)]}'
{
  "commit": "d5b8aa1d246fddfe4042be6f6eb169efa5cfbb94",
  "tree": "5c0be52e7fba52ba38c8cd606b5c15e2435e343a",
  "parents": [
    "85a43a9edaf5b541381acbf4061bace1121d6ef0"
  ],
  "author": {
    "name": "jamal",
    "email": "hadi@cyberus.ca",
    "time": "Sun Jun 26 08:13:54 2011 +0000"
  },
  "committer": {
    "name": "David S. Miller",
    "email": "davem@davemloft.net",
    "time": "Mon Jun 27 00:14:10 2011 -0700"
  },
  "message": "net_sched: fix dequeuer fairness\n\nResults on dummy device can be seen in my netconf 2011\nslides. These results are for a 10Gige IXGBE intel\nnic - on another i5 machine, very similar specs to\nthe one used in the netconf2011 results.\nIt turns out - this is a hell lot worse than dummy\nand so this patch is even more beneficial for 10G.\n\nTest setup:\n----------\n\nSystem under test sending packets out.\nAdditional box connected directly dropping packets.\nInstalled prio qdisc on the eth device and default\nnetdev default length of 1000 used as is.\nThe 3 prio bands each were set to 100 (didnt factor in\nthe results).\n\n5 packet runs were made and the middle 3 picked.\n\nresults\n-------\n\nThe \"cpu\" column indicates the which cpu the sample\nwas taken on,\nThe \"Pkt runx\" carries the number of packets a cpu\ndequeued when forced to be in the \"dequeuer\" role.\nThe \"avg\" for each run is the number of times each\ncpu should be a \"dequeuer\" if the system was fair.\n\n3.0-rc4      (plain)\ncpu         Pkt run1        Pkt run2        Pkt run3\n\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\ncpu0        21853354        21598183        22199900\ncpu1          431058          473476          393159\ncpu2          481975          477529          458466\ncpu3        23261406        23412299        22894315\navg         11506948        11490372        11486460\n\n3.0-rc4 with patch and default weight 64\ncpu \t     Pkt run1        Pkt run2        Pkt run3\n\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\u003d\ncpu0        13205312        13109359        13132333\ncpu1        10189914        10159127        10122270\ncpu2        10213871        10124367        10168722\ncpu3        13165760        13164767        13096705\navg         11693714        11639405        11630008\n\nAs you can see the system is still not perfect but\nis a lot better than what it was before...\n\nAt the moment we use the old backlog weight, weight_p\nwhich is 64 packets. It seems to be reasonably fine\nwith that value.\nThe system could be made more fair if we reduce the\nweight_p (as per my presentation), but we are going\nto affect the shared backlog weight. Unless deemed\nnecessary, I think the default value is fine. If not\nwe could add yet another knob.\n\nSigned-off-by: Jamal Hadi Salim \u003cjhs@mojatatu.com\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "b4c680900d7a8c9b417d1c976ef838f9eb5eb960",
      "old_mode": 33188,
      "old_path": "net/sched/sch_generic.c",
      "new_id": "d253c16a314ce85838ba24d448f90fe1ed3fe865",
      "new_mode": 33188,
      "new_path": "net/sched/sch_generic.c"
    }
  ]
}