Aaron Beitch and Brandon Liu and Timothy Yung and Rean Griffith and Armando Fox and David A. Patterson

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2010-14

February 10, 2010

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2010/EECS-2010-14.pdf

Existing workload generation tools are not flexible enough to generate workloads that vary in three key dimensions relevant for making resource allocation decisions for Cloud Computing applications – variations in the amount of load, variations in the mix of operations performed by clients (e.g. changes in reads vs. writes or customer usage-patterns) and variations in the popularity of the data accessed, i.e., data hotspots. In this paper we present Rain, a workload generation toolkit designed to address this lack of flexibility. Rain allows for the use of probability distributions as first class elements that describe variations in the three key dimensions of workload Its architecture supports multiple workload generation strategies (open-loop, closed loop and partly-open loop), easy extensibility via user-defined request generators targeting new systems/applications – we demonstrate targeting the Olio Web 2.0 application. Rain also supports trace generation, which allows it to be used in conjunction with high-performance load-replay systems, e.g., httperf.

---

BibTeX citation:

@techreport{Beitch:EECS-2010-14,
    Author= {Beitch, Aaron and Liu, Brandon and Yung, Timothy and Griffith, Rean and Fox, Armando and Patterson, David A.},
    Title= {Rain: A Workload Generation Toolkit for Cloud Computing Applications},
    Year= {2010},
    Month= {Feb},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2010/EECS-2010-14.html},
    Number= {UCB/EECS-2010-14},
    Abstract= {Existing workload generation tools are not flexible
enough to generate workloads that vary in three key dimensions relevant for making resource allocation decisions
for Cloud Computing applications – variations in
the amount of load, variations in the mix of operations
performed by clients (e.g. changes in reads vs. writes or
customer usage-patterns) and variations in the popularity
of the data accessed, i.e., data hotspots.
In this paper we present Rain, a workload generation
toolkit designed to address this lack of flexibility. Rain
allows for the use of probability distributions as first class elements that describe variations in the three key dimensions of workload Its architecture supports multiple
workload generation strategies (open-loop, closed loop
and partly-open loop), easy extensibility via user-defined
request generators targeting new systems/applications –
we demonstrate targeting the Olio Web 2.0 application.
Rain also supports trace generation, which allows it to be
used in conjunction with high-performance load-replay
systems, e.g., httperf.},
}

EndNote citation:

%0 Report
%A Beitch, Aaron 
%A Liu, Brandon 
%A Yung, Timothy 
%A Griffith, Rean 
%A Fox, Armando 
%A Patterson, David A. 
%T Rain: A Workload Generation Toolkit for Cloud Computing Applications
%I EECS Department, University of California, Berkeley
%D 2010
%8 February 10
%@ UCB/EECS-2010-14
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2010/EECS-2010-14.html
%F Beitch:EECS-2010-14