MGARD: A Multilevel Technique for Compression of
Floating-Point Data
Author/Presenters
Event Type
Workshop
TimeFriday, November 17th9:50am -
10:10am
Location302-303
DescriptionMGARD (MultiGrid Adaptive Reduction of Data) is a
recently developed technique for multilevel lossy
compression and reduction of scientific data. The
technique is based on the theory of multigrid methods,
which have found widespread application in the
computational solution of partial differential
equations. MGARD allows the user to specify either a
lossiness tolerance level or a size constraint and
produces a quasioptimal reduction. Moreover, the data is
stored in a hierarchical decomposition well-suited to
heterogeneous storage systems.
In this study, we outline algorithms implementing MGARD and perform computational experiments demonstrating its effectiveness. We apply MGARD and state of the art compression
tools to two datasets, one generated to prescribed smoothness and another obtained from power grid monitoring micro-phasor measurement units installed at Lawrence Berkeley National Laboratory. The results indicate that in its current preliminary state MGARD offers a competitive alternative to existing state of the art compression tools.
In this study, we outline algorithms implementing MGARD and perform computational experiments demonstrating its effectiveness. We apply MGARD and state of the art compression
tools to two datasets, one generated to prescribed smoothness and another obtained from power grid monitoring micro-phasor measurement units installed at Lawrence Berkeley National Laboratory. The results indicate that in its current preliminary state MGARD offers a competitive alternative to existing state of the art compression tools.
