Workload Optimized Systems - overview
As the growth rate of microprocessor frequencies is diminishing, use of parallel-processing and new architecture features, such as transactional memory and FPGA, is becoming essential for continuous performance improvements. The workload optimized system is an approach to co-optimize the whole system stack, which include compilers, language runtimes, and middleware, by exploiting such features for our target workloads ranging from commercial server applications, business analytics, and HPC applications. Using the experience and skills gained from the development of the world's fastest Java JIT compiler, IBM Research - Tokyo is advancing research and development for system software on workload optimized systems.
Competency fields
- High Performance Computing and Analytics Optimization technologies for applications and system software are crucial for supercomputers, such as IBM Blue Gene/Q Supercomputer (Sequoia), which achieved 20 petaflops of peak performance in 2012 at Lawrence Livermore National Laboratory. Our research topics include application-specific optimization technologies, especially for quantum chromodynamics, fast Fourier transforms, and computational fluid dynamics. We are also focusing on technology for scalable analytics framework, which include distributed language X10, and optimization technique for Jaql scripting language.
- Systems of Engagement We are seeing a historic change in IT systems from Systems of Record to Systems of Engagement. While we have been optimizing systems of record vertically from processors, OS, Java, Middleware, to applications, we must optimize systems of engagement, systems horizontally for networked middleware to efficiently and continuously establish more interactions and collaborations with users and partners. For example, in Smarter Commerce, our clients provide e-commerce sites on WebSphere Commerce with Unica to build rich client relationships, and with Sterling Commerce to manage their efficient supply chains. Our team is doing research on new methodologies to optimize workloads for these networked middleware in systems of engagement for better scalability, lower latency, higher throughputs, and smaller footprint.
- High Performance Commercial Systems We are pursuing performance acceleration for commercial systems by exploiting new hardware features, by developing advanced compilation techniques, and by analyzing the performance characteristics extensively. Those new hardware features include SIMD instructions, hardware transactional memory, performance monitoring hardware, and FPGA. We are using the experience and skills gained from the development of the world's fastest Java JIT compiler. In addition to traditional scale-up and scale-out techniques, we are investigating "scale-in" techniques where we integrate key components, such as storage, memory and processing, tightly in a single chase, closer to the data for performance acceleration.