|
By Tom Groenfeldt
Firms that should be updating and recompiling their software are buying faster hardware to improve performance, according to Ambreesh Khanna, Sun Microsystems' group director of global financial services.
"I say most of the bottlenecks are still in the software, not in the hardware," said Khanna. "Everyone buys Nehalem for the 10 to 20 percent boost. They are popular because of poorly written applications--that is where the problem lies."
He recently visited a large Wall Street firm that was moving to Intel's Nehalem processor from AMD. The investment bank had achieved a 30 percent performance boost with Intel.
Khanna learned that they had achieved a 10 percent increase in speed-huge, in his opinion-by recompiling their code. But the developers worried that it might not be bug-free, so they returned to the old version and were depending on new hardware to attain the desired speed.
"They would rather spend a couple million to buy new hardware for the same sort of boost," he said. ISVs that had expected faster and faster CPUs weren't prepared for more cores and multi-threading, he added. Sun has a CPU that runs 64 threads at once, while a Nehalem processor with six to eight cores can run 12 to 16 threads at one time.
"The ISVs should be looking at this and asking how to get the applications to take advantage of more threads; instead they keep adding new functionality." So in finance they wind up with applications that use just one or two cores for the application and another for messaging. Or, as one high-tech software marketing executive said, companies call an application multi-threaded when it uses one thread for the application and a second for the GUI.
Khanna said that developers should be working on applications that can provide linear scalability for today's high-density environments, which can provide 3,000 cores in a single data center container. Ideally, they should be thinking about scalability that can take advantage of cloud computing with 10,000 or more cores.
Capital markets firms, usually a leader in adopting new technology, have not taken up the challenge, he added. Sun has developed a programming language which is multi-threaded out of the box.
"All I need is for developers to get interested in it," he said. "When I speak to programmers they are stuck in a C++ and Java world, but none of these languages can scale to 10,000 CPUs."
Sun offers a parallel programming language called Fortress, along with guides and a community web site.
It can be used for high-performance computing and less demanding applications that require parallel processing, reports HPCwire:
"And even though Fortress specifically targets high-end technical computing, it is also applicable to large-scale parallel applications of almost any type. 'We were looking for a language that was good for multicore, for supercomputing, and for everything in between,' explains Eric Allen, principal investigator of the programming languages research group at Sun Labs."
Continues HPCwire:
"The project began in 2003 and was originally funded out of DARPA's High Productivity Computing System (HPCS) program. When Sun was dropped from HPCS in Phase III of the program, Sun Labs took over the Fortress R&D completely. But since Sun has made Fortress an open source project, the company has received a lot of outside help from universities and other researchers that have contributed to the design and implementation of the language. .."
Intel is also providing help to developers in building applications for multiple threads, although perhaps not to the scale-like 10,000 cores in the cloud-that Khanna is talking about. The company runs periodic Web training for developers and offers a variety of aids.
It has also worked with leading software vendors, such as SunGard Data Systems, to help them take advantage of multi-core processors. A year ago Intel trained developers of SunGard's BancWare asset-liability management software, which is used to manage the balance sheets at some of the world's largest banks. Banks want to run 10,000 or 20,000 simulations across millions of positions. SunGard managed to use 512 cores last year and was able to run 20,000 full balance sheet simulations in just under five hours for one of the world largest multinational banks. Further improvements in performance are underway.
Only registered users can write comments.
Please login or register. |
