Hi.

I had yesterday a strange crash in my BOINC 6.10.36.
It was jammed on GPUGRID task uploading and suddenly
the was a huge amount of tasks in my GPUGRID account's
task list. None of these task are seen by BOINC manager. I will downgrade to BOINC 6.10.18.
(Platform i7 / Windows 7 Ultimate / GeForce GTS240)

Here are the ghost tasks:

Task ID Work unit ID
2040467 1285089
2038771 1274213
2038769 1281015
2038743 1284145
2038671 1282857
2038664 1284085
2038610 1284041
2038595 1281767
2038575 1283266
2038574 1282928
2038536 1283984
2038485 1283862
2038461 1283935
2038409 1283845
2038366 1283758
2038346 1277802
2038296 1282110
2038280 1282511
2038082 1280593
2037991 1277349
2037950 1273218
2037939 1283570
2022829 1274230

BR Pentti

It did it again, now 10 ghost tasks.

Sorry, 6.10.56 did not help. Got 16 ghosts again today.
(Platform Intel I7 / GTS240 / Win 7 64-bit Ultimate).

So guys, now I ask your advice. Because these ghost tasks
can never be seen on Boinc manager, I don't have
possibility to cancel them, when this kind of ghost
task burst occurs. There are two possibilities.
Either to accept, that some tasks are delayed by
waiting timeout (no response), or switch my GPU
crunching other more well-behaving projects.
Which alternative you think it is better to do?

/Pentti

Your computers are hidden, that makes it really hard (impossible?) for us to figure out what may be giving you problems.

Thanks. Now I see: Detach clears the buffer on server side.

BTW: I have unhided my computers now.

I do not use proxy. But yes, I have some kind
of communication problem with Boinc.
Boinc manager sometimes says that it does not
have network connection, although same time
I can access Internet and email server. The
only way to get Boinc communicate again is the reboot.
Could it be something to do with strict firewall rules?

Hi.
I'll detach and keep some days pause on GPUGRID to
see, if the commmunication problems are tied to GPUGRID only.
See U...

My problem seems to be solved now.
My fault. And yes, firewall was the
reason, but not the firewall I thought.

I got a new ADSL modem, which had a
second firewall I didn't think at all.
It was not preconfigured into bridged mode as I thought,
but into NAT mode. And that caused my problems.

Thanks everybody.

/Pentti

I've seen a BOINC project which handles ghost workunits by detecting them somehow, and then giving the client computer those same workunits again the next time it asks for any workunits before the deadline. Could GPUGRID implement something like that?

One way to do it would to provide small workunits designed to do little more than return a report on which other workunits the client machine knows it has, and give only the new small workunits to machines that are otherwise at their limit for workunits per day or limit for workunits currently in progress,
possibly with a limit of just one of these new small workunits until either a ghost workunit is detected or some other workunit is returned.