[OLSR-users] Routes momentarily disappearing

Thomas Lopatic (spam-protected)
Fri Dec 30 11:38:16 CET 2005

Hi Karl,

You might want to try setting the HelloValidityTime parameter to a
higher than default value. In your case, with links there are two
mechanisms at work - (1) link detection and (2) link quality. The link
quality case is what you care for with the large window size. However,
in your configuration file you do not change the link detection
parameters. By default, every two seconds a node broadcasts a HELLO
message to announce its existence to its neighbours. The corresponding
configuration line would be "HelloInterval 2.0". In its HELLO messages
it also says "if you do not hear a HELLO message for six seconds,
consider me dead". The corresponding configuration line would be
"HelloValidityTime 6.0". So, if three HELLOs in a row are lost, then the
link is considered down, irrespective of the link quality window size.

So, what this all boils down to is that we recommend setting the
"HelloValidityTime" to "HelloInterval x LinkQualityWinSize", i.e. to use
"HelloValidityTime 120.0" on *all* nodes in your mesh. (Remember that
setting this value on node X influences the behaviour of X's neighbours,
i.e. if you set the validity time to 120 seconds on node X, all of X's
neighbours will consider the link to X existent as long as there's at
least one HELLO from X every 120 seconds.)

In this way, we guarantee that link detection will not remove a link
before link quality has had a chance to let it's link quality decay to
the worst possible case, which is that LinkQualityWinSize x
HelloInterval packets have been lost. So, basically, what we do is, we
switch off link detection and leave it all to link quality.

So, I'd recommend adding "HelloValidityTime 120.0" to your interface

BTW, "HelloInterval x LinkQualityWinSize" is the default setting for
HelloValidityTime in 0.4.10, which is used if the user doesn't specify
anything else. So for 0.4.10 your configuration will work as is.


More information about the Olsr-users mailing list