Platform's Grid Software Moves Processes Closer to Data Sources
Platform Computing has added data access, scheduling and scaling features to version 5 of its Symphony grid computing software, which is coming out today. (For those unfamiliar with grid computing, grid software takes a group of servers, virtualizes them, spreads the processing of work across them and allows you to treat them as one cohesive whole.) Users of Symphony include Citigroup Capital Markets, Societe Generale and Sal. Oppenheim. Wall Street clients typically use the product for compute-intensive applications such as risk processing, indexing, pricing and analytics, or to share resources among lines of business. "Our clients reap benefit from this because they drive utilization levels way up; they don't need to find more hardware to get the results they need and they're able to utilize compute capacity that exists elsewhere in the enterprise," says Jim Mancuso, general manager, financial services.
Most of the development effort that went into this new release focused on improving data access by placing workloads on systems that are physically close to the required data, to decease bottlenecks and improve performance. This could be particularly helpful in data-heavy applications such as pre-trade risk analysis. The software has also been modified to help applications run on multi-core processors, even if they haven't been written in parallel.
Purists say it's impossible to fully realize the benefits of four- or six-core processors without writing or rewriting a program to run in parallel. The chief architect of a major Wall Street firm recently said that while putting intermediary grid or virtualization management layers in between non-parallel applications and new multi-core chips may help distribute certain tasks such as calculating algorithms, it isn't a complete answer and memory management problems can arise. "If you do a little, you get a little," he said.
"That's accurate, that's an intelligent statement," Mancuso concedes. "What you put into it is what you get out of it. Taking advantage of new hardware and horsepower is a big issue in the industry. Not all clients want to do this work. Some clients are focused on taking advantage of multithreading and such and for others, it doesn't have as much meaning in their business right now."
Version 5 has two add-on products. Data Affinity allows the software to schedule application tasks while taking into account data locality, to place workloads in a more cost efficient way. Dynamic Service is designed to drive up utilization within a single blade and reduce memory contention for data I/O intensive applications by sending only one set of data, enabling the software to scale in mixed environments with old (two-core) and new (multi-core) hardware.
Tweet This