Hi all:

I administer several HPC clusters; unfortunately, its quite often that a couple of them are unloaded (completely). I'd like to put that to use by doing boinc compute projects, but only until a "real job" is scheduled. (One of the clusters hasn't had a "real job" in 3 months! And its 196 AMD 2.4Ghz cores!; the other cluster has 4x the cores, but sees use for a few days a month on average.).

The only way I know about handling this presently would be to create 48 seperate BOINC directories (one for each node) with seperate installs or hard links for much of the installs; configure each install, then launch the boinc process for each node, such that it will suspend if non-boinc usage exceeds a set point.

This is possible, but this is a lot of work to donate cpu cycles...If there was a way to NOT have to create 48 seperate directories (i.e., allow BOINC to share a directory), that would be helpful. Is there an "easy way" to do this?

Thanks!
--Jim

Hi all:

I administer several HPC clusters; unfortunately, its quite often that a couple of them are unloaded (completely). I'd like to put that to use by doing boinc compute projects, but only until a "real job" is scheduled. (One of the clusters hasn't had a "real job" in 3 months! And its 196 AMD 2.4Ghz cores!; the other cluster has 4x the cores, but sees use for a few days a month on average.).

The only way I know about handling this presently would be to create 48 seperate BOINC directories (one for each node) with seperate installs or hard links for much of the installs; configure each install, then launch the boinc process for each node, such that it will suspend if non-boinc usage exceeds a set point.

This is possible, but this is a lot of work to donate cpu cycles...If there was a way to NOT have to create 48 seperate directories (i.e., allow BOINC to share a directory), that would be helpful. Is there an "easy way" to do this?

Thanks!
--Jim

I am not certain about directories, but you are not the first to do this on a cluster, and there are some settings specifically designed to deal with clusters. First is that you have to set the flag that allows multiple BOINC instances to run on a single machine. Second, you can set "report results immediately", and last [edit] that I know about, I may have missed one [/edit] there is a setting that will halt BOINC execution for a given node once the task is done so that it can be assigned to the next task (whatever that happens to be).

---

BOINC WIKI