[HTCondor-users] group quotas and "resources". What is a resource ?

[SCHAER Frederic <frederic.schaer@xxxxxx>]

Hi,

 

I have the following condor_userprio output :

 

# condor_userprio -most

Last Priority Update:  3/6  13:50

Group                                Config     Use    Effective   Priority   Res   Total Usage  Time Since Requested

  User Name                           Quota   Surplus   Priority    Factor   In Use (wghted-hrs) Last Usage Resources

----------------------------------- --------- ------- ------------ --------- ------ ------------ ---------- ----------

group_cms                                0.21 Regroup              100000.00     28      6366.64      <now>         26

  production@domain                              50000.00 100000.00      0       596.83    0+00:02

  admin@domain                                  100000.00 100000.00      1       703.86      <now>

  pilot@domain                                 2688074.75 100000.00     27      5065.94      <now>

group_alice                              0.05 Regroup              100000.00    140     59494.25      <now>        169

  pilot@domain                                   50000.00 100000.00      0        53.44    0+00:02

  admin@domain                                14000264.00 100000.00    140     59440.74      <now>

group_atlas                              0.33 Regroup              100000.00     48    266212.78      <now>        537

  admin@domain                                   50000.00 100000.00      0       660.46    0+00:02

  pilot@domain                                   50000.00 100000.00      0       641.35    0+00:03

  production@domain                            4799960.00 100000.00     48    264917.69      <now>

<none>                                   0.00 yes                  100000.00      0       595.39    0+00:33          0

  group_ops.admin@domain                         50000.00 100000.00      0       542.80    0+00:33

----------------------------------- --------- ------- ------------ --------- ------ ------------ ---------- ----------

Number of users: 9                            Regroup                           216    332623.12    0+23:59

 

 

In there, you can see that atlas consumed about 80% of the “Total usage”.

But thing is that we see in our monitoring that  the most consuming people are –by far- the alice people.

But they are consuming 2048MB of RAM per Cpu, whereas atlas is “consuming” 40GB (yes..) per 8 cpus (multicore jobs).

In fact the atlas jobs are requesting 40gb of RAM, but they are not using that much.

 

I’m therefore wondering what is a “resource” in condor terms, when we speak about group quotas ?

I see that a slot classad declares this : MachineResources = "Cpus Memory Disk Swap"

 

Are the accounting groups CPU related ?

Is the memory taken into account in group quotas ?

(How) can I make sure that only Cpu resources are taken into account in the fair share ?

 

Thanks

 

Frederic