Well, we're back in the regime of work supply exceeding transport (download) capacity that prompted this thread in the first place, but it's not as severe as it was then. Once again, Linux clients seem to be better at hanging on to a connection once it's been established, than Windows clients.

Man, this is so annoying... now I remember part of the reason I quit WCG years ago. And now I'm kinda regretting coming back...

At any rate, the solution for now is to brute force and keep retrying the transfers until they go through. To help automate that process, here's a super dumb batch file that will automatically retry all transfers once every 12 seconds (for about 3 hours):



Save that as a .bat and run it. And good luck.

The download issues are pathetic, I have only a few hours typically when electricity is cheap, so I only want to run GPU jobs that finish quickly and result in many points per WU. I would start thinking about reworking the project to only require one download per WU and unzip the individual files on the clients, this would lower the number of requests to your servers that are clearly not up to the task.

The NEVR ENDING DOWNLOAD PROBLEMS are a real pain in the rear end.
Krembil does not seem to think it is a big issue.
Lack of REGULAR COMMUNICATION with users is another area Krembil is lacking in.
I could add more but that is enough for now.

But quite frankly, your timing sucks! Opening up the flood gates and then just call it a day and show up the next morning to see all the angry and frustrated responses certainly doesn't help to either fix the issue nor to preserve Krembil's credibility, as little as there is left in the DC community...

Ralf