Losing WorkUnits, part 2
Anonymous
20 Apr 2005 17:37:49 UTC
Topic 12927
(moderation:
We're trying to track this down, just added some debug output to the scheduler that will hopefully reveal something - we just didn't get hold of the bug yet.
BM
Losing WorkUnits, part 2
)
David Anderson and I have put this onto our 'to-do' list, but it will take some time to do. Our plan is to modify the client and scheduler so that the client reports the WU that it currently has to the scheduler. The scheduler will check this against the list of 'in progress' WU for that host. If there are some WU 'in progress' which are absent from the host, then they will be re-sent to the host.
It may take some time for us to implement this, as it requires both client- and server-side changes. But it should make the sending of WU to a host be 'reliable' because there will be a handshake for it.
NOTE ADDED MAY 13, 2005
If you are using a proxy server or have your machine networked using some other machine as a gateway or proxy, be sure to (1) set the proxy timeout to a good long value (several minutes) and (2) use a recent version of BOINC. In particular, BOINC version 4.19 had problems with some proxy servers. These problems can give rise to the phantom/lost WU.
Cheers,
Bruce