Network Monitoring: Application Level
As discussed in the PIPVIC deliverable "Collection and Analysis of Mbone
Traffic Statistics", there are three possible places at which it is
possible to take loss measurements: application level, IP level and link
level. This page discusses application level monitoring, using RTCP
The audio and video tools used in the PIPVIC project are based around the
IETF standard real-time transport protocol (RTP), and the associated RTP
control protocol (RTCP). The control protocol, RTCP, is used to convey
information about each participant to all other participants: it is, for
example, the means by which the personnal names displayed by the media
tools are distributed.
Each application periodically multicasts an RTCP reception report to the
group. This reception report includes, for each active sender, summary
statistics for packet loss and jitter since the last report. By plotting
the information contained in these reception reports, it is possible to
display an overview of the network performance during a conference.
The graphs below show the reception quality observed, at UCL, during the
project meeting held on Friday 1st May 1998. There is one graph for each
participant sending video during the meeting. Within each graph, there is a
row for each receiver, with a coloured bar indicating the quality that
receiver was reporting. This bar is green if less than 5% packet loss was
observed, orange for 5-10% loss, and red when more than 10% loss occured.
Periods where no reception reports were received are white. This could mean
one of two things: either the source stopped transmitting, or the reception
report packets were lost.
Apologies for the size of the text in these graphs: if I scale them up big
enough the make the text easy to read, the files become huge and we run out
of disk space on our web server....
Readers may also be interested in the
Mbone loss measurements taken by
Mark Handley, which show a couple of different ways of measuring
- Paul Hogan appears multiple times. This is because his video tool
was restarted (the information in an RTCP RR packet is indexed by
SSRC value, and restarting a tool results in a new SSRC being chosen).
- The mrouter linking UCL to the rest of the world crashed a number of
times during the session. This is clearly visible, since no reception
reports were received from outside UCL during this period (ie: the big
white gaps on the right of the graphs).
- Reception quality was generally very good, with the exception of video
data coming from Dave Price (the second graph is mostly orange, which
indicates that all receivers where seeing between 5% and 10% loss from
him). The probably indicates a problem near to Aberystwyth.
- There is periodic loss on the link from Westminster. Notice the white
regions on the first graph, showing the reception of David Hearnshaw,
and the gaps in his reception reports for other people. All receivers
are showing the same pattern, indicating that the problem is near the
sender. In this case, it appears that packets are not getting out of
Westminster, so there is nothing for the other participants to report
- The reception reports from Angela Sasse show consistently poor quality
(all rows labelled with her name are mostly red). This occurs even for
those senders within UCL, on the same logical network segment as her.
I would guess this means that Angela's workstation is overloaded, and
is dropping packets from it's input queue, hence her video tools
to main PIPVIC page
Colin Perkins, 4 May 1998.