BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:1.0
BEGIN:VEVENT
DTSTART:20141119T200000Z
DTEND:20141119T203000Z
LOCATION:391-92
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: Glasswing is a MapReduce framework that uses OpenCL to exploit multi-core CPUs and accelerators. However, compute device capabilities vary significantly and require targeted optimization. Similarly, availability of memory, storage and interconnects impacts job performance. In this paper, we present and analyze how MapReduce applications can improve their horizontal and vertical scalability using a well controlled mixture of coarse- and fine-grained parallelism. We discuss the Glasswing pipeline and its ability to overlap computation, communication, memory transfers and disk access. We show how Glasswing adapts to the distinct capabilities of a variety of compute devices by employing fine-grained parallelism. We experimentally evaluated the performance of five applications and show that Glasswing outperforms Hadoop on a 64-node multi-core CPU cluster by factors between 1.2 and 4, and factors from 20 to 30 on a 23-node GPU cluster. Similarly, we show that Glasswing is at least 1.5x faster than GPMR on the GPU cluster.
SUMMARY:Scaling MapReduce Vertically and Horizontally
PRIORITY:3
END:VEVENT
END:VCALENDAR