In this project we study the application of the restart method to systems with service-oriented architecture (SOA). We consider services composed from several independent services, where all access to a service happens over the public Internet and service-internals are hidden from the user. In this scenario, service response-times and service-availability suffer from faults and failures both in the Internet and on the machines hosting the services. The client applies restart, i.e. resubmits service-invokations that do not finish in time, in order to reduce service response times and improve service-availability. We develop a test-bed for multi-level fault-injection in SOA systems and derive stochastic fault-models from measurements. We also quantify the adaptivity of restart algorithms, that is, their ability to adapt to system and network conditions.
In this DFG-funded project we investigate the use and impact of the restart method in distributed environments. In these environments several independent clients contend for the same resource (e.g. a Web Server). Clients may or may not employ restart, i.e. repeat their requests, restart strategies may differ, and the number of clients may potentially be huge. Clients do not communicate directly, thus interacting only through access to the same shared resource. A major question in this scenario is the fairness of restart algorithms, i.e. whether clients will attain a steady state where no client suffers starvation and where the server is not overloaded. Through analysis of stochastic models we aim to confirm the fairness property of the restart algorithm by investigating the timing behaviour of a simulated network in which several clients employ restart. In this project we cooperate with Prof. Leila Kloul from PRISM (France).
Packet-switched networks pose new challenges for clock synchronisation in cellular telephony infrastructure. Clock synchronisation requires strict bounds on the packet-delay variation (PDV) of the synchronisation stream. Packet-switching introduces complex timing behaviour, thereby affecting synchronisation accuracy. In fact, in some packet-switched networks clock synchronisation may be impossible. In cooperation with highstreet technologieswe analyse the timing behaviour of complex packet-switched networks through detailed simulation using the network simulator ns-2. We provide guidelines for suitable network setups and work on improving synchronisation protocols.
System evaluation through experimentation in test-beds, simulation, and analysis requires fault-models that capture faults well and enable efficient reproduction of faults. In cooperation with Prof. Miklos Telek and Prof. Gabor Horvath of the Technical University of Budapest we develop techniques for efficent and accurate stochastic fault-models. In particular, we focus on random-number generation from Phase-type and Matrix-Exponential distributions. Our results increase simulation speed and help to avoid artifacts in fault-injection experiments.
In this project we study Quality of Service in IEEE 802.11e Wireless LANS. We develop methods for enhancing QoS, using e.g. priority manipulations, and integrate them into a model-driven QoS management framework.
If you are interested in doing your thesis with us, you may come to our Monday meetings (2:15pm room 137).
We offer a variety of topics for Bachelor, Master, and Diploma theses, e.g.
Your thesis should be written using LaTeX type-setting (latex-template). You should present your work at least twice at a Wednesday meeting and pre-submit your thesis for proof-reading 2-3 weeks before your deadline - or earlier.
We sometimes use our Wiki page.
If you are interested in working on dependable systems, do not hesitate to contact us.