Akihiro Nomura
Associate Professor for Institute Management
Global Scientific Information and Computing Center
Tokyo Institute of Technology
I was Ph.D student in Parallel and Distributed System Software Laboratory (Yutaka Ishikawa lab) in The University of Tokyo
Japanese Page
Note: My given name is "Akihiro" and my family name is "Nomura."
Work
Efficient Scheduling of Supercomputer Jobs
Supercomputers are shared with many users, and arbitration of their job to the limited resource is not a trivial problem.
To minimize waste of machine time, we need to pack jobs both spatially (node partitioning) and temporally (exact estimation of execution time).
High performance collective operation
Non-blocking collective communications are going to be introduced in new MPI standards.
Normally, communication threads are necessary to implement Non-blocking collectives in order to make them progress even if application thread does not return to MPI library, but the costs of context switch between computation threads and communication threads are not negligible.
I propose a method to pivot communication threads' functionality into OS kernel context in order to avoid context switches.
Modeling of collective communications in commodity network environment
Recently, commodity network environment, especially TCP/IP on Ethernet, is becoming used in cloud computing environment.
Unlike in the interconnects which is commonly used in cluster computing, like Infiniband, the behavior of communication in TCP/IP network is complicated because of various reasons.
I believe that building the method to predict the behavior of communication will be a stepping stone to solving various problems on the performance of cloud computing.
Publications
Please check recent publications from T2R2.
High performance collective operation
- Akihiro Nomura, Yutaka Ishikawa, Naoya Maruyama, Satoshi Matsuoka
"Design and Implementation of Portable and Efficient Non-blocking collective Communication"
IEEE/ACM CCGrid 2012, May 2012
Paper
Slides
- Akihiro Nomura, Yutaka Ishikawa
"Study in Implementing Asynchronous MPI Collective Communications" (in Japanese)
IPSJ SIG Technical Reports (HPC-130, SWoPP 2011) Jul 2010
- Akihiro Nomura, Yutaka Ishikawa
"Design of Kernel-level Asynchronous Collective Communication"
EuroMPI 2010 pp.92-101, Sep 2010
Paper
Slides
- Akihiro Nomura, Yutaka Ishikawa
"Design and Implemenation of Kernel-level Asynchronous MPI Collective Communications" (in Japanese)
IPSJ SIG Technical Reports (HPC-126, SWoPP 2010) Aug 2010
- Akihiro Nomura, Hiroya Matsuba, Yutaka Ishikawa
"High-Performance Non-blocking Collective Communications in MPI" (Poster)
Supercomputing 2009 (SC|09), November 2009
Modeling of collective communications in commodity network environment
- Akihiro Nomura, Hiroya Matsuba, Yutaka Ishikawa
"Network Performance Model for TCP/IP Based Cluster Computing"
Cluster 2007 pp.194-203, September 2007
- Akihiro Nomura, Yutaka Ishikawa
"Design of Communication Model Considering Retransmission Control" (in Japanese)
IPSJ SIG Technical Reports (HPC-107, SWoPP 2006) pp. 175-180, July 2006
OS-level support for realtime communication
- Akihiro Nomura, Yutaka Ishikawa
"Communication Control of Real-time Packets" (in Japanese)
IPSJ SIG Technical Reports (OS-100, SWoPP 2005) pp. 79-84, August 2005
- Akihiro Nomura, Yutaka Ishikawa
"Transmission Control on Real-time Packets" (in Japanese)
IPSJ SIG Technical Reports (OS-99) pp. 161-162, May 2005
Contact Information
Akihiro Nomura
(Sorry, this is a picture to avoid spams.)