From: Andrew Morton <akpm@linux-foundation.org> To: Rik van Riel <riel@redhat.com> Cc: elladan@eskimo.com, peterz@infradead.org, linux-kernel@vger.kernel.org, tytso@mit.edu, kosaki.motohiro@jp.fujitsu.com, linux-mm@kvack.org Subject: Re: [PATCH] vmscan: evict use-once pages first (v2) Date: Thu, 30 Apr 2009 19:54:39 -0700 [thread overview] Message-ID: <20090430195439.e02edc26.akpm@linux-foundation.org> (raw) In-Reply-To: <20090430215034.4748e615@riellaptop.surriel.com> On Thu, 30 Apr 2009 21:50:34 -0400 Rik van Riel <riel@redhat.com> wrote: > > Which would cause exactly the problem Elladan saw? > > Yes. It was not noticable in the initial split LRU code, > but after we decided to ignore the referenced bit on active > file pages and deactivate pages regardless, it has gotten > exacerbated. > > That change was very good for scalability, so we should not > undo it. However, we do need to put something in place to > protect the working set from streaming IO. > > > > Currently the kernel has no effective code to protect the > > > page cache working set from streaming IO. Elladan's bug > > > report shows that we do need some kind of protection... > > > > Seems to me that reclaim should treat swapcache-backed mapped mages in > > a similar fashion to file-backed mapped pages? > > Swapcache-backed pages are not on the same set of LRUs as > file-backed mapped pages. yup. > Furthermore, there is no streaming IO on the anon LRUs like > there is on the file LRUs. Only the file LRUs need (and want) > use-once replacement, which means that we only need special > protection of the working set for file-backed pages. OK. > When we implement working set protection, we might as well > do it for frequently accessed unmapped pages too. There is > no reason to restrict this protection to mapped pages. Well. Except for empirical observation, which tells us that biasing reclaim to prefer to retain mapped memory produces a better result.
WARNING: multiple messages have this Message-ID (diff)
From: Andrew Morton <akpm@linux-foundation.org> To: Rik van Riel <riel@redhat.com> Cc: elladan@eskimo.com, peterz@infradead.org, linux-kernel@vger.kernel.org, tytso@mit.edu, kosaki.motohiro@jp.fujitsu.com, linux-mm@kvack.org Subject: Re: [PATCH] vmscan: evict use-once pages first (v2) Date: Thu, 30 Apr 2009 19:54:39 -0700 [thread overview] Message-ID: <20090430195439.e02edc26.akpm@linux-foundation.org> (raw) In-Reply-To: <20090430215034.4748e615@riellaptop.surriel.com> On Thu, 30 Apr 2009 21:50:34 -0400 Rik van Riel <riel@redhat.com> wrote: > > Which would cause exactly the problem Elladan saw? > > Yes. It was not noticable in the initial split LRU code, > but after we decided to ignore the referenced bit on active > file pages and deactivate pages regardless, it has gotten > exacerbated. > > That change was very good for scalability, so we should not > undo it. However, we do need to put something in place to > protect the working set from streaming IO. > > > > Currently the kernel has no effective code to protect the > > > page cache working set from streaming IO. Elladan's bug > > > report shows that we do need some kind of protection... > > > > Seems to me that reclaim should treat swapcache-backed mapped mages in > > a similar fashion to file-backed mapped pages? > > Swapcache-backed pages are not on the same set of LRUs as > file-backed mapped pages. yup. > Furthermore, there is no streaming IO on the anon LRUs like > there is on the file LRUs. Only the file LRUs need (and want) > use-once replacement, which means that we only need special > protection of the working set for file-backed pages. OK. > When we implement working set protection, we might as well > do it for frequently accessed unmapped pages too. There is > no reason to restrict this protection to mapped pages. Well. Except for empirical observation, which tells us that biasing reclaim to prefer to retain mapped memory produces a better result. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
next prev parent reply other threads:[~2009-05-01 2:59 UTC|newest] Thread overview: 336+ messages / expand[flat|nested] mbox.gz Atom feed top 2009-04-28 4:44 Swappiness vs. mmap() and interactive response Elladan 2009-04-28 5:35 ` KOSAKI Motohiro 2009-04-28 5:35 ` KOSAKI Motohiro 2009-04-28 6:36 ` Elladan 2009-04-28 6:36 ` Elladan 2009-04-28 6:52 ` KOSAKI Motohiro 2009-04-28 6:52 ` KOSAKI Motohiro 2009-04-28 7:26 ` Elladan 2009-04-28 7:26 ` Elladan 2009-04-28 7:44 ` KOSAKI Motohiro 2009-04-28 7:44 ` KOSAKI Motohiro 2009-04-28 7:48 ` Peter Zijlstra 2009-04-28 7:48 ` Peter Zijlstra 2009-04-28 7:58 ` Balbir Singh 2009-04-28 7:58 ` Balbir Singh 2009-04-28 8:11 ` Peter Zijlstra 2009-04-28 8:11 ` Peter Zijlstra 2009-04-28 8:23 ` KAMEZAWA Hiroyuki 2009-04-28 8:23 ` KAMEZAWA Hiroyuki 2009-04-28 8:25 ` Balbir Singh 2009-04-28 8:25 ` Balbir Singh 2009-04-28 8:03 ` KOSAKI Motohiro 2009-04-28 8:03 ` KOSAKI Motohiro 2009-04-28 9:09 ` Wu Fengguang 2009-04-28 9:09 ` Wu Fengguang 2009-04-28 9:26 ` Wu Fengguang 2009-04-28 9:26 ` Wu Fengguang 2009-04-28 12:08 ` Theodore Tso 2009-04-28 12:08 ` Theodore Tso 2009-04-29 5:51 ` KOSAKI Motohiro 2009-04-29 5:51 ` KOSAKI Motohiro 2009-04-29 6:34 ` Andrew Morton 2009-04-29 6:34 ` Andrew Morton 2009-04-29 7:47 ` KOSAKI Motohiro 2009-04-29 7:47 ` KOSAKI Motohiro 2009-04-30 4:14 ` Elladan 2009-04-30 4:14 ` Elladan 2009-04-30 4:43 ` Andrew Morton 2009-04-30 4:43 ` Andrew Morton 2009-04-30 4:55 ` KOSAKI Motohiro 2009-04-30 4:55 ` KOSAKI Motohiro 2009-04-30 4:55 ` Elladan 2009-04-30 4:55 ` Elladan 2009-04-29 7:48 ` KOSAKI Motohiro 2009-04-29 7:48 ` KOSAKI Motohiro 2009-04-30 11:59 ` KOSAKI Motohiro 2009-04-30 11:59 ` KOSAKI Motohiro 2009-04-30 13:46 ` Elladan 2009-04-30 13:46 ` Elladan 2009-05-06 11:04 ` KOSAKI Motohiro 2009-05-06 11:04 ` KOSAKI Motohiro 2009-04-28 15:28 ` Rik van Riel 2009-04-28 23:29 ` [PATCH] vmscan: evict use-once pages first Rik van Riel 2009-04-28 23:29 ` Rik van Riel 2009-04-29 3:36 ` Elladan 2009-04-29 3:36 ` Elladan 2009-04-29 17:06 ` Christoph Hellwig 2009-04-29 17:06 ` Christoph Hellwig 2009-04-29 6:42 ` Peter Zijlstra 2009-04-29 6:42 ` Peter Zijlstra 2009-04-29 13:30 ` Rik van Riel 2009-04-29 13:30 ` Rik van Riel 2009-04-29 15:47 ` [PATCH] vmscan: evict use-once pages first (v2) Rik van Riel 2009-04-29 15:47 ` Rik van Riel 2009-04-29 16:07 ` KOSAKI Motohiro 2009-04-29 16:07 ` KOSAKI Motohiro 2009-04-29 16:18 ` Rik van Riel 2009-04-29 16:18 ` Rik van Riel 2009-04-29 17:14 ` [PATCH] vmscan: evict use-once pages first (v3) Rik van Riel 2009-04-29 17:14 ` Rik van Riel 2009-04-30 0:39 ` KOSAKI Motohiro 2009-04-30 0:39 ` KOSAKI Motohiro 2009-04-30 8:10 ` Johannes Weiner 2009-04-30 8:10 ` Johannes Weiner 2009-05-01 22:32 ` Andrew Morton 2009-05-01 22:32 ` Andrew Morton 2009-05-01 23:05 ` Rik van Riel 2009-05-01 23:05 ` Rik van Riel 2009-05-01 23:25 ` Andrew Morton 2009-05-01 23:25 ` Andrew Morton 2009-05-03 1:28 ` Wu Fengguang 2009-05-03 1:28 ` Wu Fengguang 2009-05-03 1:15 ` Wu Fengguang 2009-05-03 1:15 ` Wu Fengguang 2009-05-03 1:33 ` Rik van Riel 2009-05-03 1:33 ` Rik van Riel 2009-05-03 1:46 ` Wu Fengguang 2009-05-03 1:46 ` Wu Fengguang 2009-04-29 16:10 ` [PATCH] vmscan: evict use-once pages first (v2) Peter Zijlstra 2009-04-29 16:10 ` Peter Zijlstra 2009-04-30 7:20 ` Elladan 2009-04-30 7:20 ` Elladan 2009-04-30 13:08 ` Rik van Riel 2009-04-30 13:08 ` Rik van Riel 2009-04-30 14:00 ` Elladan 2009-04-30 14:00 ` Elladan 2009-05-01 0:45 ` Andrew Morton 2009-05-01 0:45 ` Andrew Morton 2009-05-01 0:59 ` Rik van Riel 2009-05-01 0:59 ` Rik van Riel 2009-05-01 1:13 ` Andrew Morton 2009-05-01 1:13 ` Andrew Morton 2009-05-01 1:50 ` Rik van Riel 2009-05-01 1:50 ` Rik van Riel 2009-05-01 2:54 ` Andrew Morton [this message] 2009-05-01 2:54 ` Andrew Morton 2009-05-01 14:05 ` Rik van Riel 2009-05-01 14:05 ` Rik van Riel 2009-05-01 18:04 ` Ray Lee 2009-05-01 18:04 ` Ray Lee 2009-05-01 19:34 ` Rik van Riel 2009-05-01 19:34 ` Rik van Riel 2009-05-01 19:44 ` Ray Lee 2009-05-01 19:44 ` Ray Lee 2009-05-01 20:08 ` Rik van Riel 2009-05-01 20:08 ` Rik van Riel 2009-05-01 20:17 ` Elladan 2009-05-01 20:17 ` Elladan 2009-05-01 19:35 ` Andrew Morton 2009-05-01 19:35 ` Andrew Morton 2009-05-01 20:05 ` Rik van Riel 2009-05-01 20:05 ` Rik van Riel 2009-05-01 20:45 ` Andrew Morton 2009-05-01 20:45 ` Andrew Morton 2009-05-01 21:46 ` Rik van Riel 2009-05-01 21:46 ` Rik van Riel 2009-05-03 3:15 ` Wu Fengguang 2009-05-03 3:15 ` Wu Fengguang 2009-05-03 3:24 ` Rik van Riel 2009-05-03 3:24 ` Rik van Riel 2009-05-03 3:43 ` Wu Fengguang 2009-05-03 3:43 ` Wu Fengguang 2009-05-04 10:23 ` Peter Zijlstra 2009-05-04 10:23 ` Peter Zijlstra 2009-05-07 12:11 ` [PATCH -mm] vmscan: make mapped executable pages the first class citizen Wu Fengguang 2009-05-07 12:11 ` Wu Fengguang 2009-05-07 13:39 ` Christoph Lameter 2009-05-07 13:39 ` Christoph Lameter 2009-05-07 14:15 ` Peter Zijlstra 2009-05-07 14:15 ` Peter Zijlstra 2009-05-07 14:18 ` Christoph Lameter 2009-05-07 14:18 ` Christoph Lameter 2009-05-07 14:38 ` Peter Zijlstra 2009-05-07 14:38 ` Peter Zijlstra 2009-05-07 15:36 ` Christoph Lameter 2009-05-07 15:36 ` Christoph Lameter 2009-05-07 15:59 ` Rik van Riel 2009-05-07 15:59 ` Rik van Riel 2009-05-07 15:06 ` Rik van Riel 2009-05-07 15:06 ` Rik van Riel 2009-05-07 16:00 ` Lee Schermerhorn 2009-05-07 16:00 ` Lee Schermerhorn 2009-05-07 16:32 ` Christoph Lameter 2009-05-07 16:32 ` Christoph Lameter 2009-05-07 17:11 ` Rik van Riel 2009-05-07 17:11 ` Rik van Riel 2009-05-08 3:40 ` Elladan 2009-05-08 3:40 ` Elladan 2009-05-08 16:04 ` Rik van Riel 2009-05-08 16:04 ` Rik van Riel 2009-05-09 4:04 ` Elladan 2009-05-09 4:04 ` Elladan 2009-05-08 17:18 ` Christoph Lameter 2009-05-08 17:18 ` Christoph Lameter 2009-05-09 10:20 ` KOSAKI Motohiro 2009-05-09 10:20 ` KOSAKI Motohiro 2009-05-08 17:37 ` Alan Cox 2009-05-08 17:37 ` Alan Cox 2009-05-07 15:10 ` Johannes Weiner 2009-05-07 15:10 ` Johannes Weiner 2009-05-07 15:17 ` Peter Zijlstra 2009-05-07 15:17 ` Peter Zijlstra 2009-05-07 15:21 ` Rik van Riel 2009-05-07 15:21 ` Rik van Riel 2009-05-08 3:30 ` Wu Fengguang 2009-05-08 3:30 ` Wu Fengguang 2009-05-08 4:17 ` [RFC][PATCH] vmscan: report vm_flags in page_referenced() Wu Fengguang 2009-05-08 4:17 ` Wu Fengguang 2009-05-08 12:09 ` Minchan Kim 2009-05-08 12:09 ` Minchan Kim 2009-05-08 12:15 ` Wu Fengguang 2009-05-08 12:15 ` Wu Fengguang 2009-05-08 14:01 ` Minchan Kim 2009-05-08 14:01 ` Minchan Kim 2009-05-09 6:56 ` Wu Fengguang 2009-05-09 6:56 ` Wu Fengguang 2009-05-10 23:45 ` Minchan Kim 2009-05-10 23:45 ` Minchan Kim 2009-05-17 11:25 ` Wu Fengguang 2009-05-17 11:25 ` Wu Fengguang 2009-05-07 20:44 ` [PATCH -mm] vmscan: make mapped executable pages the first class citizen Andrew Morton 2009-05-07 20:44 ` Andrew Morton 2009-05-08 8:16 ` Wu Fengguang 2009-05-08 8:16 ` Wu Fengguang 2009-05-08 8:28 ` Wu Fengguang 2009-05-08 8:28 ` Wu Fengguang 2009-05-08 19:58 ` Andrew Morton 2009-05-08 19:58 ` Andrew Morton 2009-05-08 22:00 ` Alan Cox 2009-05-08 22:00 ` Alan Cox 2009-05-08 22:15 ` Andrew Morton 2009-05-08 22:15 ` Andrew Morton 2009-05-08 22:53 ` Elladan 2009-05-08 22:53 ` Elladan 2009-05-08 22:20 ` Rik van Riel 2009-05-08 22:20 ` Rik van Riel 2009-05-10 8:59 ` KOSAKI Motohiro 2009-05-10 8:59 ` KOSAKI Motohiro 2009-05-10 9:07 ` Peter Zijlstra 2009-05-10 9:07 ` Peter Zijlstra 2009-05-10 9:35 ` Wu Fengguang 2009-05-10 9:35 ` Wu Fengguang 2009-05-10 10:06 ` KOSAKI Motohiro 2009-05-10 10:06 ` KOSAKI Motohiro 2009-05-10 9:36 ` KOSAKI Motohiro 2009-05-10 9:36 ` KOSAKI Motohiro 2009-05-10 13:45 ` Alan Cox 2009-05-10 13:45 ` Alan Cox 2009-05-10 13:56 ` KOSAKI Motohiro 2009-05-10 13:56 ` KOSAKI Motohiro 2009-05-10 14:51 ` Rik van Riel 2009-05-10 14:51 ` Rik van Riel 2009-05-10 14:59 ` KOSAKI Motohiro 2009-05-10 14:59 ` KOSAKI Motohiro 2009-05-10 20:13 ` Alan Cox 2009-05-10 20:13 ` Alan Cox 2009-05-10 20:37 ` Rik van Riel 2009-05-10 20:37 ` Rik van Riel 2009-05-10 21:23 ` Arjan van de Ven 2009-05-10 21:23 ` Arjan van de Ven 2009-05-11 10:03 ` Johannes Weiner 2009-05-11 10:03 ` Johannes Weiner 2009-05-10 21:29 ` Alan Cox 2009-05-10 21:29 ` Alan Cox 2009-05-10 9:20 ` Wu Fengguang 2009-05-10 9:20 ` Wu Fengguang 2009-05-10 9:29 ` KOSAKI Motohiro 2009-05-10 9:29 ` KOSAKI Motohiro 2009-05-10 10:03 ` Wu Fengguang 2009-05-10 10:03 ` Wu Fengguang 2009-05-10 10:15 ` KOSAKI Motohiro 2009-05-10 10:15 ` KOSAKI Motohiro 2009-05-10 11:21 ` Wu Fengguang 2009-05-10 11:21 ` Wu Fengguang 2009-05-10 11:39 ` KOSAKI Motohiro 2009-05-10 11:39 ` KOSAKI Motohiro 2009-05-10 11:44 ` Wu Fengguang 2009-05-10 11:44 ` Wu Fengguang 2009-05-10 12:19 ` Peter Zijlstra 2009-05-10 12:19 ` Peter Zijlstra 2009-05-10 12:39 ` KOSAKI Motohiro 2009-05-10 12:39 ` KOSAKI Motohiro 2009-05-10 13:17 ` Peter Zijlstra 2009-05-10 13:17 ` Peter Zijlstra 2009-05-12 2:50 ` Wu Fengguang 2009-05-12 2:50 ` Wu Fengguang 2009-05-12 4:35 ` Wu Fengguang 2009-05-12 4:35 ` Wu Fengguang 2009-05-12 13:20 ` Rik van Riel 2009-05-12 13:20 ` Rik van Riel 2009-05-16 9:26 ` Wu Fengguang 2009-05-16 9:26 ` Wu Fengguang 2009-05-12 2:51 ` [PATCH -mm] vmscan: report vm_flags in page_referenced() Wu Fengguang 2009-05-12 2:51 ` Wu Fengguang 2009-05-12 6:23 ` Peter Zijlstra 2009-05-12 6:23 ` Peter Zijlstra 2009-05-12 6:44 ` Minchan Kim 2009-05-12 6:44 ` Minchan Kim 2009-05-12 11:44 ` Wu Fengguang 2009-05-12 11:44 ` Wu Fengguang 2009-05-12 2:52 ` [PATCH -mm] vmscan: make mapped executable pages the first class citizen Wu Fengguang 2009-05-12 2:52 ` Wu Fengguang 2009-05-12 3:00 ` KOSAKI Motohiro 2009-05-12 3:00 ` KOSAKI Motohiro 2009-05-12 20:54 ` [PATCH -mm] vmscan: protect a fraction of file backed mapped pages from reclaim Christoph Lameter 2009-05-12 20:54 ` Christoph Lameter 2009-05-12 17:06 ` Rik van Riel 2009-05-12 17:06 ` Rik van Riel 2009-05-12 21:20 ` Christoph Lameter 2009-05-12 21:20 ` Christoph Lameter 2009-05-12 17:39 ` Rik van Riel 2009-05-12 17:39 ` Rik van Riel 2009-05-12 22:02 ` Christoph Lameter 2009-05-12 22:02 ` Christoph Lameter 2009-05-12 20:17 ` Rik van Riel 2009-05-12 20:17 ` Rik van Riel 2009-05-12 20:26 ` Christoph Lameter 2009-05-12 20:26 ` Christoph Lameter 2009-05-13 0:45 ` KOSAKI Motohiro 2009-05-13 0:45 ` KOSAKI Motohiro 2009-05-14 20:14 ` Christoph Lameter 2009-05-14 20:14 ` Christoph Lameter 2009-05-14 23:28 ` KOSAKI Motohiro 2009-05-14 23:28 ` KOSAKI Motohiro 2009-05-14 23:42 ` Rik van Riel 2009-05-14 23:42 ` Rik van Riel 2009-05-15 18:09 ` Christoph Lameter 2009-05-15 18:09 ` Christoph Lameter 2009-05-16 8:54 ` Wu Fengguang 2009-05-16 8:54 ` Wu Fengguang 2009-05-12 8:17 ` [PATCH -mm] vmscan: make mapped executable pages the first class citizen Minchan Kim 2009-05-12 8:17 ` Minchan Kim 2009-05-12 2:53 ` [PATCH -mm] vmscan: merge duplicate code in shrink_active_list() Wu Fengguang 2009-05-12 2:53 ` Wu Fengguang 2009-05-12 2:58 ` KOSAKI Motohiro 2009-05-12 2:58 ` KOSAKI Motohiro 2009-05-12 3:03 ` Wu Fengguang 2009-05-12 3:03 ` Wu Fengguang 2009-05-12 7:26 ` Minchan Kim 2009-05-12 7:26 ` Minchan Kim 2009-05-12 11:48 ` Wu Fengguang 2009-05-12 11:48 ` Wu Fengguang 2009-05-12 11:57 ` Minchan Kim 2009-05-12 11:57 ` Minchan Kim 2009-05-12 13:32 ` Rik van Riel 2009-05-12 13:32 ` Rik van Riel 2009-05-16 9:30 ` Wu Fengguang 2009-05-16 9:30 ` Wu Fengguang 2009-05-08 3:02 ` [PATCH -mm] vmscan: make mapped executable pages the first class citizen Wu Fengguang 2009-05-08 3:02 ` Wu Fengguang 2009-05-08 7:30 ` Minchan Kim 2009-05-08 7:30 ` Minchan Kim 2009-05-08 8:09 ` Wu Fengguang 2009-05-08 8:09 ` Wu Fengguang 2009-05-08 9:34 ` Minchan Kim 2009-05-08 9:34 ` Minchan Kim 2009-05-08 14:25 ` Christoph Lameter 2009-05-08 14:25 ` Christoph Lameter 2009-05-08 14:34 ` Rik van Riel 2009-05-08 14:34 ` Rik van Riel 2009-05-08 17:41 ` KOSAKI Motohiro 2009-05-08 17:41 ` KOSAKI Motohiro 2009-05-04 8:04 ` [PATCH] vmscan: evict use-once pages first (v2) Peter Zijlstra 2009-05-04 8:04 ` Peter Zijlstra 2009-05-01 3:09 ` Elladan 2009-05-01 3:09 ` Elladan
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20090430195439.e02edc26.akpm@linux-foundation.org \ --to=akpm@linux-foundation.org \ --cc=elladan@eskimo.com \ --cc=kosaki.motohiro@jp.fujitsu.com \ --cc=linux-kernel@vger.kernel.org \ --cc=linux-mm@kvack.org \ --cc=peterz@infradead.org \ --cc=riel@redhat.com \ --cc=tytso@mit.edu \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.