========================================================
E2emonit: available bandwidth
EGEE JRA4 Face to Face meeting, Pisa, 24-25 October 2005
========================================================

Present
-------

UEDIN: Alistair Phipps (AKP), Andy Jackson (AJ)
CNRS:  Sophie Nicoud (SN), Benjamin Vial (BV)
DFN:   Robert Stoy (RS)

Actions arising
---------------

Discussion
----------

Iperf wrapper - measures available bandwidth using UDP mode.

Wrapper requires as input the approximate speed of the slower interface at the
end-hosts.  Where packet loss over a certain threshold is encountered, the speed
is halved, otherwise the speed is set to 1.5x the previous value.  Two break
conditions: 1) no. of tests, 2) packet loss percentage small

RS says that it does not work in some cases, but he does not know why - needs to
determine this.

RS says that fluctuations in available bandwidth generally occur quite slowly,
so this will not affect the measurement.

In case of inherent loss, packets discarded independent of rate or arbitrarily,
then the available bandwidth cannot be measured by this method.

The packet loss output from these runs is not useful as an indication of loss on
the link - loss is expected at high loads.  Instead, only the pinger output
should be used (ICMP packet loss).

For jitter, the average jitter from all measurements (at different bandwidths)
should be used, as jitter should be independent of utilisation.

For out of order packets, RS suggests to use the value when the available
bandwidth is used.

AKP 31/10/2005