Combining Parallel and Sequential Workloads on a Network of Workstations

Remzi H. Arpaci, Amin M. Vahdat, Thomas Anderson and David A. Patterson

EECS Department
University of California, Berkeley
Technical Report No. UCB/CSD-94-838
October 1994

http://www2.eecs.berkeley.edu/Pubs/TechRpts/1994/CSD-94-838.pdf

This paper examines the plausibility of using a network of workstations (NOW) for a mixture of parallel and sequential jobs. Through trace-driven simulation, our study identifies a number of results that should be of interest to NOW system designers. First, it is not sufficient to use workstation resources to provide a supercomputer only by night. Next, parallel programs can cause a significant number of lengthy delays to interactive users. Finally, simple scheduling techniques can identify available workstations and minimize user delays while providing parallel program performance comparable to a dedicated massively parallel processor. If these scheduling policies are employed, parallel programmers and interactive users can peacefully coexist on a NOW.


BibTeX citation:

@techreport{Arpaci:CSD-94-838,
    Author = {Arpaci, Remzi H. and Vahdat, Amin M. and Anderson, Thomas and Patterson, David A.},
    Title = {Combining Parallel and Sequential Workloads on a Network of Workstations},
    Institution = {EECS Department, University of California, Berkeley},
    Year = {1994},
    Month = {Oct},
    URL = {http://www2.eecs.berkeley.edu/Pubs/TechRpts/1994/5748.html},
    Number = {UCB/CSD-94-838},
    Abstract = {This paper examines the plausibility of using a network of workstations (NOW) for a mixture of parallel and sequential jobs. Through trace-driven simulation, our study identifies a number of results that should be of interest to NOW system designers. First, it is not sufficient to use workstation resources to provide a supercomputer only by night. Next, parallel programs can cause a significant number of lengthy delays to interactive users. Finally, simple scheduling techniques can identify available workstations and minimize user delays while providing parallel program performance comparable to a dedicated massively parallel processor. If these scheduling policies are employed, parallel programmers and interactive users can peacefully coexist on a NOW.}
}

EndNote citation:

%0 Report
%A Arpaci, Remzi H.
%A Vahdat, Amin M.
%A Anderson, Thomas
%A Patterson, David A.
%T Combining Parallel and Sequential Workloads on a Network of Workstations
%I EECS Department, University of California, Berkeley
%D 1994
%@ UCB/CSD-94-838
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/1994/5748.html
%F Arpaci:CSD-94-838