Internation Telecommunication Union
Les attended the ITU meeting in Orlando Florida and made an invited presentation on Internet QoS Measurements. He was
also a member of a panel on QoS. The ITU is particularly interested in Internet telephony,
so prior to the meeting, we
captured and analyzed Internet telephony packet streams and read up on the
protocols. We also studied what performance metrics are critical to Internet
telephony and what reasonable thresholds values are, and started to define and make
some measurements on jitter.
Provided assistance to Colorado State and ESnet to understand the performance between
Colorado State and SLAC in light of some chages in peering.
Les & Warren attended a voice conference of the XIWT. One topic was running
UDP-echo servers at XIWT sites to compare with ping echo. After some
discussion with security folks SLAC decided to look for another mechanism
to compare with ping-echo. The goal is to see whether ping-echos are being
deprecated (e.g. by traffic shaping of ICMP).
Our poster paper submission to the organizers of the INET'99
conference to be held in
San Jose in June 1999, was accepted.
Les & Warren met with Chris Jones of CERN/Roche and two others from Roche
pharmeceuticals to discuss details of Internet performance and to provide
help with PingER.
Warren completed the work on remaking the PingER code. It calculates values for
the 5 metrics (loss, response, quiescence, unreachability and
unpredictability) fo 100 and 1000 byte packets for all nodes and all sites
(nodes at the same site are aggregated in the latter case). The results are
also split up into daily reports with hourly ticks, monthly reports with
daily ticks and summary reports with monthly ticks. He has also added a new
metric that we are calling the "jitter". It is the IQR of the RTT of the 10
Warren updated ping.table.pl to allow selection of all
the new options. The user can now select the metric, the monitoring site group, the remote site
group, the packet size, the tick type, whether to do by site or by node, to exclude groups and
to show beacon sites only. Occasionally it times out before all the page is completed.
Warren has been monitoring the icmp-seq order of the packets in our ping sample to find the magnitude
of out of order packets. This can have a significant effect on Internet telephony. So far it
appears roughly 1 in 10,000 samples has an out of order packet. This small effect should have no
noticeable effect on Internet telephony or video conferencing. See
http://www.slac.stanford.edu/comp/net/wan-mon/pathology.html for more information.
In order to try and understand the impact of setting IP precedence bits we are placing a
monitoring machine behind a router running a new release of the Cisco IOS that can set precedence
Back to Top
Revised 23 December 1998