Zero to infinity...

But, in practice, if "Connected..." > deadline, you will only get work if idle cpu, so for WCG any setting > 10 days will mean zero cached work...

The highly recommended, and the BOINC-default is, if you've got a permanent connection, "Computer is connected to the Internet about every N days" should be set to zero. (BOINC Manager has a little different wording...) Use "Additional..." to set desired cache-size.

For users with non-permanent connections, the recommended is to set "Connected..." >= expected disconnected-interwall, since "Connected..." is used in cpu-scheduling-decisions. Also, it will stop client downloading work with too short deadline to be returned due to the disconnect-interwall.


"Connected..." is used when it comes to:
1: Cache-size, when cached work drops below the sum of "Connected..." + "Additional..." more work is asked.
2: When needs to run "high priority". If at all possible, BOINC-client will try to finish all tasks atleast "Connected..." before the deadline.
3: Any "ready to report"-task is reported if less than "Connected..." to deadline.

There are some differences between BOINC-versions on the behaviour, these are:
a: BOINC-version v5.10.13 and earlier, any "ready to report"-task is reported if been "ready to report" for "Connected...". (some users uses this with example 0.01-days to report results nearly immediately after finished, but the config-option report_results_immediately was introduced to handle this).
b: v6.6.xx and later clients, always try to keep "Connected..." work on client, even if this means asking for work from low-resource-share projects with large negative LTD.

If you're not running multi-project, I'm not aware of any significant differences in how "Connected..." behaves on any of the v5.10.14 - v6.10.xx-clients.

twylith,

That config option was specifically (re)activated for those that for instance re-image machines daily i.e. if they shut down they get wiped. Do not use if you don't re-image. Hammers the servers. (Think a longer while back the same discussion about multiple RtR's took place with you).

Similarly, there is an abort on exit, so results that were not finished are signaled back to the server as never getting completed so they can be immediately reissued rather than sitting there to expire.

WCG sets the preset profiles for Connect to 0.1 and cache to 0.2, which results in a theoretical total work in progress (remaining time) + cached work of around 0.3 days i.e. 7.2 hours.

What the bottom value is when it looses meaning. Well, the actual result files are uploaded anyhow asap when there is a connection regardless of this setting. It just controls the reporting to the scheduler, phase 2 of the cycle, with about 7-10 rules attached to this how soon, such as new work fetch and the 24 hours completed rules. See FAQ's for this.

RtR is Ready to Report, which are the lines you see building in the Tasks view. There are the 7-10 rules depending on the BOINC version. By itself RtR's are not immediately reported simply because there is no need and at that improves efficiency at the server side.

As for the default, just checked in a test profile and it's 0.2 days connect and 0.3, so 0.5 total.

I just wonder if this lowest value causes the core client to do allot more, taking cycles. Certainly, if always connected is on, it "connect every..." looses its value and just becomes a part of the total buffer equation, but Spock logic does not necessarily apply in Berkeley.

Again, the Result files you see in the transfer window are not held. They are send up ASAP. It's the RtR's that will collect so the server scheduler does not have to do all the whirring of updating the DB and setting the validators in motion on each completion. Latter are paused at times anyhow, to allow other processes to catch up... imagine 425,000 separate reporting contacts... one every 0.2 seconds. Imagine a project that has few minute tasks facing immediately clearing, so thanks for keeping that option off.

To make it simple, for WCG-only members with a permanent connection to the network the "Connect" parameter has no influence on when and how often the BOINC client will contact the servers.
The result files for a job completing will be packed and uploaded immediately after.
And the reporting of jobs "ready to report" will be done the next time the client will need to ask for more jobs.
The client will not report jobs "ready to report" without requesting new work unless one of these jobs is close to its deadline or if there has not been any need for such contacts during 24 hours.

So, for such members, to avoid undesirable effects of high values of this parameter on job scheduling and jobs delivery by the server, simply set the "Connect" parameter to a low value (0 is fine) and forget it.

A setting of example 0.1 have little or no practical effects, if you're not either running multiple projects, or is running an old BOINC-client (example v5.10.13).

A setting of example 9.7 on the other hand will have significant effects, since all WCG-tasks will run "High priority", and all RtR's will be reported immediately, due to there being less than 9.7 days until the deadline...

As for "cache size", this is the sum of "Connected..." and "Additional...".

So, if example one of them is 1 day and the other is 4 days, the cache-size is 5 days.