I can see that the download problems are a real issue for fast machines and short WUs, but since ARP WUs take me 14 hours each and I can only run 2 at a time. I would really rather have to deal with download issues a couple of times (minor really, as it isn't as bad as it used to be), than not have anything to chew on.

ARP has the end of its project in sight if they can keep the WUs flowing. It is good for ARP and once ARP is done then that issue goes away, at least until the next big download project comes along.

Well, I got not even two dozens of them and I have MASSIVE download problems, though mostly this shows with OPNG and OPN1 WUs...

Too bad it is still just crickets from WCG....


Ralf

Good question.

For most people, ARP and MCM have been running dry or almost dry on their machines. For MCM1 this means that the one large file needs to be downloaded again. For ARP1 this means that for each task a number of large files need to be downloaded. It takes some time to download big files from the server; the more machines are busy downloading, the more server connections are busy and taken. With the number of server connections growing and growing, the download speed is dropping and the number of free server connections (there is a limit!) is also becoming less. So, when too many devices are asking for work, their connection to the server is refused. For the machines that are downloading, their download speed will be low or lower than what is believed to be normal.

So, are they related? It depends, roughly speaking, on the amount of work being released and the download speed. With high download speed, the number of busy server connections will be low. With low download speed, the number of busy server connections will be high.

Adri

I don't think that anyone suggested that this is a client side issue, but rather a server side issue. That when the server is busy with several large downloads (like those 102MB MCM1 files, or ARP1 multiple +10MB files) AND at the same time has to serve a lot of connection requesting the tiny OPN1/OPNG files, that this could be a possible source for the overall download issues.

Ralf

ARP1 work units are taking about 9 hours each on my machine. I have it set up to do only 4 WCG work units at a time. Sometimes I run out of patience and push the downloads by hitting Retry button, but overnight, mostly they come through all right. But only mostly.

I allow 4 Rosetta tasks to run at a time too.

I also allow 4 ClimatePrediction tasks to run at a time, but that makes no difference since I have received no work units from them since late July.