Advanced search

Message boards : Technical Support : camb_boinc2docker fails after 15 minutes

Author Message
FlashGordon
Send message
Joined: 20 Nov 07
Posts: 1
Credit: 63,466
RAC: 0
Message 21457 - Posted: 30 May 2017, 8:37:26 UTC

Hi,

I just saw that on my machine the docker tasks all fail after 15 minutes of doing nothing. One example is here:

https://www.cosmologyathome.org/result.php?resultid=54228148

I do not know if this part is the reason:



2017-05-30 15:44:36 (12380): Guest Log: net.ipv4.ip_forward = 1
2017-05-30 15:44:36 (12380): Guest Log: sysctl: cannot stat /proc/sys/net/ipv6/conf/all/forwarding: No such file or directory
2017-05-30 15:44:36 (12380): Guest Log: sysctl: setting key "cannot stat %s": No such file or directory
2017-05-30 15:44:36 (12380): Guest Log: sysctl: "cannot stat %s" is an unknown key
2017-05-30 15:44:36 (12380): Guest Log: sysctl: setting key "cannot stat %s": No such file or directory
2017-05-30 15:44:36 (12380): Guest Log: Segmentation fault



2017-05-30 15:44:46 (12380): Guest Log: Tue May 30 15:44:35 UTC 2017 dhcp -------------------------------
2017-05-30 16:00:06 (12380): Powering off VM.



Other BOINC Vbox tasks (for example LHC) work fine and have network access. Any idea what is wrong?

Jesse Viviano
Send message
Joined: 29 Nov 14
Posts: 28
Credit: 418,604
RAC: 699
Message 21498 - Posted: 18 Aug 2017, 14:00:50 UTC

Go to the project's directory. On Windows 10, it will be located at C:\ProgramData\BOINC\projects\www.cosmologyathome.org. Then create a file named "app_config.xml" and put the following code into it:

<app_config> <app_version> <app_name>camb_boinc2docker</app_name> <plan_class>vbox64_mt</plan_class> <avg_ncpus>8</avg_ncpus> </app_version> </app_config>


This will force the Cosmology@home application to use no more than 8 virtual CPUs, or vCPUs for short. There is a bug in Cosmology@home where it will crash if it uses more than 8 vCPUs per task. I noticed that your Ryzen machine has 6 physical cores, so it can provide up to 12 vCPUs thanks to its SMT capability. The project's administrators should configure the project to use no more than 8 vCPUs to work around this bug.

bzaborow
Send message
Joined: 9 Apr 17
Posts: 1
Credit: 732,240
RAC: 1,952
Message 21502 - Posted: 28 Aug 2017, 20:04:56 UTC - in response to Message 21498.

Just figured out myself that there is a problem with using >8 cores by one VM. And looking for a solution I came here - tried it and it works, thanks!

Just one notice to any people using this approach: the app_config is not reloaded before every task, so you need to restart boinc to make changes work.

Message boards : Technical Support : camb_boinc2docker fails after 15 minutes