Takeshi Ogasawara  Takeshi Ogasawara photo       

contact information

IBM Research - Tokyo, Japan


Professional Associations

Professional Associations:  ACM  |  ACM SIGPLAN

more information

More information:  International  |  In Japanese


I am interested in performance optimization of programs by compilers, their runtime systems, and specialized hardware accelerators. My recent interests is acceleration of genome analysis pipelines. The following includes brief descriptions of my research so far.


Workload characterization of server-side frameworks for script languages

As modern frameworks for easily building target programs in script languages are widely used,  development of server-side code in script languages has been popular. Node.js is a popular framework for  JavaScript code. The performance of JavaScript was significantly improved by dynamic compilers to optimize complex applications on Web browsers. Node.js was built on one of major JavaScript compilers, V8.

Is V8 optimizing Node.js applications efficiently as it optimizes Web applications? Through detailed workload characterization of various Node.js applications, we found that performance improvements by compiler optimization are limited because the library code of Node.js and V8 written in C++ largely consumed the CPU time while V8 canot optimize C++ code.


Scalability of system software

The processors based on the non-uniform-memory-access (NUMA) architecture are becoming popular on the server side. A NUMA-aware memory management is critical for high performance of Java programs in which many threads are running across multiple CPU chips. I proposed a technique that can find the CPU chips for objects with low overhead, based on a widely-used generational GC.

When I worked on the project that investigated bottlenecks in the software stack (particularly, our middleware and Java virtual machine) with many cores, I researched the end-to-end problems, including threading, lock contention, and garbage collection.


Java just-in-time compiler

In the past I worked on the project, the Java just-in-time compiler. I researched the dynamic optimization of exception handling, the memory management of the JIT, lock optimization, and optimized code generation for x86 platforms.

High Performance Fortran compiler

When I worked on the high performance fortran compiler project, I studied the use of collective communication for HPF.