BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:2.0
BEGIN:VEVENT
DTSTART:20151117T163000Z
DTEND:20151117T170000Z
LOCATION:18CD
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Modern cosmology and plasma-physics codes are now capable of=0Asimulating trillions of particles on petascale-systems. Each timestep=0Aoutput from such simulations is on the order of 10s of TBs. Summarizing=0Aand analyzing raw particle data is challenging, and scientists=0Aoften focus on density structures, whether in the real 3D=0Aspace, or a high-dimensional phase space. In this work, we develop=0Aa highly scalable version of the clustering algorithm DBSCAN, and=0Aapply it to the largest datasets produced by state-of-the-art codes.=0AOur system, called BD-CATS, is the first one capable of performing=0Aend-to-end analysis at trillion particle scale. We show analysis of 1.4=0Atrillion particles from a plasma-physics simulation, and a 10,240^3=0Aparticle cosmological simulation, utilizing ~100,000 cores in 30=0Aminutes. BD-CATS is helping infer mechanisms behind particle=0Aacceleration in plasma-physics and holds promise for qualitatively=0Asuperior clustering in cosmology. Both of these results were previously=0Aintractable at the trillion-particle scale.
SUMMARY:BD-CATS: Big Data Clustering at Trillion Particle Scale
PRIORITY:3
END:VEVENT
END:VCALENDAR