Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
The Case for Malleable Stream Architectures Christopher Batten 1, Hidetaka Aoki 2, Krste Asanovic 3
 

Summary: The Case for Malleable Stream Architectures
Christopher Batten 1, Hidetaka Aoki 2, Krste Asanovi´c 3
1 Massachusetts Institute of Technology, Cambridge, MA, USA ­ cbatten@mit.edu
2 Hitachi, Ltd., Tokyo, Japan ­ hidetaka.aoki.rt@hitachi.com
3 University of California, Berkeley, CA, USA ­ krste@eecs.berkeley.edu
Stream workloads vary widely, as do proposed stream ar-
chitectures. We argue that stream processors should prior-
itize efficient temporal and spatial data-parallel execution,
while not ignoring support for temporal and spatial kernel-
parallel execution. We introduce a new malleable stream
architecture with data- and kernel-parallel mechanisms that
can be reconfigured as needed by stream applications.
1. The Landscape of Stream Workloads
Many applications can be expressed as streams of ele-
ments flowing between computational kernels. Although
stream programs are more structured than general-purpose
programs, they include surprisingly diverse forms of paral-
lelism and communication patterns. For example, consider
the toy application in Figure 1. Kernels A­C exhibit data-
level parallelism (DLP) meaning the kernel execution for one

  

Source: Asanović, Krste - Computer Science and Artificial Intelligence Laboratory & Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT)

 

Collections: Computer Technologies and Information Sciences