- All Implemented Interfaces:
SolverStrategy
public class Pcm2RegExStrategy
extends Object
implements SolverStrategy
This is an excerpt of Heiko's dissertation (see below for link)
The Stochastic Regular Expression (SRE) model is an analytical performance
model in the class of semi-Markov processes [Tri01]. It consists of a
discrete time Markov-chain (DTMC) to model state transitions, but the sojourn
time in each state can follow arbitrary probability distributions instead of
being limited to exponential distributions as in Markov chains. Furthermore,
SREs are hierarchically structured and do not allow cycles in the embedded
DTMC for more accurate predictions. Chapter 6.3.3 will provide the syntax and
semantics of SREs, afterwards Chapter 6.3.4 shows how to compute overall
sojourn times with SREs. Only a partial transformation of PCM instances to
SREs is possible, because of the model's limited expressiveness. The
transformation is straight-forward, as the control flow modelling of PCM
instances and SREs are closely aligned. Chapter 6.3.5 will describe the
transformation PCM2SRE. While allowing accurate predictions by supporting
arbitrary distribution functions for timing values, SRE are limited to
analysing single-user scenarios. They do not include queues or control flow
forks, and cannot express contention effects due to concurrent requests.
However, they provide a fast method of producing performance predictions
during early development stages, as they are usually more quickly solved than
running a simulation. Chapter 6.3.6 discusses the assumptions underlying SREs
in detail. The SRE model will be used for a performance prediction in a case
study in Chapter 7.3.3.