Herbert Schultz, IBM Manager HPC Solutions, met with Doug Beary, Account Executive, Technology Infrastructure Solutions, Datatrend Technologies, Inc., an IT Management Business Partner, to discuss Doug’s perspective on how clients today are taking advantage of HPC.
Doug, tell us a little about your role at Datatrend Technologies.
Herb, I support the HPC division within Datatrend. I consider myself a student of how end customers use HPC. I am really a sales guy that genuinely likes solving problems. I cover accounts that range from small and General Businesses to large and complex accounts with thousands of employees and students,
What types of industries do you see taking advantage of high performance computing? Most people typically think of government labs and universities.
I see in my customer set, large and small universities, as you would expect, but I also see medium sized companies that are starting to take advantage of the data that is locked in their systems and is valuable but unused. For example, I recently worked with a medium sized agricultural company with 5000 employees and 6 plants. They had installed SAP 4 years ago and had collected and stored a mountain of data but had not been using it. They were looking to unlock some of the value in that data and we worked with them to design an appliance based system that uses HPC under-the-covers to swiftly analyze their data. They are just a small example of where larger businesses are seeing value locked up in their existing data and are using HPC to exploit that data.
A number of my clients just want raw compute horsepower and want to run parallel codes and SMP (Symmetric Multi-Processor) codes. We have designed systems for these customers too. We work with IBM® hardware because it provides us with a lot of flexibility. We can choose the right IBM optimized system based on the specific client workload requirements. We use IBM servers as building blocks to make large or small HPC systems.
Manufacturers are now using HPC servers for design engineering, including applications like crash-test simulations. Petroleum companies use HPC for solutions for reservoir simulation. And Wall Street is adopting HPC in a big way.
People don’t typically think of Financial Services companies using high performance computing. Can you give me an example?
Just recently I spoke at the Linux on Wall Street show where we discussed an HPC solution for high frequency trading. This solution used the IBM System x® 3850 X5 server that can scale to large single system images and very large memory sizes. Normally, these types of systems are used in Life Sciences for Medical and Academic applications. But this Financial Services client was using the IBM x3850 X5 system to handle many data feeds and trading algorithms at the same time. The client required low latency network performance as well as an extremely large memory size and the ability to handle several processes and threads at the same time. This form of High Performance computing is really a high performance system that is “tuned” to optimize on a specific problem. That is really what HPC is now becoming, highly tuned systems for specific problem areas.
So to take advantage of high performance computing, it looks like you don’t always need to have large, complex systems. Do you have any other examples you can share with us?
Sure. Recently I worked with East Carolina University (ECU) to build an IBM System x Intelligent Cluster™ solution that was just a single IBM BladeCenter® with several HS22 blades installed. This system is simple to install and use, and was really inexpensive. The value in the system to ECU was its ability to solve problems quickly. Researchers at ECU had been struggling to run massively parallel computations in a timely fashion. The existing departmental solution was small, inefficient, and required too much management. This new solution allows the professors to ask questions based upon the answers they just received… instead of just waiting for answers. They could work as they received answers in real time. This quote from Professor Jason Bond really succinctly expresses the value of the new system to ECU: ”There are some analyses that make use of all 96 cores. Previously, a task of this magnitude might have taken a full day of computation to complete. With the IBM Intelligent Cluster, it takes just minutes.”
Doug, there is a growing concern about power and cooling in data centers. Do you have any recommendations if someone has a need for larger clusters for their workloads and are experiencing these issues?
The IBM iDataplex® is designed to optimize on performance density, efficiency and price. These are all attributes that work well in HPC and technical computing environments. IBM designed the iDataplex to be very efficient, both in space and in energy efficiency. If you look at the industry rankings for Top 500 Green HPC systems, IBM dominates the list. Over half of the systems are IBM and most are iDataplex. IBM even thought to make the airflow more efficient in these systems and therefore needed less power to drive the fans! This is how well engineered these systems are. I am able to build medium sized (500+ cores) systems in the space of a normal 42U rack footprint. The systems are typically sold by the rack and IBM builds them to order.
Can you tell us more about how your company can help clients who have a need for HPC?
I would be very interested to discuss HPC or Technical Computing business IT needs with prospective clients. Datatrend designs, builds, and installs turnkey solutions. We can help with everything from the purchase of hardware, to software installation from IBM or ISVs like ScaleMP or Platform Computing. We also support existing systems and can tune systems and storage for optimal performance . I can be reached at 919 -961-4777, [email protected].