Hi,

we have a working FS configured in our cluster. We have 3 big QoS who
share resources, and works fine. Now, we'd like to give more prio to
some specific users in a determinate QoS. So, following doc, we should
define those users and give some shares to them, so, our QoS section for
that "group" should look like:

GROUPCFG[cmt1a]         FSTARGET=5      QDEF=lhccms
GROUPCFG[cmt1p]         FSTARGET=75     QDEF=lhccms
GROUPCFG[cmprd]         FSTARGET=20     QDEF=lhccms
GROUPCFG[sgmcm]         PRIORITY=1000   QDEF=lhccms   MAXPROC=3
USERCFG[cmprd002]       FSTARGET=2000   QDEF=lhcms
USERCFG[cmprd001]       FSTARGET=2000   QDEF=lhcms
QOSCFG[lhccms]          FSTARGET=24.89

*We have added both USERCFG lines.


FS policy is defined like:

# Set up Fair Share weight component
FSWEIGHT                1
FSUSERWEIGHT            2
FSGROUPWEIGHT           10
FSQOSWEIGHT             100
FSDEPTH                 4
FSINTERVAL              12:00:00
FSDECAY                 0.5
FSPOLICY                DEDICATEDPS%


But there's something wrong in this conf cause when sending jobs from
user cmprd003 (not explicitly defined with a USERCFG in that QoS) it has
more prio than cmprd001/cmprd002 jobs.


diagnose -Q is not giving relevant info:

# diagnose -Q
QOS Status

System QOS Settings:  QList: DEFAULT (Def: DEFAULT)  Flags: 0

Name                * Priority QTWeight QTTarget XFWeight XFTarget     QFlags   
JobFlags Limits

DEFAULT                      0        0        0        0     0.00     [NONE]   
  [NONE] [NONE]
[ALL]                        0        0        0        0     0.00     [NONE]   
  [NONE] [NONE]
[...]
lhccms                       0        0        0        0     0.00     [NONE]   
  [NONE] [NONE]
  Groups:    sgmcm cmt1p cmt1a cmprd
lhclhcb                      0        0        0        0     0.00     [NONE]   
  [NONE] [NONE]
  Groups:    lhcb lhsgm lhpilot lhprd
lhcms                        0        0        0        0     0.00     [NONE]   
  [NONE] [NONE]
  Users:     cmprd002 cmprd001
[...]

diagnose -f:


# diagnose -f|grep cmpr
cmprd008          0.00 ------- ------- -------    0.00 -------
cmprd003          0.00 -------    0.00    0.00 ------- -------
cmprd002*         0.45 2000.00    0.84    0.34    0.31    0.69
cmprd001*         0.00 2000.00 ------- ------- ------- -------
cmprd*            0.45  20.00     0.84    0.34    0.32    0.69


So, do we have to define ALL users in cmprd* pool in order to make
USERWEIGHT working?

am I missunderstanding FS doc?
http://www.adaptivecomputing.com/resources/docs/maui/6.3fairshare.php 

Any other way for doing so?

Many thanks in advance,
Cheers,
Arnau
_______________________________________________
mauiusers mailing list
[email protected]
http://www.supercluster.org/mailman/listinfo/mauiusers

Reply via email to