article propagation delay
Pavel V. Knyazev
pasha at surnet.ru
Tue Mar 12 12:46:53 UTC 2002
Imagine the following ...
I have a lot of incoming peers, some of them are ordinary
usenet sites, some of them are the largest sites (that have
more than 100 incoming peers). Naturally the latters offer
a lot of articles earlier than ordinary sites, because they are
closer to the core of usenet. BUT! As larger the site is -
- under heavier load it works. What it means? It can OFFER
articles EARLIER than other sites, but actually SEND the articles'
bodies LATER, than, for example, other sites can do (because they
are operating under easier load).
This is an example (ms are milliseconds):
0 ms: peer asked me if i had a particular message (check)
30 ms: i answered i do not (238)
after this moment i refuse all articles with the same id (438)
... a BIG DELAY happens ... (i guess this peer is overloaded)
3000 ms: peer gave me the body (takethis)
3030 ms: i got the message (239)
It seems ok at a first look. But, this message is doomed to NOT be
propagated to my outgoing peers because they could already receive
this article from another faster peers.
How could i to catch up this slow-but-fast peer, that gives me
a lot of articles (because it offers articles earlier than others)
but hinders me propagating articles (because it sends the body
later than the others could do).
I can't leave wipcheck and wipexpire set to 0.
This will make a lot of dups.
What do you think about this?
More information about the inn-workers