Hello,

I am using Maui 3.2.6p16 with Torque 2.1.1 on an AMD 
Opteron (64 bit) running CentOS (~RedHat).

I have been using CONSUMEDWEIGHT to share the CPUs 
in use between the different users.  This used to 
work nicely in Maui 3.2.6p13.  But during one of the 
upgrades since then it broke.  I am using:

FSWEIGHT              1
CREDWEIGHT            1 
RESOURCEWEIGHT        1 
SERVICEWEIGHT         0 
USAGEWEIGHT           1 

CONSUMEDWEIGHT        -10

ENABLENEGJOBPRIORITY           TRUE


But with `diagnose -p` I see that the "USAGE" never 
makes a contribution, e.g.:
Job                    PRIORITY*   Cred(Class)    FS(Group)   Res( Proc)   
Res(Cons :Rem  :PerC )
             Weights   --------       1(    1)     1(  100)     1(    5)     1( 
  10:   10:   10)

419618                     -149     0.0(  0.0)  55.2(-788.)  44.8(640.0)   0.0( 
 0.0:  0.0:  0.0)
422524                     -880     0.0(  0.0)  86.7(-1039)  13.3(160.0)   0.0( 
 0.0:  0.0:  0.0)
422528                     -880     0.0(  0.0)  86.7(-1039)  13.3(160.0)   0.0( 
 0.0:  0.0:  0.0)

Percent Contribution   --------     0.0(  0.0)  74.9( 74.9)  25.1( 25.1)   0.0( 
 0.0:  0.0:  0.0)



Does anyone have a clue what I'm doing wrong / what 
has changed.

Best regards,

Simon.


maui.cfg

NODEACCESSPOLICY      SHARED

RMPOLLINTERVAL        00:01:30
NODEPOLLFREQUENCY     3

SERVERPORT            42559
SERVERMODE            NORMAL

LOGFILE               maui.log
LOGFILEMAXSIZE        10000000
LOGLEVEL              4

CLIENTTIMEOUT         00:01:00
DEFAULTDOMAIN         ALiCEnext

# Node configuration

JOBNODEMATCHPOLICY EXACTNODE
NODEMAXLOAD                     2.5  # don't use nodes with unusually high loads
NODEALLOCATIONPOLICY        FIRSTAVAILABLE

# Priority Weights

FSWEIGHT              1
CREDWEIGHT            1 # For idle queue weights
RESOURCEWEIGHT        1 # For parallel jobs
SERVICEWEIGHT         0 # For Xfactor (queue time based) boost
USAGEWEIGHT           1 # Priority based on number of running jobs (and 
walltime used)

QUEUETIMEWEIGHT       0
XFACTORWEIGHT         100
XFXAP                 10000

CONSUMEDWEIGHT        10

# Credential Priorities (used to supress the idle queue)
CLASSWEIGHT            1
CLASSCFG[DEFAULT] PRIORITY=0

# Try and stop jobs getting stuck waiting for crashed nodes:
QOSCFG[DEFAULT]         QFLAGS=IGNHOSTLIST
USERCFG[DEFAULT]        QLIST=DEFAULT QDEF=DEFAULT

# Resources priorities (for parallel jobs)
PROCWEIGHT        5

# Give idle a low priority:
CLASSCFG[idle]    MAXJOB=400 MAXJOBPERUSER=400 PRIORITY=-500000
CLASSCFG[short]   MAXJOB=250 MAXJOBPERUSER=250
CLASSCFG[medium]  MAXJOB=450 MAXJOBPERUSER=350
CLASSCFG[large]   MAXJOB=450 MAXJOBPERUSER=300
CLASSCFG[huge]    MAXJOB=120 MAXJOBPERUSER=60
CLASSCFG[theory]  MAXJOB=100 MAXJOBPERUSER=50 PRIORITY=-1000000

# Limit number of jobs for specific users
USERCFG[andorada] MAXJOB=50

# Standing reservation (small jobs during office hours)
SRCFG[quick]    PERIOD=DAY DAYS=MON,TUE,WED,THU,FRI
SRCFG[quick]    STARTTIME=11:00:00 ENDTIME=19:00:00
SRCFG[quick]    MAXTIME=1:01:00
SRCFG[quick]    TASKCOUNT=2 RESOURCES=PROCS:2
SRCFG[quick]    DEPTH=2 PARTITION=DEFAULT PLIST=DEFAULT PDEF=DEFAULT

SYSCFG[base] PLIST=DEFAULT
CLASSCFG[suse] PDEF=suse PLIST=suse&
#
# For testing purposes:
NODECFG[n512] PARTITION=simon  # MySQL tests
CLASSCFG[simon] PDEF=simon PLIST=simon&

# FairShare

FSPOLICY              DEDICATEDPS
FSDEPTH               14
FSINTERVAL            24:00:00
FSDECAY               0.90
FSGROUPWEIGHT         100
FSUSERWEIGHT          0
FSACCOUNTWEIGHT       0
FSQOSWEIGHT           0
FSCLASSWEIGHT         0

# JOB Deferal settings

DEFERTIME 00:05:00
DEFERCOUNT 1000

# Policies

REJECTNEGPRIOJOBS              FALSE
ENABLENEGJOBPRIORITY           TRUE
ENABLEMULTINODEJOBS            TRUE

BACKFILLPOLICY              FIRSTFIT
RESERVATIONPOLICY           CURRENTHIGHEST
RESERVATIONDEPTH            20 # how many advanced system reservations to make

MAXJOBPERUSERPOLICY         ON
MAXJOBPERUSERCOUNT          99999

#MAXPROCPERUSERPOLICY        OFF
#MAXPROCPERUSERCOUNT         256

MAXPROCSECONDPERUSERPOLICY  OFF
MAXPROCSECONDPERUSERCOUNT   36864000

MAXJOBQUEUEDPERUSERPOLICY   OFF
MAXJOBQUEUEDPERUSERCOUNT    2

#MAXPROCPERGROUPPOLICY       OFF
#SMAXPROCPERGROUPCOUNT       128
#MAXPROCPERGROUPCOUNT        160

JOBMAXOVERRUN               01:00:00  # Allow jobs to overrun wall time by 1 
hour
JOBAGGREGATIONTIME          00:00:10  # Maui is patient with bursts of jobs

(... fairshare weights)
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers

Reply via email to