Program

Overview Monday Tuesday
Convention Hall A, 2F ( Dec. 11 )
10:00 - 10:20 Opening
10:20 - 12:10 Session 1 (CREST/SPPEXA) Chair:Yaoko Nakagawa
10:20 - 10:50 "Software Technology that Deals with Deeper Memory Hierarchy in Post-petascale Era"
Toshio Endo
abstract
Abstract Coming soon.
10:50 - 11:20 "Power Management Framework for Post-Petascale Supercomputers"
Masaaki Kondo
abstract
Abstract Coming soon.
11:20 - 11:50 "Project CASSIA: Framework for Exhaustive and Large-scale Social Simulation"
Itsuki Noda
abstract
Abstract Coming soon.
11:50 - 12:10 "High Performance, High Productivity Programming through Domain Specialization"
Naoya Maruyama
abstract
Abstract This talk will present an overview of some of our past research for achieving high performance and high productivity at heterogeneous large-scale systems. It is generally considered that performance and productivity are conflicting trade-off in HPC applications, especially when using accelerators such as GPUs. To address this challenge, we have been working on domain-specific programming frameworks that significantly ease the problem of achieving both performance and productivity. In this talk, we will show that by designing frameworks customized for specific computation patterns it is possible to automate most of the manual programming burden for parallelization and optimization.
12:10 - 13:30 Lunch Break
13:30 - 14:30 Session 2 (Keynote) Chair:Mitsuhisa Sato
"Exascale Computing Project: Software Technology Perspective"
Rajeev Thakur
abstract
Abstract Coming soon.
14:30 - 15:30 Session 3 (Invited Talk 1) Chair:Hiroshi Nakashima
14:30 - 15:00 "Advanced Loop Transformation for Scalable Automatic GPU Mapping"
Tobias Grosser
abstract
Abstract Coming soon.
15:00 - 15:30 "PowerStack: Enabling Efficient Power Management through Hierarchical Design"
Tapasya Patki
abstract
Abstract Coming soon.
15:30 - 15:50 Break @ Hall B
15:50 - 16:50 Session 4 (Invited Talk 2) Chair:Hiroaki Kobayashi
15:50 - 16:20 "Social Physics: Data-Driven Discovery of Social Connectome"
Kimmo Kaski
abstract
Abstract While Information Communication Technology (ICT) has offered us new ways to communicate and socially interact, it leaves behind digital traces of our individual behaviour as records of ever-growing datasets. The study of such large-scale or Big Data using high-performance computational analysis and modeling with Network Theory approach can give us unprecedented insight into human sociality and to the structures and processes of social life and the society. This is well-demonstrated by our analysis of the dataset of mobile phone communication-logs, confirming the Granovetterian picture for the social network structure, i.e. being modular showing communities with strong internal ties and weaker external ties linking them. More recently the same dataset, but with additional data of the gender and age of the service subscribers, has allowed us to look at the nature of social interaction in more detail and from a different Dunbarian egocentric viewpoint. With this we have got a deeper insight into the gender and age-related social behaviour patterns and dynamics of close human relationships. Our analysis results demonstrate sex differences in the gender-bias of preferred relationships that reflect the way the reproductive investment strategies of both sexes change across their lifespan. We have also investigated the influence of seasonally and geographically related daily dynamics of daylight and ambient temperature on human resting patterns and observed two daily inactivity periods in the population-wide mobile phone calling patterns. The nocturnal resting period was found to be influenced by the length of daylight, and that its seasonal variation depends on the latitude of the phone users. In addition, the duration of the afternoon resting period was found influenced by the temperature, beyond certain threshold value, and that the yearly dynamics of the afternoon and nocturnal resting periods appear to be counterbalancing each other. These empirical findings inspired us to take the next step in network theory, namely developing models to catch some salient features of social networks and processes of human sociality. One of our first models, based on network sociology mechanisms for making friends, turned out to produce many empirically observed Granovetterian features of social networks, like meso-scale community and macro-scale topology formation. The modeling has subsequently been extended to take into account social networks being layered, multiplexing or context based, geography dependent, and having relationships between people changing in time. To summarize we believe that Social Physics’ large-scale data-driven analytics and modelling approaches to social systems opens up an unprecedented perspective to gain understanding of human sociality from individual to societal level, due to availability of Big Data and ever-increasing power of High Performance Computing, which together with methodological and algorithmic development would eventually lead to tools of social and societal design.
16:20 - 16:50 "OpenARC: Extensible Compiler Framework for Directive-based, Efficient Heterogeneous Computing Study"
Seyong Lee
abstract
Abstract Directive-based, accelerator programming models such as OpenACC have arisen as an alternative solution to program emerging Scalable Heterogeneous Computing (SHC) platforms. However, the increased complexity in the SHC systems incurs several challenges in terms of portability and productivity. OpenARC is an open-sourced, very High-Level Intermediate Representation (HLIR)-based, extensible compiler, which serves as an extensible research framework to address these issues in the directive-based accelerator programming. OpenARC is the first OpenACC compiler supporting Altera FPGAs, in addition to NVIDIA GPUs, AMD GPUs, and Intel Xeon Phis. OpenARC’s high-level representations (HLIR) allows to generate human-readable output code (either CUDA or OpenCL), which can be viewed and modified further by programmers if necessary. OpenARC offers device-aware OpenACC extensions, with which users can express architecture-specific features at high-level to achieve performance portability across diverse architectures. Several on-going research, where various performance optimizations, traceability mechanisms, fault tolerance techniques, etc., are developed for better performance/debuggability/resilience, demonstrates the efficacy of OpenARC as a research framework for directive-based, high-level programming study on the complex accelerator computing.
16:50 - 17:50 Panel Discussion Moderator: Mitsuhisa Sato
16:50 - 17:50 "International Collaboration and Competition for Exascale Computing and Beyond"

Panelists: Rajeev Thakur, Tobias Grosser, Achim Basermann, Itsuki Noda, Taisuke Boku, Masaaki Kondo
17:50 - 18:00  
18:00 - 19:30 Poster Session & Reception @ Hall B