From: Martin Brandenburg <martin@omnibond.com>
To: devel@lists.orangefs.org, linux-fsdevel@vger.kernel.org,
linux-kernel@vger.kernel.org, hubcap@omnibond.com
Cc: Martin Brandenburg <martin@omnibond.com>
Subject: [PATCH 18/19] orangefs: use client-core buffer size to determine writepages count
Date: Sun, 7 Oct 2018 23:27:35 +0000 [thread overview]
Message-ID: <20181007232736.3780-19-martin@omnibond.com> (raw)
In-Reply-To: <20181007232736.3780-1-martin@omnibond.com>
The previous fixed count of 128 was arbitrary.
I see about a 10% performance increase on large block size I/O since
the count is now 1024 (given the default four megabyte client-core
buffer).
Signed-off-by: Martin Brandenburg <martin@omnibond.com>
---
fs/orangefs/inode.c | 27 ++++++++++++++++++++-------
1 file changed, 20 insertions(+), 7 deletions(-)
diff --git a/fs/orangefs/inode.c b/fs/orangefs/inode.c
index 9a41b7d2ce54..20950f3f758a 100644
--- a/fs/orangefs/inode.c
+++ b/fs/orangefs/inode.c
@@ -15,8 +15,6 @@
#include "orangefs-kernel.h"
#include "orangefs-bufmap.h"
-#define ORANGEFS_WRITEPAGES_COUNT 128
-
static int orangefs_writepage_locked(struct page *page,
struct writeback_control *wbc)
{
@@ -237,9 +235,10 @@ struct orangefs_writepages {
size_t len;
kuid_t uid;
kgid_t gid;
- struct page *pages[ORANGEFS_WRITEPAGES_COUNT];
+ int maxpages;
int npages;
- struct bio_vec bv[ORANGEFS_WRITEPAGES_COUNT];
+ struct page **pages;
+ struct bio_vec *bv;
};
static int orangefs_writepages_work(struct orangefs_writepages *ow,
@@ -324,7 +323,7 @@ static int orangefs_writepages_callback(struct page *page,
}
if (!uid_eq(ow->uid, wr->uid) || !gid_eq(ow->gid, wr->gid)) {
orangefs_writepages_work(ow, wbc);
- memset(ow, 0, sizeof *ow);
+ ow->npages = 0;
ret = -1;
goto done;
}
@@ -340,9 +339,9 @@ static int orangefs_writepages_callback(struct page *page,
mapping_set_error(page->mapping, ret);
unlock_page(page);
} else {
- if (ow->npages == ORANGEFS_WRITEPAGES_COUNT) {
+ if (ow->npages == ow->maxpages) {
orangefs_writepages_work(ow, wbc);
- memset(ow, 0, sizeof *ow);
+ ow->npages = 0;
}
}
}
@@ -358,6 +357,18 @@ static int orangefs_writepages(struct address_space *mapping,
ow = kzalloc(sizeof(struct orangefs_writepages), GFP_KERNEL);
if (!ow)
return -ENOMEM;
+ ow->maxpages = orangefs_bufmap_size_query()/PAGE_SIZE;
+ ow->pages = kcalloc(ow->maxpages, sizeof(struct page *), GFP_KERNEL);
+ if (!ow->pages) {
+ kfree(ow);
+ return -ENOMEM;
+ }
+ ow->bv = kcalloc(ow->maxpages, sizeof(struct bio_vec), GFP_KERNEL);
+ if (!ow->bv) {
+ kfree(ow->pages);
+ kfree(ow);
+ return -ENOMEM;
+ }
mutex_lock(&ORANGEFS_SB(mapping->host->i_sb)->writepages_mutex);
blk_start_plug(&plug);
ret = write_cache_pages(mapping, wbc, orangefs_writepages_callback, ow);
@@ -365,6 +376,8 @@ static int orangefs_writepages(struct address_space *mapping,
ret = orangefs_writepages_work(ow, wbc);
blk_finish_plug(&plug);
mutex_unlock(&ORANGEFS_SB(mapping->host->i_sb)->writepages_mutex);
+ kfree(ow->pages);
+ kfree(ow->bv);
kfree(ow);
return ret;
}
--
2.19.0
next prev parent reply other threads:[~2018-10-08 6:37 UTC|newest]
Thread overview: 23+ messages / expand[flat|nested] mbox.gz Atom feed top
2018-10-07 23:27 [PATCH 00/19] [V2] orangefs: page cache Martin Brandenburg
2018-10-07 23:27 ` [PATCH 01/19] orangefs: implement xattr cache Martin Brandenburg
2018-10-07 23:27 ` [PATCH 02/19] orangefs: do not invalidate attributes on inode create Martin Brandenburg
2018-10-07 23:27 ` [PATCH 03/19] orangefs: simplify orangefs_inode_getattr interface Martin Brandenburg
2018-10-07 23:27 ` [PATCH 04/19] orangefs: update attributes rather than relying on server Martin Brandenburg
2018-10-07 23:27 ` [PATCH 05/19] orangefs: hold i_lock during inode_getattr Martin Brandenburg
2018-10-07 23:27 ` [PATCH 06/19] orangefs: set up and use backing_dev_info Martin Brandenburg
2018-10-07 23:27 ` [PATCH 07/19] orangefs: let setattr write to cached inode Martin Brandenburg
2018-10-07 23:27 ` [PATCH 08/19] orangefs: reorganize setattr functions to track attribute changes Martin Brandenburg
2018-10-07 23:27 ` [PATCH 09/19] orangefs: remove orangefs_readpages Martin Brandenburg
2018-10-07 23:27 ` [PATCH 10/19] orangefs: service ops done for writeback are not killable Martin Brandenburg
2018-10-07 23:27 ` [PATCH 11/19] orangefs: migrate to generic_file_read_iter Martin Brandenburg
2018-10-07 23:27 ` [PATCH 12/19] orangefs: implement writepage Martin Brandenburg
2018-10-07 23:27 ` [PATCH 13/19] orangefs: skip inode writeout if nothing to write Martin Brandenburg
2018-10-07 23:27 ` [PATCH 14/19] orangefs: write range tracking Martin Brandenburg
2018-10-09 23:43 ` invalidatepage questions (was Re: [PATCH 14/19] orangefs: write range tracking) martin
2018-10-07 23:27 ` [PATCH 15/19] orangefs: avoid fsync service operation on flush Martin Brandenburg
2018-10-07 23:27 ` [PATCH 16/19] orangefs: use kmem_cache for orangefs_write_request Martin Brandenburg
2018-10-07 23:27 ` [PATCH 17/19] orangefs: implement writepages Martin Brandenburg
2018-10-07 23:27 ` Martin Brandenburg [this message]
2018-10-07 23:27 ` [PATCH 19/19] orangefs: do writepages_work if a single page must be written Martin Brandenburg
2018-10-08 20:43 ` [PATCH 00/19] [V2] orangefs: page cache Mike Marshall
2018-10-08 23:16 ` martin
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=20181007232736.3780-19-martin@omnibond.com \
--to=martin@omnibond.com \
--cc=devel@lists.orangefs.org \
--cc=hubcap@omnibond.com \
--cc=linux-fsdevel@vger.kernel.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).