An inquiry into PBNM system performance required for massive scale telecommunication applications

Publication Type:
Issue Date:
Full metadata record
Files in This Item:
Filename Description SizeFormat
01Front.pdf219.92 kBAdobe PDF
02Whole.pdf5.17 MBAdobe PDF
PBNM systems have been proposed as a feasible techology for managing massive scale applications including telecommunication service management. What is not known is how this class of system performs under carrier-scale traffic loads. This research investigates this open question and concludes, subject to the considerations herein, this technology can provide services to large scale applications. An in depth examination of several inferencing algorithms is made using experimental methods. The inferencing operation has been implicated as the major source of performance problems in rule based systems and we examine this. Moreover, these algorithms are of central importance to current and future context-aware, pervasive, mobile services. A novel algorithm, JukeBox, is proposed that is a correct, general and pure bindspace conjunctive match algorithm. It is compared to the current state of the art algorithm - Rete. We find that Rete is the superior algorithm when implemented using the hashed-equality variant. We also conclude that IO is an important cause of PBNM system performace limitations and is perhaps of more significance than the implicated inferencing operations. However, inferencing can be a bottleneck to performance and we document the factors associated with this. We describe a generally useful policy system benchmarking procedure that provides a visible, repeatable and measurable process for establishing a policy server's service rate characteristics. The service rate statistics, namely (mu) and (sigma), establish the limitations to policy system throughput. Combined with the offered traffic load to the server, using the statistic (lambda), we can provide a complete characterisation of system performance using the Pollaczek-Khinchine function. This characterisation allows us to make simple design and dimensioning heuristics that can be used to rate the policy system as a whole.
Please use this identifier to cite or link to this item: