Opened 5 years ago

Closed 4 years ago

#1526 closed defect (invalid)

Bug writing pe_hostfile binding strategy ?

Reported by: Didier.Rebeix@… Owned by:
Priority: normal Milestone:
Component: sge Version: 8.1.8
Severity: minor Keywords:
Cc:

Description

Hi there,

Trying to use SGE core binding feature, I'm facing strange binding strategy in the generated pe_hostfile.

If I submit a dmp job with " -binding pe linear:slots " every node in the pe_hostfile seems to get the same binding strategy.

Below are 2 examples of strange pe_hostfiles and corresponding qsub options :

#################### example 1 ####################
qsub -q batch -pe dmp* 64 -binding pe linear:slots simple.job

part061.u-bourgogne.fr 12 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5
part081.u-bourgogne.fr 12 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5
part065.u-bourgogne.fr 12 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5
part060.u-bourgogne.fr 11 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5
part083.u-bourgogne.fr 10 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5
part082.u-bourgogne.fr 7 batch@… 0,0:0,1:0,2:0,3:0,4:0,5:1,0:1,1:1,2:1,3:1,4:1,5

#################### example 2 ####################
qsub -q batch -pe dmp* 18 -binding pe linear:slots simple.job

part065.u-bourgogne.fr 6 batch@… 1,0:1,1:1,2:1,3:1,4:1,5
part061.u-bourgogne.fr 12 batch@… 1,0:1,1:1,2:1,3:1,4:1,5

It looks like the first binding strategy for the first host is calculated right but is wrongly applied to all other nodes.

I'm using sge-8.1.8.

All my dmp PEs (1 per IB switch) are configured the same way :

# qconf -sp dmp_swib1
pe_name dmp_swib1
slots 1000
user_lists NONE
xuser_lists NONE
start_proc_args /usr/ccub/sge/pe/dmp/startdmp.sh -catch_rsh $pe_hostfile
stop_proc_args /usr/ccub/sge/pe/dmp/stopdmp.sh
allocation_rule $fill_up
control_slaves TRUE
job_is_first_task FALSE
urgency_slots min
accounting_summary FALSE
qsort_args NONE

Feature or bug ?

Thank !

--

Didier Rebeix

Centre de Calcul et Messageries
Université de Bourgogne
Maison de l’université
Esplanade Erasme - BP 27877
21078 Dijon Cedex

TEL : 03.80.39.52.05 / FAX : 03.80.39.52.69

Change History (1)

comment:1 Changed 4 years ago by dlove

  • Resolution set to invalid
  • Status changed from new to closed

Sorry this got missed originally.

The entries in the pe_hostfile aren't used to set up the binding -- that's worked out by the execd.

Note: See TracTickets for help on using tickets.