I have also noticed that it isnt connecting to report completed units unless it wants to grab more work... but at least i am able to walk by my computers every 3-5 hours and click the UPDATE button so it sends them...

if your computer needs more work when it connects it will also send the completed one but for some reason right now mine are not connecting to send complete work unless it also wants more work

normaly i try to make sure all my work units are up in time for the next "stats update" phase and so i routinely update them 1 hour before the update starts (for me that 5am and 5pm) cause the starts update starts at 6 my time

It can look like it's not reporting them, but they are reported, even if you're not asking for more work. Reporting of work happens:
1; Then client asks for more work.
2; If a result has been "ready to report" for 24 hours. (changed in v5.10.14).
3; If less than "Computer is connected to the Internet about every N days" until deadline.
4; If less than 24 hours until deadline.
5; To send trickle-up message (CPDN only).
6; Manually triggered by user hitting "update".
7; If user is using <report_results_immediately>. This option is not recommended, since adds unnessesary many extra connections.
8; If project is using <next_rpc_delay>, client will make a connection regardless of have anything to report or not. WCG is using 4 days here, while some other BOINC-projects has used settings as short as 1 hour.
9; Not so common, but a connection to Account Manager can in some instances trigger connection to multiple projects.

#2 was in v5.10.13 and earlier: If a task has been "ready to report" for more than "Computer is connected to the Internet about every N days".


In most instances, #1 will be in effect, but incorrect estimates can screw-up this, especially for users that "needs" a large cache.
Even if project set to "no new work", #2, #3 and #4 makes sure all work is reported before the deadline, as long as it's finished before the deadline that is, and connection isn't being blocked for some reason.

You're not the 1st. one that's made the wrong assumption with "connect every N days", especially if you've used older clients like v5.10.13 and earlier that did report results based on this setting. It's not helping that WCG isn't showing the full text for the preferences either:
"Computer is connected to the Internet about every N days" (Leave blank or 0 if always connected. BOINC will try to maintain at least this much work.)

As for why many users is suddenly seeing this, for one thing, 1-hour tasks compared to more normal 6 hour tasks is much easier to spot, since you'll have so many of them in a short time, especially on quad and 8-way systems.

Still, there wouldn't have been such a large effect, if the initial HCMD2-estimates had more to do with reality. The initial estimate was... 7 hours or something, while computer used maybe only 1 hour. After a string of these wu's, the BOINC-client had corrected for this, and expected the work takes 1 hour. And, WCG suddently changed the estimates, so non-corrected it would take 1 hour... For someone with the already corrected, the esimates suddenly showed maybe 10 minute/wu, so client in a short time asked for much more work, to fill-up enough.

Finish a single of these new shorter-estimated wu's, and suddenly the 10-minute estimate is corrected back to 1 hour, and all cached work immediately is estimated to take 6x longer...

Meaning, if you had 1 day cached, with the new estimates it grew to 6 days cached. And, since 6 days > 1 day, there's currently no reason for client to ask for more work.


With your 0.3 days cache-setting, the effects is smaller, but still the esimated run-time of work suddenly increased to maybe 2 days or something, and for this reason you've got a ton of work to run-through before drops below 0.3 days again.

< silver (after update)

22 pages of PVs