RE: too much migrations?
"Walker, Bruce J (HP-Labs)" <[email protected]>
| Newsgroups | gmane.linux.cluster.ssic.user |
|---|---|
| Message-ID | <3689AF909D816446BA505D21F1461AE405A4637B@cacexc04.americas.cpqcorp.net> |
To add to what Roger said: If, with the exception of one pretty busy process, the nodes are balanced, then that one process could keep moving back and forth. The problem is that with it running on one node, that node looks relatively overloaded but then just moving it makes the other node overloaded. You need more load levelable load to allow the system to balance stuff out. With just the one process there is no great place to put it. To completely solve this one would have to carefully assess the load a given process is generating and then assess whether moving it to an unloaded node would just cause that node to now be the loaded one. The currently load balancing algorithm is not quite that sophisticated. Bruce -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Roger Tsang Sent: Friday, April 14, 2006 9:54 AM To: Maurice Libes Cc: [email protected] Subject: Re: [SSI-users] too much migrations? To take advantage of process load-balancing you want to start n+1 parallel processes. Roger On 4/14/06, Maurice Libes <[email protected]> wrote: > > hi to all > > i ask again an old question because i don't know if such an amount of > migration (here below) between nodes is normal? (on oSSI 1.2 on FC2) > > i have loadlevelled a process with the > loadlevel -p <pid> command > > to my opinion, i am astonished to see that this process is migrating > too frequently i am afraid that this process spends too much computing > time in migrating from node to nodes.. > > according to you is that a normal situation? > > example here below : > the process 181870(symphonie) leaves node 2 at > 1145007809 and come back on the same node immediately 54seconds later > at 1145007863, then restart at 1145007872 and come back at 1145007918 > > do i miss something or something goes wrong? thanks for your advice > > [root@comclust2 root]# cat /proc/cluster/loadlog | grep 181870 > 1145007809 loadbl:pid 181870(symphonie) -> node 1 mem 49780 my load 30 > node1 load 25 > 1145007863 loadlb:pid 181870(symphonie) <- node 1 mem 115296 my load > 15 > node1 load 52 > 1145007872 loadbl:pid 181870(symphonie) -> node 1 mem 98228 my load 31 > node1 load 27 > 1145007918 loadlb:pid 181870(symphonie) <- node 1 mem 84680 my load 16 > node1 load 30 > 1145007920 loadbl:pid 181870(symphonie) -> node 1 mem 84248 my load 31 > node1 load 22 > 1145007935 loadlb:pid 181870(symphonie) <- node 1 mem 80048 my load 16 > node1 load 39 > 1145007938 loadbl:pid 181870(symphonie) -> node 1 mem 79448 my load 31 > node1 load 23 > 1145008011 loadlb:pid 181870(symphonie) <- node 1 mem 57536 my load 15 > node1 load 35 > 1145008060 loadbl:pid 181870(symphonie) -> node 1 mem 50284 my load 30 > node1 load 25 > 1145008110 loadlb:pid 181870(symphonie) <- node 1 mem 49488 my load 15 > node1 load 55 > 1145008153 loadbl:pid 181870(symphonie) -> node 1 mem 49688 my load 31 > node1 load 24 > 1145008177 loadlb:pid 181870(symphonie) <- node 1 mem 123756 my load > 15 > node1 load 83 > 1145008188 loadbl:pid 181870(symphonie) -> node 1 mem 117192 my load > 30 > node1 load 27 > 1145008219 loadlb:pid 181870(symphonie) <- node 1 mem 71652 my load 15 > node1 load 34 > 1145008221 loadbl:pid 181870(symphonie) -> node 1 mem 71256 my load 32 > node1 load 24 > 1145008323 loadlb:pid 181870(symphonie) <- node 1 mem 153988 my load > 14 > node1 load 70 > 1145008339 loadbl:pid 181870(symphonie) -> node 1 mem 56220 my load 23 > node1 load 25 > 1145008374 loadlb:pid 181870(symphonie) <- node 1 mem 56296 my load 14 > node1 load 69 > 1145008416 loadbl:pid 181870(symphonie) -> node 1 mem 55296 my load 23 > node1 load 24 > 1145008445 loadlb:pid 181870(symphonie) <- node 1 mem 55636 my load 14 > node1 load 37 > 1145008652 loadbl:pid 181870(symphonie) -> node 1 mem 50864 my load 23 > node1 load 25 > 1145008689 loadlb:pid 181870(symphonie) <- node 1 mem 52524 my load 14 > node1 load 81 > 1145008844 loadbl:pid 181870(symphonie) -> node 1 mem 49280 my load 23 > node1 load 24 > 1145008864 loadlb:pid 181870(symphonie) <- node 1 mem 50552 my load 14 > node1 load 46 > 1145015003 loadbl:pid 181870(symphonie) -> node 3 mem 50064 my load 43 > node3 load 30 > 1145015189 loadlb:pid 181870(symphonie) <- node 3 mem 238076 my load > 15 > node3 load 41 > 1145015330 loadbl:pid 181870(symphonie) -> node 3 mem 49732 my load 41 > node3 load 30 > 1145016442 loadlb:pid 181870(symphonie) <- node 3 mem 48652 my load 35 > node3 load 28 > 1145016515 loadbl:pid 181870(symphonie) -> node 3 mem 47164 my load 53 > node3 load 35 > 1145017288 loadlb:pid 181870(symphonie) <- node 3 mem 47132 my load 38 > node3 load 28 > 1145017301 loadbl:pid 181870(symphonie) -> node 3 mem 47536 my load 61 > node3 load 35 > 1145017435 loadlb:pid 181870(symphonie) <- node 3 mem 58580 my load 16 > node3 load 39 > 1145017491 loadbl:pid 181870(symphonie) -> node 3 mem 47760 my load 50 > node3 load 32 > 1145017711 loadlb:pid 181870(symphonie) <- node 3 mem 47148 my load 41 > node3 load 31 > 1145017714 loadbl:pid 181870(symphonie) -> node 3 mem 48420 my load 51 > node3 load 38 > 1145018049 loadlb:pid 181870(symphonie) <- node 3 mem 47100 my load 37 > node3 load 28 > 1145018051 loadbl:pid 181870(symphonie) -> node 3 mem 48484 my load 45 > node3 load 37 > 1145018331 loadlb:pid 181870(symphonie) <- node 3 mem 46836 my load 32 > node3 load 28 > 1145018370 loadbl:pid 181870(symphonie) -> node 3 mem 48336 my load 69 > node3 load 35 > 1145018945 loadlb:pid 181870(symphonie) <- node 3 mem 70116 my load 18 > node3 load 71 > 1145019030 loadbl:pid 181870(symphonie) -> node 3 mem 48880 my load 43 > node3 load 37 > 1145019126 loadlb:pid 181870(symphonie) <- node 3 mem 48940 my load 21 > node3 load 41 > 1145019176 loadbl:pid 181870(symphonie) -> node 3 mem 49200 my load 46 > node3 load 35 > 1145019455 loadlb:pid 181870(symphonie) <- node 3 mem 362708 my load > 14 > node3 load 42 > 1145019572 loadbl:pid 181870(symphonie) -> node 3 mem 172132 my load > 36 > node3 load 30 > 1145019738 loadlb:pid 181870(symphonie) <- node 3 mem 351956 my load > 15 > node3 load 39 > 1145019868 loadbl:pid 181870(symphonie) -> node 3 mem 187400 my load > 40 > node3 load 30 > 1145020041 loadlb:pid 181870(symphonie) <- node 3 mem 304048 my load > 16 > node3 load 58 > 1145020173 loadbl:pid 181870(symphonie) -> node 3 mem 108976 my load > 36 > node3 load 30 > 1145020344 loadlb:pid 181870(symphonie) <- node 3 mem 337896 my load > 17 > node3 load 62 > 1145020475 loadbl:pid 181870(symphonie) -> node 3 mem 159416 my load > 41 > node3 load 30 > 1145020639 loadlb:pid 181870(symphonie) <- node 3 mem 286468 my load > 14 > node3 load 73 > 1145020653 loadbl:pid 181870(symphonie) -> node 3 mem 143900 my load > 48 > node3 load 35 > 1145020813 loadlb:pid 181870(symphonie) <- node 3 mem 242244 my load > 14 > node3 load 71 > 1145020947 loadbl:pid 181870(symphonie) -> node 3 mem 49452 my load 37 > node3 load 30 > 1145021116 loadlb:pid 181870(symphonie) <- node 3 mem 228888 my load > 16 > node3 load 52 > 1145021264 loadbl:pid 181870(symphonie) -> node 3 mem 48624 my load 57 > node3 load 35 > 1145021378 loadlb:pid 181870(symphonie) <- node 3 mem 230680 my load > 14 > node3 load 53 > > > > -- > Maurice Libes > Tel : +33 (04) 91 82 93 25 Centre d'Oceanologie de Marseille > Fax : +33 (04) 91 82 65 48 UMS2196 CNRS- Campus de Luminy, > Case 901 > mailto:[email protected] F-13288 Marseille cedex 9 > Annuaire : http://annuaire.univmed.fr/showuser.php?uid=libes > > > ------------------------------------------------------- This SF.Net email is sponsored by xPML, a groundbreaking scripting language that extends applications into web and mobile media. Attend the live webcast and join the prime developer group breaking into this new coding territory! http://sel.as-us.falkag.net/sel?cmd=k&kid0944&bid$1720&dat1642 _______________________________________________ Ssic-linux-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ssic-linux-users ------------------------------------------------------- This SF.Net email is sponsored by xPML, a groundbreaking scripting language that extends applications into web and mobile media. Attend the live webcast and join the prime developer group breaking into this new coding territory! http://sel.as-us.falkag.net/sel?cmd=lnk&kid0944&bid$1720&dat1642