Hello,

Thank you for the quick answer. Unfortunately I'm still lost a little bit:

If I understand correctly you're saying that I should make it so that the image from which the Google Cloud creates new instances (the "master"/"gold" image) does not contain BOINC set up, but rather I should link BOINC with WCG the first time such an instances starts up. Also, that I should change "report_results_immediately" to 1.

These are great suggestions and I shall implement them promptly. In fact this may explain the fact that I'm seeing less results from these machines than I was expecting.

You are also correct that these instances "remain intact" after being stopped and they could be re-used as per the documentation (https://cloud.google.com/compute/docs/instanc...ible#preemption_process):

"""
Preempted instances still appear in your project, but you are not charged for the instance hours while it remains in a TERMINATED state. You can access and recover data from any persistent disks that are attached to the instance, but those disks still incur storage charges until you delete them.
"""

The trick is that I'm using Preemtible instances together with Managed Instance Groups. This acts like a watchdog / scheduler - ie. I set the number of instances I want and if instances die, it tries to start new ones to take its place. I don't think that the scheduler re-uses TERMINATED instances, but rather starts fresh ones (although I will try to clarify this with Google Support). So re-using older instances is not a very viable option.

Do you have any idea:

- if having many devices associated with an account would be a problem for the WCG database? (just a quick calculation: if I run 100 instances for one month, restarting every 24 hours, that would be roughly 3000 devices)
- is there a way to permanently remove devices from the WCG database? or perhaps to "merge" multiple devices into one (even if it's after the fact - for example I could try to periodically merge old instances into one).

Could you please point me to the "code to abort unfinished tasks on shutdown and force a communication to tell the project these need immediate reassignment"?

Thank you,
Attila

Thank you again for all your help and advice. I've been crunching now for almost 30 days.

I do have a follow up question though if you have the time: 3 days ago I increased my cluster size (because I received confirmation that we can dedicate more resources to it). And I see corresponding increase in my "Results returned" graph, however no increase in the "Total Run Time" or "Points generated" graph.

Any idea why that might be? Based on the BOINC logs those instances do seem to finish / upload WUs, so I'm not sure what's going on...

Thanks for the info. On viewMemberStatHistory.do I see:

10/21/2017 0:001:08:19:36 5,884 29
10/20/2017 1:041:13:00:25 1,903,436 9,174
10/19/2017 1:018:20:46:03 1,786,841 9,018
10/18/2017 1:044:02:53:36 1,892,786 9,244
10/17/2017 1:014:04:20:42 1,925,279 6,626
10/16/2017 1:012:07:59:34 2,005,855 7,179
10/15/2017 1:015:00:33:44 1,968,455 6,743
10/14/2017 1:020:07:57:36 1,958,492 6,670
10/13/2017 1:016:20:17:13 1,939,777 6,712

As you can see the "no. results returned" jumps by ~3k on the 18th (due to the increase in cluster size), however "points generated" / "total runtime" stay the same. The quorum explanation for MCM seems like a very plausible explanation. I also checked and I have ~400 WUs "pending verification". So, perhaps once those are verified, the statistics will be updated.

Cheers.