Thu, 13 Apr 1995 00:34:44 +0200
|
On Apr 10 (this Monday), SUNET's backup server was reconfigured to run a
maximum of two parallel backups over our main ethernet, rather than as
many as the ethernet would take. I believe this to have solved the
duplicate problem by preventing the ethernet from getting saturated and
triggering a link flap which in turn may trigger duplicate copies,
normally with an incidence of 1 in 10,000 (naturally if the duplicated
message goes to 200 users, 200 extra copies are generated). At any rate,
there has been no link flap between SEARN and SEGATE since Apr 9 at
around 9pm local. If there are still duplicates, it means there is a
second problem.
If any of you have received duplicates bearing a "Date:" field that says
Apr 10 or later (Swedish time in case your mail system adjusts for local
time), please notify me *in private*. Again, the date that matters is the
date the message was distributed to the list, and not the date you
received it. Finally, if you were thinking of flaming me or pointing out
that messages shouldn't be duplicated when a connection goes down, well,
I agree. The people who designed the protocols made a mistake a there is
a time window where the receiving site has accepted the message but the
sending site thinks the transfer was aborted because the connection was
lost. I did not design these protocols, and I did my best to address the
problem as soon as it was reported.
Eric
|
|
|