Unexpected number of core used on remote server

#1

Hello,
I’m currently working on a remote server of 16 cores but I want to use only 4 of them (because everyone in my lab work on this server). Open Mole is not respecting the limit on the number of core I gave him in the SSHEnvironment declaration, it’s using all the 16 cores. Does anyone has experienced something like this ?

I’m using the following script (I skipped the model definition part) :

val nsga2 = NSGA2(
mu = 50,
genome = Seq(T0 in (0.0, 3.0), cexp in (0.8, 1.0), pas in (0.1, 5.0), niter in (100.0, 10000.0)),
objectives = Seq(doubleValue)
)

val evolution = SteadyStateEvolution(
algorithm = nsga2,
evaluation = AckleyFunction – toDouble,
parallelism = 10,
termination = 30
)

val savePopulation = SavePopulationHook(evolution, workDirectory / “evolution/”)

val env =
SSHEnvironment(
“login”,
“host”,
4 // The number of cores you want to use on the remote server
)

(evolution hook savePopulation on env)

I’m adding a screenshot of what I experienced this morning on the server :

openmole_load.png1514x440 284 KB

#2

Hi Leonard,

indeed the SSHEnvironment is buggy in the version 7. I redesigned all the environments in the version 8. It might be not that stable but the solution for now would be to use the dev version for now:
https://next.openmole.org/

Could you confirm that you don’t get the same pb with version if you try it?

Romain

#3

PS: That would be very hard to fix in the version 7, that’s why I had to rework the environment in depth in the upcoming version.

#4

Hello,
I tried the same workflow on the same remote server with OpenMole 8, this is what I observe with htop :

openmolev8.png961x146 67.7 KB

This is more or less the same at any time : all cores are used by openMole but less than with the older version.

It seems also that a small CPU work is still going on even after the cancellation of the Open Mole Task. It’s around 10% on all cores and it was already happening with v7.1 :s

EDIT
It’s actually difficult to say what is happening with OMv8, same workflow, same server, I observe that the number of core used varies from 2 to 10 but for each core the workload on each core is small. Most of the cores are around 7%