Speaker 1 : Amna Abdul Wahid
Title : Optimising the Sampling Rate when Measuring Packet-level Performance in Networks
Optimal sampling of a network for measurement of parameters like mean delay and loss probability is an active area of research. The current industry standard for sampling the network is ‘uniform’ sampling – perhaps every second or every minute. Academic research has been exploring alternatives to this, e.g. packet train models and provably optimal patterns of sampling that are built on a mathematical model. Sometimes these approaches suggest a deterministic (fixed) sampling rate, while other studies conclude that non-deterministic sampling is optimal. My research targets the packet loss probability, building on work by Parker et al, and Roughan. My aim is to identify the nature of the relationship between error in the measured PLP and the sampling interval for a range of likely traffic metrics, e.g. load, buffer length, traffic burstiness, then to relate this to practical networking examples such as to yield industry standard guidelines.
Speaker 2 : Keith Jones
Title : WRED in a multiclass environment
Tradition in service provider networks dictates that VoIP traffic and video traffic should be mapped to a policed EF class. Users suffer poor voice quality when it is policed. Moving VoIP traffic from the EF class to a lower QoS class and applying WRED improves user QoE.
Date: 23rd October, 2013.
Time: 14.00-15.00 hrs
Venue: QMUL Maths:1.03