Monitoring for the Test VoIP Network

SLAC is participating (Map) in a trial of the Voice over IP (VoIP) application.

This report details the research conducted to determine how setting of type of service (TOS) bits in the IP header combined with weighted fair queuing (WFQ) in routers can provide a high quality VoIP conversation.

Characterization of Observed Jitter

The throughput, round trip time (RTT) and packet loss are all important factors, but primarily the important metric is jitter, where we define jitter to be the difference between arrival times for each packet.

We begin by plotting the jitter for 10,000 pings from a workstation at SLAC to a Workstation at LBNL along test link.

	This graph for two separate set of pings (10,000 and 7,848) taken about 10 days apart shows that most RTT are about 0.5ms longer than the previous, this is offset by a large number of RTT being around 2.2 ms less than the previous. This may seem suprising, if the jitter is randomly caused by noise on the network line is would seem plausible that the jitter is normally distributed around zero, or at least symmetric around zero.
	This graph attempts to shed some light on the peculiar distribution observed previously. The bars show the minimum, median and maximum RTT in each minute for 7,848 consecutive pings (ie between packet loss). There appears to be a pattern, where for 20 minutes or so, the minimum and maximum constantly decrease before quickly increasing again.
	The Median RTT value for each minute appears to be close to a normal distribution.

Statistical Analysis of Ping Data

Revised 20 August 1999
URL: http://www-iepm.slac.stanford.edu/monitoring/voip/index.html
Comments to iepm-l@slac.stanford.edu