Hello,

I noticed point diffrences. Some work units works on a single core for 24 hours and gets 60 points some others about 1000. As I see points are not calculated on time basis. Those results are also belongs to the same computer so its not about the hardware either.

---

Perhaps it could be diffraction spikes instead, except not any notion of time you could make it.

---

I think the question should be, "why do some tasks only get 60 credits"? And to answer that, it would be helpful if you could provide links to some.

---

Rosetta Moderator: Mod.Sense

I couldnt find such low points at my last wus but here is another sample. Similar calculations times with huge point differences.

https://boinc.bakerlab.org/rosetta/workunit.php?wuid=1003697077

https://boinc.bakerlab.org/rosetta/workunit.php?wuid=1003558915

---

The low credit task reports that it was shutdown by the "watchdog". Meaning it did not end normally, but did get some useful work done before that point, and so the watchdog cut off the work unit and reported the successful portion.

As search algorithms mature, such WU shutdowns become much less frequent. The way the credit system works, if the batch of work has caused the watchdog to kick in on several previous completed WUs, they actually start to pull more points per completed model. The fact that you both got that watchdog shutdown, and low credit would tend to mean that watchdog shutdowns are not very common in the results coming back for that batch of work.

The credit system is based on the average of the claimed credit, per completed model, of all of the other machines working on the same batch. The claimed credit is based on the BOINC benchmarks for the machine, which are largely based on raw CPU time (which is not a good predictor for all types of work, especially memory intensive work). So, in broad strokes, the credit system is based on actual completed work, and therefore is not easy to spoof. It is not looking at the CPU seconds directly, but that factored in to the claimed credit. One machine might get the same number of completed models done in an hour, that another takes 5 hours to complete, but the resulting work units will receive the same credit award for the work done, because they both completed the same number of models (decoys).

---

Rosetta Moderator: Mod.Sense

That's only part of the problem. The real problem is the code was programmed wrong and David Baker has known about the defective code for years. Baker should explain when he'll fix it. It's so cavalier to waste patrons money.

---

You say it's rare but almost all of my wus are like that,. should I fix smth with my boinc setup or smth? I need to browse 4 pages of results to find smth over 1000points. When I check details all of them says watchdog ended. Strangely all my computers are same.

---

The low credit task reports that it was shutdown by the "watchdog". Meaning it did not end normally, but did get some useful work done before that point, and so the watchdog cut off the work unit and reported the successful portion.

As search algorithms mature, such WU shutdowns become much less frequent. The way the credit system works, if the batch of work has caused the watchdog to kick in on several previous completed WUs, they actually start to pull more points per completed model. The fact that you both got that watchdog shutdown, and low credit would tend to mean that watchdog shutdowns are not very common in the results coming back for that batch of work.

The credit system is based on the average of the claimed credit, per completed model, of all of the other machines working on the same batch. The claimed credit is based on the BOINC benchmarks for the machine, which are largely based on raw CPU time (which is not a good predictor for all types of work, especially memory intensive work). So, in broad strokes, the credit system is based on actual completed work, and therefore is not easy to spoof. It is not looking at the CPU seconds directly, but that factored in to the claimed credit. One machine might get the same number of completed models done in an hour, that another takes 5 hours to complete, but the resulting work units will receive the same credit award for the work done, because they both completed the same number of models (decoys).

How long does the watchdog wait before shutting down the WU? I am assuming it shuts it down because there is some sort of compute error on that machine - either the WU itself failed, or some sort of hardware issue, such as excessive memory swap, did the same?
I come back to this project every so often and wonder how often my compute cycles sit in limbo - especially if I use the 24 hour workunits, which I have heard net the best results. It would also be even more of a shame to see those fail and waste a day of computing, essentially.
Hopefully I'm understanding this correctly.

I don’t see premature end when I check boinc. They start from 0 to 100 and takes about 24hours to complete. When the work unit is sent I get low points.

All my devices generates similar results.

1) imac
2) macbook pro
3) ryzen 9 3900x
4) 2 x intel workstations.

There are no similar hardware and they are in different locations too.

---