From: Fengguang Wu <wfg@mail.ustc.edu.cn>
To: Andrew Morton <akpm@osdl.org>
Cc: linux-kernel@vger.kernel.org
Subject: [PATCH 3/6] readahead: context based method: update ra_min
Date: Sat, 06 Jan 2007 15:26:29 +0800 [thread overview]
Message-ID: <368068421.78607@ustc.edu.cn> (raw)
Message-ID: <20070106072729.755581784@mail.ustc.edu.cn> (raw)
In-Reply-To: 20070106072626.911640026@mail.ustc.edu.cn
[-- Attachment #1: readahead-context-based-method-update-ra_min.patch --]
[-- Type: text/plain, Size: 3342 bytes --]
Classify the 4 cases into 2 classes, and assign proper ra_min for them.
Also update comments correspondly.
Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
---
mm/readahead.c | 49 +++++++++++++++++++++++++++++++----------------
1 file changed, 33 insertions(+), 16 deletions(-)
--- linux-2.6.20-rc3-mm1.orig/mm/readahead.c
+++ linux-2.6.20-rc3-mm1/mm/readahead.c
@@ -1285,17 +1285,26 @@ static int
try_context_based_readahead(struct address_space *mapping,
struct file_ra_state *ra,
struct page *page, pgoff_t offset,
- unsigned long ra_min, unsigned long ra_max)
+ unsigned long req_size, unsigned long ra_max)
{
pgoff_t start;
+ unsigned long ra_min;
unsigned long ra_size;
unsigned long la_size;
- /* Check if there is a segment of history pages, and its end index.
+ /*
+ * Check if there is a segment of history pages, and its end index.
* Based on which we decide whether and where to start read-ahead.
- *
- * Case 1: we have a current page.
- * Search forward for a nearby hole.
+ */
+
+ /*
+ * Select a reasonable large initial size for sequential reads.
+ */
+ ra_min = min(req_size * 4, mapping->backing_dev_info->ra_pages0);
+
+ /*
+ * Case s1: we have a current page.
+ * =======> Search forward for a nearby hole.
*/
read_lock_irq(&mapping->tree_lock);
if (page) {
@@ -1308,8 +1317,9 @@ try_context_based_readahead(struct addre
return -1;
}
- /* Case 2: current page is missing; previous page is present.
- * Just do read-ahead from the current index on.
+ /*
+ * Case s2: current page is missing; previous page is present.
+ * =======> Just do read-ahead from the current index on.
* There's clear sign of sequential reading. It can be
* a) seek => read => this read
* b) cache hit read(s) => this read
@@ -1321,26 +1331,33 @@ try_context_based_readahead(struct addre
goto has_history_pages;
}
- /* Case 2x: the same context info as 2.
- * It can be the early stage of semi-sequential reads(interleaved/nfsd),
- * or an ugly random one. So be conservative.
+ /*
+ * Not an obvious sequential read:
+ * select a conservative initial size, plus user prefered agressiveness.
+ */
+ ra_min = min(req_size, MIN_RA_PAGES) +
+ readahead_hit_rate * 8192 / PAGE_CACHE_SIZE;
+
+ /*
+ * Case r1: the same context info as s2, but not that obvious.
+ * =======> The same action as s2, but be conservative.
+ * It can be the early stage of intermixed sequential reads,
+ * or an ugly random one.
*/
if (readahead_hit_rate && __probe_page(mapping, offset - 1)) {
start = offset;
- if (ra_min > 2 * readahead_hit_rate)
- ra_min = 2 * readahead_hit_rate;
goto has_history_pages;
}
- /* Case 3: no current/previous pages;
- * sparse read-ahead is enabled: ok, be aggressive.
- * Check if there's any adjecent history pages.
+ /*
+ * Case r2: no current/previous pages; sparse read-ahead is enabled.
+ * =======> Do sparse read-ahead if there are adjecent history pages.
*/
if (readahead_hit_rate > 1) {
start = radix_tree_scan_data_backward(&mapping->page_tree,
offset, ra_min);
if (start != ULONG_MAX && offset - start < ra_min) {
- ra_min += offset - start;
+ ra_min *= 2;
offset = ++start; /* pretend the request starts here */
goto has_history_pages;
}
--
next prev parent reply other threads:[~2007-01-06 7:27 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
[not found] <20070106072626.911640026@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 0/6] adaptive readahead update Fengguang Wu
[not found] ` <20070106072729.455570443@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 1/6] readahead: sysctl parameters: set readahead_hit_rate=1 Fengguang Wu
[not found] ` <20070106072729.598977589@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 2/6] readahead: min/max sizes: remove get_readahead_bounds() Fengguang Wu
[not found] ` <20070106072729.755581784@mail.ustc.edu.cn>
2007-01-06 7:26 ` Fengguang Wu [this message]
[not found] ` <20070106072729.877602418@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 4/6] readahead: context based method: remove readahead_ratio Fengguang Wu
[not found] ` <20070106072730.019364791@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 5/6] readahead: call scheme: remove get_readahead_bounds() Fengguang Wu
[not found] ` <20070106072730.157093477@mail.ustc.edu.cn>
2007-01-06 7:26 ` [PATCH 6/6] readahead: nfsd case: remove ra_min Fengguang Wu
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=368068421.78607@ustc.edu.cn \
--to=wfg@mail.ustc.edu.cn \
--cc=akpm@osdl.org \
--cc=linux-kernel@vger.kernel.org \
/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: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).