I'm new to this project, I've now processed about a dozen Binary Radio Pulsar Search (Arecibo) v1.51 () arm-android-linux-gnu tasks on my phone and every single one has errored after 41,417 secs of run time, which seems a bit of a coincident. Do these tasks work is there any point in carrying on processing any more.
Copyright © 2024 Einstein@Home. All rights reserved.
pascali wrote:... every
)
Hi pascali, Welcome to Einstein.
If you look at your tasks list for that device on the website, there are actually 3 that have been completed and 2 of those are validated. So, these tasks do indeed work OK.
The identical processing time for the others is a sign that they reached the allowed time limit and BOINC terminated crunching in the belief that the tasks might be stuck. I have no experience with devices like phones so can't comment on what is causing the slow crunching. The ones that did finish were still very slow but significantly faster than the others that hit the time limit.
Usually, the time limit is several times longer than what the device should take. Maybe the Einstein servers have rated the device to be faster than it really is. Another possibility is that the device is throttling itself due to excessive heat. Maybe you could try keeping the device as cool as possible to see if that allows it to run faster. Crunching is compute intensive and phones aren't really designed with scientific number crunching in mind :-).
When problems like this occur, just click on the Task ID link for a failed task and you can see what caused the failure. For this task you can see the line:
Exit status: 197 (0x000000C5) EXIT_TIME_LIMIT_EXCEEDED
near the top of the page. If you scroll down a bit you will find a heading "Stderr output" which is a complete log of what happened during crunching. I chose this one at random and it shows that crunching was stopped and restarted several times before the time limit was reached. This is by design since state is saved at regular intervals (in things called checkpoints) which allows restarting from the last saved checkpoint with minimal loss of progress.
So the only issue is the device being a lot slower than the project thinks it should be.
Cheers,
Gary.
Gary Roberts wrote:So the
)
Rerun the CPU benchmarks, the measured integer speed on that device is nearly 3 times of that of your Core i5, which is likely completely wrong and the reason for all those errors. You are not the first with that issue.
.
Yes, the Boinc client
)
Yes, the Boinc client benchmarks are ludicrous with Android devices. Best solution is to remove the benchmark values from the client_state.xml file and set
<skip_cpu_benchmarks>1</skip_cpu_benchmarks>
in the cc_config.xml so that benchmarks are never run and the client defaults to the base benchmark numbers.
Then you won't get the exceeded time limit errors.