ECM Errors !!!

[technik007.cz]

I haven't posted link to any of workunits because all of ecm_nr fails across all of my about 50 devices same way. They run for about 30mins in average and then stop for some reason. Also stderr out is nearly the same.

Do you know what means parameter -maxmem 4000 ? What units does it use?
All of them were executed with this parameter, except other ones. But I don't have 4000MB of RAM on any of my devices.

[yoyo]

Without any link I can't tell.
Each batch of workunits has different memory requirements.

-maxmem 4000 means that in no case the workunit should use more than 4 GB. But this doesn't mean that the workunit needs so much. The real requirement is attached to the workunit and different for each batch. But without any link, I can't tell.

[technik007.cz]

I don't have my computers visible for others. Is there any other methods how-to find out?

BTW Desktop computers have usually large amount of RAM like 8GB and more plus large swap where little errors in available memory calculations can easily hide. But arm devices have mostly RAM sizes between 512MB-1GB with swap turned off because you don't want to kill your microSD card due to frequent writtes.

[yoyo]

Please sent me some links to such error workunits.

[technik007.cz]

https://www.rechenkraft.net/yoyo/workun ... d=78944958
https://www.rechenkraft.net/yoyo/result ... =111728771
https://www.rechenkraft.net/yoyo/result ... =111766364
https://www.rechenkraft.net/yoyo/result ... =111677207

[yoyo]

I'm looking into the 3 results.

https://www.rechenkraft.net/yoyo/result ... =111728771
This workunits needs 1,1 GB RAM and was running on a system which has 1,9 GB RAM, 0 GB swap and 6 cores

https://www.rechenkraft.net/yoyo/result ... =111766364
This wu needs 0,2 GB RAM and was running on a system which has 0,9 GB RAM, 0 GB swap and 4 cores

https://www.rechenkraft.net/yoyo/result ... =111677207
This wu needs 0,2 GB RAM and was running on a system which has 0,9 GB RAM, 16 GB swap and 4 cores

You might see the problem.
Your BOINC client downloads the result only if the needed RAM is free.
Boinc client also starts the result only if the needed RAM is free.
But it happens that BOINC starts many wus in parallel, because currently during start the RAM is available. If than these running workunits allocate the RAM, there is not enough available RAM for all. I don't know how often BOINC checks it and suspending the wu with keep in RAM doesn't free any RAM. At the end the kernel dosn't know what to do, there is also 0 swap. At the end the workunit crashes.

I can only say that for such systems with less RAM, there should be swap of 2 times RAM.

yoyo

[technik007.cz]

You were unable to obtain details of the WU for a device with 0.5GB RAM. I guess it also runs wus with a 0.2GB RAM requirement like these =>
https://www.rechenkraft.net/yoyo/result ... =111663745
https://www.rechenkraft.net/yoyo/result ... =111816779
https://www.rechenkraft.net/yoyo/result ... =111665291

However, as I mentioned, you cannot enable swap on these devices because it will negatively impact the lifetime of microSD cards. I have only a few with SSDs like that one with 16GB swap, but it failed here anyway.

What is the difference between ecm_cn and ecm_nr? Ecm_cn does not fail even on a device with 0.5GB of RAM see here =>
https://www.rechenkraft.net/yoyo/result ... =111681456
https://www.rechenkraft.net/yoyo/result ... =111681430
https://www.rechenkraft.net/yoyo/result ... =111681288
https://www.rechenkraft.net/yoyo/result ... =111681184

[zombie67]

https://www.rechenkraft.net/yoyo/result ... =111829019

[yoyo]

You were unable to obtain details of the WU for a device with 0.5GB RAM. I guess it also runs wus with a 0.2GB RAM requirement like these =>
https://www.rechenkraft.net/yoyo/result ... =111663745
https://www.rechenkraft.net/yoyo/result ... =111816779
https://www.rechenkraft.net/yoyo/result ... =111665291

The computer has 469 MB RAM and 50 MB swap. The results are tagged to use up to 200 MB. This doesn't mean that they need 200 MB. They might need much less.

However, as I mentioned, you cannot enable swap on these devices because it will negatively impact the lifetime of microSD cards. I have only a few with SSDs like that one with 16GB swap, but it failed here anyway.

I just can repeat here. Running a system without swap is like driving a car without airbag and without seat bells, especially if the system has so less memory per thread. If the system/car runs into edge cases you'll have problems.

What is the difference between ecm_cn and ecm_nr? Ecm_cn does not fail even on a device with 0.5GB of RAM see here =>
https://www.rechenkraft.net/yoyo/result ... =111681456
https://www.rechenkraft.net/yoyo/result ... =111681430
https://www.rechenkraft.net/yoyo/result ... =111681288
https://www.rechenkraft.net/yoyo/result ... =111681184

Those workunits are tagged to use up to 10 MB RAM.

The above mentiond workunits contains 5 individual runs. You can see this in the job.xml and that the progress indicator jumps in 20% steps.
Each run creates an output file where is stated how much memory was allocated.

yoyo

[yoyo]

https://www.rechenkraft.net/yoyo/result ... =111829019

You see in the result
Process creation (../../projects/www.rechenkraft.net_yoyo/ecmwrapper_705.02_x86_64-apple-darwin) failed: Bad CPU type in executable (errno = -1)

[zombie67]

https://www.rechenkraft.net/yoyo/result ... =111829019

You see in the result
Process creation (../../projects/www.rechenkraft.net_yoyo/ecmwrapper_705.02_x86_64-apple-darwin) failed: Bad CPU type in executable (errno = -1)

This happens with Intel x64 machines too. gemini8_[RKN] brought it to my attention with is x64 Mac machines. He suspected that the issue is that while the apps page says it is for x64 intel machines, it was actually for apple silicon machines (since the app date was new). So I tried it on my apple silicon machine to see if it worked. It should work regardless with Rosetta2, but no. The same error as his intel x64 machines.

[Michael H.W. Weber]

However, as I mentioned, you cannot enable swap on these devices because it will negatively impact the lifetime of microSD cards. I have only a few with SSDs like that one with 16GB swap, but it failed here anyway.

...ever heard of high endurance µSD cards - originally designed for surveillance cams? Use these for SBCs which use SD cards as system storage (preferentialy Samsung as they appear to guarantee longest life time).

Michael.