[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Condor-users] Condor Fair Share ?

Title: Condor Fair Share ?

Hi Guys,

Need your advice on the below issue.

I have two machines which Machine A serves as central manager (submission and exectution host) and Machine B is served as submission and execution host. Both machines have 2 cpus means I have total of 4 cpus.

The central manager policy is configured as below. 

## Condor's automatic adjustment of user priorities (SET TO 1 DAY)

# define the groups in your pool
GROUP_NAMES = group_pend, group_penn21004

# specify what share of the pool (# of batch slots) each group should have
GROUP_QUOTA_A           = 50

This means I have defined two groups (A and B).

I have created a simple vanilla job with perl script.

$indexSize = 500;
for ($i=1; $i<=$indexSize; $i++){
        my @arg = system("condor_submit name.submit");

I use the perl script to keep submitting the same job for 500 times and the script has been executed on both machines. The reason of this test is actually to verify the Condor fair-share according to the group quota defined in the central manager (condor_config) file.

Experiment 1:
I started with Machine B (group_B) to submit the jobs by executing the above perl script and then came to Machine A (group_A) and did the same procedure.

After a while I can only see the Machine B took all the resources while jobs submitted from Machine A kept idle and started to run once all the Machine B jobs finished.

Experiment 2:
I use the same procedure to submit the job but different order. This time I used to submit the jobs from Machine A first and next Machine B.

In this experiment I can see the Machine A jobs took 3 resources untill all the jobs finished and Machine B which is initially took 1 resources during Machine A jobs running then took all resources after the Machine A jobs completed.

According to the group quota defined in the policy, in whatever situation, group B should take 2 times the amount of Condor resources compared to group A? Am I right? . But the results tell different.

Is there are any other configurations that I missed out to make the group quota works properly.

Thanks in advance for your advice.