[PL #3595] some slices not given any CPU shares

Fri Jan 7 10:56:31 EST 2005

Email Recipients (see http://www.planet-lab.org/Support)
       Requestor: acb at cs.princeton.edu
       Ticket Ccs: frankeh at watson.ibm.com

==================================================

CPU reservations on these nodes are failing because there are no shares 
left in the "free" pool, even though the number of outstanding shares is 
much less than the 8192, the total number of shares.  I haven't been 
able to reproduce this problem yet.  However, after looking at the logs 
on some of these nodes, it appears that there is a strong correlation 
between CPU reservations starting to fail and the upgrade to the last 
version of resman (0.02-4).  This sounds related to ticket 3593:

https://rt.planet-lab.org/Ticket/Display.html?id=3593

I suggest that we reboot these nodes and continue to monitor the 
situation.

Andy

Andy Bavier via RT wrote:
> Email Recipients (see http://www.planet-lab.org/Support)
>        Requestor: acb at cs.princeton.edu
> 
> 
> ==================================================
> 
> Wed Jan 05 16:07:15 2005: Request 3595 was acted upon.
> Transaction: Ticket created by Andy
> 
> Subject: some slices not given any CPU shares
> 
> On 98 nodes, there are currently slices that have a CPU reservation of 
> 0.  These slices receive hardly any CPU when there is contention.  The 
> list of nodes on which I observe this, and the number of slices without 
> CPU reservations on each node, is included below.
> 
> Specifically, for a slice with no CPU reservation, 'cpulimit getlimit 
> <slicename>' returns 32 but /rcfs/taskclass/<slicename>/shares lists the 
> CPU reservation as 0.  Executing 'cpulimit on <slicename>' fails. 
> Rebooting the node appears to fix the problem (when the node comes back 
> up, all slices have the proper CPU reservations).
> 
> Andy
> 
> ---------------------
> 
> csplanetlab1.kaist.ac.kr: 10
> freedom.ri.uni-tuebingen.de: 2
> pl1.6test.edu.cn: 21
> pl1.cs.utk.edu: 18
> pl2.swri.org: 5
> planet02.csc.ncsu.edu: 10
> planet1.calgary.canet4.nodes.planet-lab.org: 16
> planet1.cavite.nodes.planet-lab.org: 5
> planet1.cs.rochester.edu: 11
> planet1.halifax.canet4.nodes.planet-lab.org: 19
> planet1.leixlip.nodes.planet-lab.org: 10
> planet1.ottawa.canet4.nodes.planet-lab.org: 26
> planet1.pittsburgh.intel-research.net: 16
> planet1.toronto.canet4.nodes.planet-lab.org: 29
> planet1.winnipeg.canet4.nodes.planet-lab.org: 22
> planet2.calgary.canet4.nodes.planet-lab.org: 19
> planet2.leixlip.nodes.planet-lab.org: 1
> planet2.pittsburgh.intel-research.net: 22
> planet2.toronto.canet4.nodes.planet-lab.org: 19
> planet3.seattle.intel-research.net: 18
> planetlab01.ethz.ch: 6
> planetlab10.millennium.berkeley.edu: 36
> planetlab11.millennium.berkeley.edu: 26
> planetlab12.millennium.berkeley.edu: 27
> planetlab13.millennium.berkeley.edu: 25
> planetlab14.millennium.berkeley.edu: 23
> planetlab15.millennium.berkeley.edu: 23
> planetlab16.millennium.berkeley.edu: 12
> planetlab1.atla.internet2.planet-lab.org: 9
> planetlab1.bgu.ac.il: 11
> planetlab1.cnds.jhu.edu: 11
> planetlab1.comet.columbia.edu: 53
> planetlab1.cs.dartmouth.edu: 32
> planetlab1.cs.umd.edu: 23
> planetlab1.cs.uoregon.edu: 34
> planetlab1.diku.dk: 31
> planetlab1.eecs.umich.edu: 34
> planetlab1.enel.ucalgary.ca: 8
> planetlab1.ewi.tudelft.nl: 6
> planetlab1.flux.utah.edu: 19
> planetlab1.hstn.internet2.planet-lab.org: 1
> planetlab1.im.ntu.edu.tw: 22
> planetlab1.info.ucl.ac.be: 14
> planetlab1.inria.fr: 21
> planetlab1.isi.jhu.edu: 7
> planetlab1.it.uts.edu.au: 32
> planetlab1.kscy.internet2.planet-lab.org: 5
> planetlab1.millennium.berkeley.edu: 21
> planetlab1.nycm.internet2.planet-lab.org: 13
> planetlab1.polito.it: 14
> planetlab1.singapore.equinix.planet-lab.org: 26
> planetlab-1.stanford.edu: 36
> planetlab1.ucsd.edu: 34
> planetlab1.unl.edu: 26
> planetlab2.atla.internet2.planet-lab.org: 14
> planetlab2.cis.upenn.edu: 21
> planetlab2.cnds.jhu.edu: 11
> planetlab2.comet.columbia.edu: 44
> planetlab2.cs.dartmouth.edu: 24
> planetlab2.cs.duke.edu: 24
> planetlab2.cs.purdue.edu: 27
> planetlab2.cs.uiuc.edu: 14
> planetlab2.cs.umd.edu: 25
> planetlab2.cs.unb.ca: 3
> planetlab2.cs.vu.nl: 16
> planetlab2.diku.dk: 24
> planetlab2.ewi.tudelft.nl: 5
> planetlab2.flux.utah.edu: 3
> planetlab2.im.ntu.edu.tw: 4
> planetlab2.ipls.internet2.planet-lab.org: 3
> planetlab2.isi.jhu.edu: 1
> planetlab2.millennium.berkeley.edu: 11
> planetlab2.polito.it: 2
> planetlab2.postel.org: 9
> planetlab2.singapore.equinix.planet-lab.org: 23
> planetlab2.sttl.internet2.planet-lab.org: 9
> planetlab2.ucsd.edu: 24
> planetlab2.unl.edu: 28
> planetlab2.wash.internet2.planet-lab.org: 4
> planetlab2.xeno.cl.cam.ac.uk: 28
> planetlab-3.cmcl.cs.cmu.edu: 29
> planetlab3.comet.columbia.edu: 11
> planetlab3.cs.duke.edu: 23
> planetlab3.cs.uoregon.edu: 17
> planetlab3.millennium.berkeley.edu: 13
> planetlab3.singapore.equinix.planet-lab.org: 11
> planetlab3.xeno.cl.cam.ac.uk: 29
> planetlab4.millennium.berkeley.edu: 20
> planetlab5.millennium.berkeley.edu: 20
> planetlab7.millennium.berkeley.edu: 20
> planetlab9.millennium.berkeley.edu: 14
> pli1-br-1.hpl.hp.com: 1
> pli1-pa-4.hpl.hp.com: 8
> pli2-br-2.hpl.hp.com: 6
> recall.snu.ac.kr: 9
> ricepl-1.cs.rice.edu: 32
> ricepl-3.cs.rice.edu: 6
> vn3.cs.wustl.edu: 7
> 
> 
> _______________________________________________
> Devel-community mailing list
> Devel-community at lists.planet-lab.org
> http://lists.planet-lab.org/mailman/listinfo/devel-community