PhD Alumni


Anselmi Jonatha

Present position: Post-doctoral fellow at INRIA and LIG Laboratory
http://home.dei.polimi.it/anselmi/
 

Thesis title:  New Analyses in BCMP Queueing Networks Theory
Advisor:  Paolo Cremonesi
Research area:  Systems Architectures
Thesis abstract:  
BCMP queueing networks are a class of Markov processes which during the last decades have been widely used in the performance analysis of computer and communication systems in order to approach capacity planning problems. Even though the literature is plenty of excellent research works, there are still some difficulties limiting their applicability to real-world scenarios. The main difficulty relies on the computational complexity given by their exact solution. In fact, this paved the way to a number of alternative analyses aimed to efficiently understand their performance behavior. These include approximate, bounding, asymptotic and bottleneck analyses. In the present dissertation, we provide new theoretical results in each of the previous alternative analyses.
The scientific contributions of this thesis are: i) new bounds on the system throughput and response time of closed, single-class networks with load-dependent stations; ii) an asymptotic equivalence among closed, open and mixed multiclass networks which holds under the assumption that the service demands of a given station, for sufficiently large population sizes, dominate the ones of all the other stations; iii) a new framework supporting the bottleneck analysis of closed, multiclass networks with large population sizes which is able to handle load-dependent stations; iv) an inequality which efficiently bounds from above the textit{partition function} of closed, multiclass networks; v) a non-iterative approximate method for the solution of closed multiclass networks containing finite capacity regions and shared constraints. It is proven theoretically or it is shown numerically that the proposed analyses either improve the accuracy or reduce the computational complexity of existing approaches.

Curriculum: