log in  |  register  |  feedback?  |  help  |  web accessibility
Logo
ArrayStore: A Storage Manager for Complex Parallel Array Processing
Presented By: Souvik Bhattacharjee - University of Maryland, College Park
3258 A.V. Williams Building (AVW)
Wednesday, October 3, 2012, 1:00-2:00 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Abstract
Authors: Emad Soroush, Magdalena Balazinska, Daniel Wang
 
 
Abstract: 
We present the design, implementation, and evaluation of ArrayStore, a new storage manager for complex, parallel array processing. ArrayStore builds on prior work in the area of multidimensional data storage, but considers the new problem of supporting a parallel and more varied workload comprising not only range-queries, but also binary operations such as joins and complex user-defined functions. This paper makes two key contributions. First, it examines several existing single-site storage management strategies and array partitioning strategies to identify which combination is best suited for the array-processing workload above. Second, it develops a new and effi cient storage management mechanism that enables parallel processing of operations that must access data from adjacent partitions. We evaluate ArrayStore on over 80GB of real data from two scientific domains and real operators used in these domains. We show that ArrayStore outperforms previously proposed storage management strategies in the context of its diverse target workload.
This talk is organized by Abdul Quamar