SC13 Denver, CO

The International Conference for High Performance Computing, Networking, Storage and Analysis

Efficient Data Compression of Time Series of Particles' Positions for High-Throughput Animated Visualization.

Authors: Katsumi Hagita (National Defense Academy), Takaaki Takeda (VASA Entertainment, Inc), Tsunehiko Kato (Hiroshima University), Hiroaki Ohtani (National Institute for Fusion Science), Seiji Ishiguro (National Institute for Fusion Science)

Abstract: We tried to improve data-reading throughput by data compression on animated visualization of time series data generated from simulations of huge particle systems. It is considered that data compression is expected to reduce problems related to relatively slow reading from storage device. In the present paper, we examined efficient data compression schemes for huge time series data of plasma particle simulation. For example, to read positions of 300,000 particles with double precision text, required Read Throughput is almost equal to that of recent SSD. Thus, we considered data compression enables visualization of larger system. For smooth animations, compression of trajectories in time order is effective, because correlation of positions of a certain particle in time orders is higher than spatial correlation. Especially, lossy compression with polynomial functions in time order shows good performances. We called this scheme as TOKI (Time-Order, Kinetic, and Irreversible) compression.

Poster: pdf
Two-page extended abstract: pdf

Poster Index