorks on demand, try it out for a month, see how much you are using, and decide whether it makes sense – whether you are using it enough to make it worthwhile to buy a cluster or not."

AM: Altair describes Compute Manager as "… the first foundational block in Altair's plan to build a modern web-based enterprise simulation platform." Can you tell what the next block will be? What is the grand plan?

BN: I can tell you a little bit. We see that there is more to the job of engineering than just submitting the job and getting results. And Compute Manager can let you view the results and take a snapshot of the job while it is running. In engineering, there is a lot more that you want to do. You want to manage your data better. You want to get metadata about what kind of jobs you're submitting. We'll be slowly rolling out more tools in that area, inside the HyperWorks enterprise suite. Also around results and visualization, it's nice to be able to extract a piece of data and show it graphically while a job is running. It would also be nice to interact with your data – maybe you don't have to move your data off of the cloud, for instance, if you're using a cloud, and you can visualize it right there.

AM: What's up ahead for PBS Works and Altair?

BN: This year is actually very exciting for us, because Compute Manager is getting a better reception than even I thought it would. So what's up ahead for us is a lot more focus on making things easier to use for end users, and Compute Manager is our way of doing it.