Re: Processed: destruction of round-robin functionality is fucking up our mirrors and making Debian suck for many people, hence fixing this is a release-critical "wish"

[Ian Jackson <ian@davenant.greenend.org.uk>]

Anthony Towns writes ("Re: Processed: destruction of round-robin functionality is fucking 	up our mirrors and making Debian suck for many people, hence fixing 	this is a release-critical "wish""):
> [stuff]

I've been skimreading this subthread and I have to say I'm not quite
clear about what question you are trying to answer and why.  I'll take
a guess but please do feel free to correct me if I'm wrong.

We know that the load on various of the Debian servers is very
asymmetrical and that this is (to say the least) unhelpful.

>From the mail exchanges it would appear that the discussion here is an
attempt to predict the visible effects of rule 9 behavour for the
relevant set of IP addresses.  Presumably this is intended to try to
understand whether the imbalance is caused by rule 9.

If I understand correctly, I think this isn't a good approach.

The behaviour of a rule 9 host depends on its view of its local IP
address.  We have no data on what proportion of the relevant host
populations have which addresses.

On the other hand, the behaviour of a round robin honouring host
depends on the frequency of DNS retries, past network topology
history, etc., in a way that may be difficult to predict.  (Note that
some webcache configurations send pings to determine which of a set of
possible upstreams are `closest', so we aren't just considering rule 9
clients and R-R clients.)

Noone has presented any other explanation besides rule 9 for the very
substantial traffic asymmetry.  We do have anecdotal evidence that it
used to be much better.

And even if this particular asymmetry isn't due to this particular
problem (or to look at it another way even if there are other things
which would also need fixing) rule 9 is still unjustifiable.

Ian.