On Tuesday 17 June 2003 08:50, Andrew Cooks wrote:
Hi
How long is the timeout on the new release?
About three and a half minutes.
I used the new CD image
(I think it's 20030607, or 20030610) and I found that I have to restart
the operation if any of the receiver machines stop receiving. "Stop
receiving" includes reboots and "No space left on device" errors.
No, you don't need to restart; you just need to be patient ;) In other
words, if your transfer has already been going on for more than 3 1/2
minutes, you win time by just waiting...
Timeout is so long in order to guard against accidental dropping of
receivers in case of temporary network slowness. We once had a 3com
switch which would pause for 30 seconds every minute during a udpcast
transfer, and introducing really _long_ timeouts was the only way to
cope...
I know there was a timeout value that was sometimes
too short and caused
"Pipeline full" errors and consequent dropping of the clients, but I
haven't had a chance to look at the source again.
Actually, "Pipeline full" warnings are only marginally related to
sender->receiver timeouts. "Pipeline full" warnings are displayed by
the receiver when flow control kicks in (mostly, because data is
received faster from the network than can be written to the disk or to
a decompressor). The main motivation of these warnings was to do
performance analysis (to determine whether any sub-optimal performance
was due to inefficiencies in the network protocol, or to local
reasons). As this message was confusing to users, it is now disabled
by default (you should only get it if you compile udpcast with the
DEBUG flag).
Regards,
Alain