- All Implemented Interfaces:
SolverStrategy
public class Pcm2LqnStrategy
extends Object
implements SolverStrategy
This is an excerpt of Heiko's dissertation (see below for link)
The Layered Queueing Network (LQN) model is a performance model in the class
of extended queueing networks. It is a popular model with widespread use
[BDIS04]. Like the PCM, it specifically targets analysing the performance of
distributed systems. While ordinary queueing networks model software
structures only implicitly via resource demands to service centers, LQNs
model a system as a layered hierarchy of interacting software entities, which
produce demands for the underlying physical resources such as CPUs or hard
disks. Therefore, LQNs reflect the structure of distributed systems more
naturally than ordinary queueing networks. In particular, they model the
routing of jobs in the network more realistically.
In the context of this work, a model transformation from PCM instances (with
computed context models) to LQNs has been implemented. The transformation
offers at least two advantages: First, it enables comparing the concepts of
the PCM with concepts of LQNs, which can be considered as a state-of-the-art
performance model. Second, the transformation makes the sophisticated
analytical solvers and simulation tools for LQNs available to the PCM. Other
than SREs, LQNs support concurrent behaviour, different kinds of workloads,
asynchronous interactions, and different scheduling strategies. Therefore, it
is possible to derive performance metrics such as resource utilizations and
throughput from PCM instances, which is not possible with SREs. However, LQNs
are restricted to exponential distributions and mean-values analysis as
discussed later.
The chapter 6.4 in Heiko's dissertation will first provide some background
about LQNs and their development in recent years (Chapter 6.4.2). Then, it
will describe the syntax and (informal) semantics of LQNs using the LQN
meta-model and several examples (Chapter 6.4.3). Chapter 6.4.4 briefly
describes two performance solvers for LQNs, before Chapter 6.4.5 presents the
mapping from PCM instances to LQN instances. Finally, Chapter 6.4.6 compares
the PCM model with the LQN model, as well as the existing PCM solvers with
two available LQN solvers.