tons of rejects (duplicates) towards fast sites

Ilya Varlashkin ilya at samara.net
Tue Mar 2 13:23:28 UTC 2004


Hello All,

I've noticed that on the connections towards fast and high-volume and
reasonable high rated (on top1000) sites my server has a lots of rejects.
About 70-90% of transmitted traffic and the better remote server the
worse the situation becomes.

I've built a test lab and could reproduce this behaviour precisely. Suppose
feeder1 gets all external (real) feeds and feeder2 has feed only from feeder1.
Both feeder1 and feeder1 have also downstream towards leafsite. On the
connection between feeder2 and leafsite I see a lot of duplicates but,
funny enough, leafsite _did_ ask feeder2 to SENDIT (response code 238).
Why is it doing this, why am I not getting 

I tried to use 'drop deffered' on feeders, tried to use 'noresendid' on
leafsite, tried to encrease wip* parameters to half-minute. Nothing helps.

Is there anything to I can to avoid dupes without filtering traffic
received from fast sites towards other fast sites?

I searched google and found usenet statistic reports of many (very many)
sites having the same problem. This means there are mega-tons of garbage
floating around the world unwanted.

There was one thread on this list couple of years ago (for INN 2.2 iirc),
but it doesn't look like anything really changed.

Any help?

Kind regards,
Ilya Varlashkin



More information about the inn-workers mailing list