)]}'
{
  "commit": "8afdcece4911e51cfff2b50a269418914cab8a3f",
  "tree": "fcfb966822f0f6c128c754f3876a80106c9cc654",
  "parents": [
    "7571966189e54adf0a8bc1384d6f13f44052ba63"
  ],
  "author": {
    "name": "Mel Gorman",
    "email": "mel@csn.ul.ie",
    "time": "Tue Mar 22 16:33:04 2011 -0700"
  },
  "committer": {
    "name": "Linus Torvalds",
    "email": "torvalds@linux-foundation.org",
    "time": "Tue Mar 22 17:44:04 2011 -0700"
  },
  "message": "mm: vmscan: kswapd should not free an excessive number of pages when balancing small zones\n\nWhen reclaiming for order-0 pages, kswapd requires that all zones be\nbalanced.  Each cycle through balance_pgdat() does background ageing on\nall zones if necessary and applies equal pressure on the inactive zone\nunless a lot of pages are free already.\n\nA \"lot of free pages\" is defined as a \"balance gap\" above the high\nwatermark which is currently 7*high_watermark.  Historically this was\nreasonable as min_free_kbytes was small.  However, on systems using huge\npages, it is recommended that min_free_kbytes is higher and it is tuned\nwith hugeadm --set-recommended-min_free_kbytes.  With the introduction of\ntransparent huge page support, this recommended value is also applied.  On\nX86-64 with 4G of memory, min_free_kbytes becomes 67584 so one would\nexpect around 68M of memory to be free.  The Normal zone is approximately\n35000 pages so under even normal memory pressure such as copying a large\nfile, it gets exhausted quickly.  As it is getting exhausted, kswapd\napplies pressure equally to all zones, including the DMA32 zone.  DMA32 is\napproximately 700,000 pages with a high watermark of around 23,000 pages.\nIn this situation, kswapd will reclaim around (23000*8 where 8 is the high\nwatermark + balance gap of 7 * high watermark) pages or 718M of pages\nbefore the zone is ignored.  What the user sees is that free memory far\nhigher than it should be.\n\nTo avoid an excessive number of pages being reclaimed from the larger\nzones, explicitely defines the \"balance gap\" to be either 1% of the zone\nor the low watermark for the zone, whichever is smaller.  While kswapd\nwill check all zones to apply pressure, it\u0027ll ignore zones that meets the\n(high_wmark + balance_gap) watermark.\n\nTo test this, 80G were copied from a partition and the amount of memory\nbeing used was recorded.  A comparison of a patch and unpatched kernel can\nbe seen at\nhttp://www.csn.ul.ie/~mel/postings/minfree-20110222/memory-usage-hydra.ps\nand shows that kswapd is not reclaiming as much memory with the patch\napplied.\n\nSigned-off-by: Andrea Arcangeli \u003caarcange@redhat.com\u003e\nSigned-off-by: Mel Gorman \u003cmel@csn.ul.ie\u003e\nAcked-by: Rik van Riel \u003criel@redhat.com\u003e\nCc: Shaohua Li \u003cshaohua.li@intel.com\u003e\nCc: \"Chen, Tim C\" \u003ctim.c.chen@intel.com\u003e\nSigned-off-by: Andrew Morton \u003cakpm@linux-foundation.org\u003e\nSigned-off-by: Linus Torvalds \u003ctorvalds@linux-foundation.org\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "c335055c4253e49a18f5513cc65db47162802bb6",
      "old_mode": 33188,
      "old_path": "include/linux/swap.h",
      "new_id": "ed6ebe690f4ac9211a253d162c92c9b5fd98d71b",
      "new_mode": 33188,
      "new_path": "include/linux/swap.h"
    },
    {
      "type": "modify",
      "old_id": "665b090b6c727369368d08ee382161b92e075baa",
      "old_mode": 33188,
      "old_path": "mm/vmscan.c",
      "new_id": "060e4c1914033fe61b2f06a445d6935b559eba2f",
      "new_mode": 33188,
      "new_path": "mm/vmscan.c"
    }
  ]
}