As you will have read, am keeping a detail log of tasks done with nativeboinc. On the now 123 results, two had this. Twelve went out with signal 11.

To continue the experiment, since the last of these signal fails, have reduced the 'while processor usage is than' to 20 percent down from 30, a value less than 100 percent of 1 core on a 4 core. Though the os is supposed to spread load over all cores, suspecting some latency still causes a core to get saturated and tasks at times to choke. The agent probably has a sampling rate slow enough to not detect the high load timely, so with 20 percent hoping to allow a better pre-empting.

At the same time increased the 'use at most nn percent of processors' from 75 to 100 percent. Surprisingly, when the 193's occurred running 4 cores, they had not come back going to 3, just the signal 11 persisted, 11 percent of results, but 5 percent of total accumulated runtime. Still not good enough for me. 12 results off 4 cores completed now and no issue yet, but the efficiency did drop from an average 96 to 93-94. Can live with that if it stops multiple hours loss to errors each pop.

Remains the other major irritation, the 'too late' conversions with 7 percent of runtime suffering this. Work units running fine, but somehow quorum not achieved, each non-failing wingman apparently just differently enough to not pass, or in a few cases, not a single wingman succeeding.

Preposterous, but with just one tablet, my rac ranking is 33 for android according nativeboinc stats out of 2518 members who've chosen to not hide their host statistics details, meaning the device stats can be connected to a member.

Detail tablet stats: http://i1383.photobucket.com/albums/ah289/lavaflow1/NativeBOINCResults.png

A device needs to run 60 days to get any semblance of correct rac. At project overall level it's thoroughly wrong though, since it seems to be the sum of individual device racs rather than a weighted average. With 1 device no probs as in my case, with multiple it's already pronouncedly off, until all run 60 days constant. Nativeboinc produces the exact same racs as boincstats i.e. they work it the same way.

It is looking like many days of NO invalid errors :-)
after a long time with many errors :-(
Running 2 core and 50% memory.

Could this be the light at the end of the tunnel of NO
errors, or is that light I see a train? Only time will tell.

Setting the cpu time to 90 percent means stop / start / stop / start on and on and on. Can't see that as conducive to stable processing for this feeble app.

An interesting stat on the nativeboinc run so far: Over 90 percent of received tasks ends up in the repair cycle. To be exact, have logged a distribution of 1 or 2 for 12 out of 126 results. Those in repair have an average distribution of 7.4 copies to include 8 that were removed from circulation altogether, no quorum possible even after 10-12 copies. Bluntly, about 90 percent of the devices are just not fit for crunching purpose with the present application. With hubble you could not see success being written as a promise in the far distance.

The whole issue is wearing me out, deeply frustrating. Going to toss this nonsense very soon and pull out. Wcg has become a serious issue grid.

Hmm, that's really a good question. It is interesting that the application runs stable and without errors if the parameters are set to 100% cores and 100% utilization of the cores. My dedicated boinc androids (odroids and minix) running fine with that.
My s2 i picked up from the drawer runs fine with that setting too. But in the last days it consumed more energy than the psu delivered. I checked all other android settings and disabled a lot of things yesterday. Now I'm back with 100% settings. Overnight I observered no energy drain.

But, and this is the issue, it disqualifies the application for all smartphone/tablett users where boinc is allowed to use the resources only if the device is not in use or boinc is configured to not interfere with the normal usage of the device .

Edit: Made clarification in the last sentence in red color. Thanks to lavaflow for pointing me to this.