A partnership between Dalhousie University computer science researchers and high-performance computing vendor Liquid Computing Inc. will help both parties learn more about how to do parallel online analytical processing (OLAP) on Liquid’s unusual fabric computing architecture, the two parties say.
Ottawa-based Liquid Computing, which completed its first financing in 2005, has developed an architecture that, according to Keith Millar, vice-president of product management, emphasized communications as much as computing power. The company’s founders came from the telecommunications industry, he explained, and they developed what they call a fabric-based computing model that is easy to scale.
The High-Performance Computing Lab at Dalhousie in Halifax works extensively with other disciplines on data-intensive problems, said Dr. Andrew Rau-Chaplin, a Dalhousie computer science professor and leader of the university’s parallel data mining and OLAP project. About half the lab’s activities are currently focused on parallel data warehousing and OLAP research, he said.
“Our interest is looking at how we can apply parallel computing techniques to these data-intensive problems,” Rau-Chaplin said. “If you want to do that, you don’t want to handicap yourself before you start by looking at the wrong kind of machine architecture. What really excites us about Liquid Computing is, I think personally they’ve got it right. You really need to focus in on network performance, in terms of bandwidth and latency. So they’re an attractive platform to work jointly with on those kinds of problems.”
Liquid Computing, meanwhile, hopes that working with the Dalhousie researchers will help its own developers understand better what parallel OLAP applications require from the hardware, so that they can further refine their product.
“What is interesting is the potential of how those algorithms can evolve on our platform,” Millar said.
Rau-Chaplin said the collaboration started with a “corridor conversation” at Canada’s annual High-Performance Computing Symposium, held last year at Memorial University in St. John’s.
“One of the great things about (HPCS) is it brings together all of the players in high-performance computing here in Canada,” said Rau-Chaplin. Liquid and Dalhousie began talking after last year’s HPCS, he said, and their formal partnership got under way in January. “As we’ve gone along,” Rau-Chaplin added, “we’ve seen more and more opportunities for dialogue on both sides.”
The partnership has no fixed term and no mandate to develop a product. “I see it as an open-ended collaboration,” Rau-Chaplin said. Liquid is providing some of its hardware to Dalhousie, Rau-Chaplin said, but the equipment is less important than the dialogue.
“Being able to talk to the guys that build the hardware and design the software systems, being able to dig deeper than you would if you just had a kind of a black box that gets delivered from some vendor, is often key to understanding the performance characteristics and bottlenecks you have in problems,” he said, “and therefore what kinds of algorithms and technical solutions will get you to that next level of performance.”
Millar said Liquid Computing has discussed similar partnerships with other universities but this is the first one the company has announced. Liquid also works with some commercial software vendors to benchmark their software on Liquid’s hardware, he said.
The Liquid architecture is well suited to database applications that require a lot of input/output, said Millar. Besides data mining, the company sees potential in the oil and gas industry.
“We’re looking for applications that rely on communications as much as they do on computation and memory,” said Millar. “That’s where we can really differentiate our product.”