Input Projection Algorithms Influence in Prediction and Optimization of QoS Accuracy

Abstract

Regardless of new achievements in the research of prediction models, QoS is still a great issue for high quality web services and remains one of the key subjects that need to be studied. We believe that QoS should not only be measured, but have to be predicted in development and implementation phases. In this paper we assess how different input projection algorithms influence the prediction accuracy of a Multi-Layer Perceptron (MLP) trained with large datasets of web services QoS values.

Influence of the QoS Measures for VoIP Traffic in a Congested Network

“…Unclassified traffic (which does not fall within any user-defined class) is sent to a special class (default) using WFQ mechanism and the weights defined by it according to the equation (2).…”

“…Each manually configured class represents a separate queue managed by a FIFO algorithm that receives a percentage of the total bandwidth depending on how the packets are chosen for transmission to the hardware buffer. After configuration, the CBWFQ algorithm works similar to WFQ, which means that it is based on the SN value defined by the same formula as in WFQ, equation (2). It can be demonstrated that any user-defined class dominates the dynamically determined flows (WFQ) almost all the time (except when certain classes are configured with a very low percentage 1.5%, which is unlikely) [13].…”

“…The QoS problem is a subject that will be studied continuously, according to other papers which studied this matter [1], [2]. There are a few challenges that concern the VoIP implementation: finding solutions to transmit a real-time service using a data network, which offers no guarantees regarding the delivery of the service, by using the best effort mode.…”