Regarding server download connection timeout times (and subsequent transient HTTP errors ensuing). I counted THREE minutes last night (making it REALLY hard to retry MCM downloads as only "abort transfer" available on BOINC 7.20.2 - never clicked this as assume it aborts the workunit) and have observed at least 12-17 seconds today (typical).

I was thinking last night this would make it really hard to download with "retry now" button ARP workunits. Anyone from Krembil reading please look into. No idea what the optimal timeout time should be but it still seems too long. It was about 4-5 seconds from memory. That's when you get the transient HTTP errors. And I was thinking about discontinuing ARP as it's taking me about an hour of "retry now" to download couple of workunits. I was going to bring another computer online soon but I can't afford to spend another hour on the ARP per day and even more downloading MCM workunits with "retry now" (at least another hour per day).

Ideally there would be NO transient errors. I'm assuming it's beyond possibility to tweak the servers so this doesn't happen and is some hardware limitation. If not maybe it can be reduced/optimised? May be some trial and error involved and I'm assuming Krembil has at least one computer running WCG to observe the effects of changes.

Anyone else notice this? Or maybe this is Krembil trying out some configuration changes. Anyone got more info or further details please post back! Anyone from Krembil please take note.