From: Anna Schumaker <anna@kernel.org>
To: linux-nfs@vger.kernel.org, trond.myklebust@hammerspace.com
Cc: anna@kernel.org
Subject: [PATCH v2 0/5] NFS: Improvements for the NFSv4.2 READ_PLUS operation
Date: Wed, 13 Jul 2022 15:08:44 -0400 [thread overview]
Message-ID: <20220713190849.615778-1-anna@kernel.org> (raw)
From: Anna Schumaker <Anna.Schumaker@Netapp.com>
Previously, decoding was a one step process that expanded holes as they
were seen in the buffer. This had a few undesireable side effects:
1) An extra READ_PLUS call was often needed to fetch any data shifted
off the end of the buffer when the last two segments are a HOLE
followed by DATA
2) We shifted the entire remaining buffer for each hole, meaning some
segments would get moved multiple times during one decode pass.
These patches attempt to address this by turning READ_PLUS decoding into
a two-step process. First, we build up a list of each segment returned
by the server, then we walk the list in reverse and move each segment
directly to their target offset. This cuts out all the extra copying,
and means we won't lose any data off of the end of the reply.
The results of my performance testing can be found here:
https://wiki.linux-nfs.org/wiki/index.php/Read_Plus_May_2022
Between these patches and the corresponding server patches, I'm seeing a
several second decrease in the amount of time that a READ_PLUS call
takes to complete even on the worst case test where pages alternate
between hole and data segments.
I also optimistically remove the CONFIG_NFS_V4_2_READ_PLUS Kconfig
option now that the known performance and correctness issues have been
resolved, but I would also be fine with changing it to default to 'y' if
there are objections to entirely dropping the option.
Please note that these patches rely on the xdr_stream_move_subsegment()
function added as the first patch of the corresponding server patches.
Changes in V2:
- Rebase on v5.19-rc6
- Update for renaming xdr_buf_move_segment() -> xdr_buf_move_subsegment()
Thoughts?
Anna
Anna Schumaker (5):
SUNRPC: Add a function for directly setting the xdr page len
SUNRPC: Add a function for zeroing out a portion of an xdr_stream
NFS: Replace the READ_PLUS decoding code
SUNRPC: Remove xdr_align_data() and xdr_expand_hole()
NFS: Remove the CONFIG_NFS_V4_2_READ_PLUS Kconfig option
fs/nfs/Kconfig | 9 --
fs/nfs/nfs42xdr.c | 168 +++++++++++++++++++------------------
fs/nfs/nfs4proc.c | 2 +-
include/linux/sunrpc/xdr.h | 5 +-
net/sunrpc/xdr.c | 111 +++++++++++-------------
5 files changed, 140 insertions(+), 155 deletions(-)
--
2.37.0
next reply other threads:[~2022-07-13 19:09 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-13 19:08 Anna Schumaker [this message]
2022-07-13 19:08 ` [PATCH v2 1/5] SUNRPC: Add a function for directly setting the xdr page len Anna Schumaker
2022-07-13 19:08 ` [PATCH v2 2/5] SUNRPC: Add a function for zeroing out a portion of an xdr_stream Anna Schumaker
2022-07-13 19:08 ` [PATCH v2 3/5] NFS: Replace the READ_PLUS decoding code Anna Schumaker
2022-07-13 19:08 ` [PATCH v2 4/5] SUNRPC: Remove xdr_align_data() and xdr_expand_hole() Anna Schumaker
2022-07-13 19:08 ` [PATCH v2 5/5] NFS: Remove the CONFIG_NFS_V4_2_READ_PLUS Kconfig option Anna Schumaker
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=20220713190849.615778-1-anna@kernel.org \
--to=anna@kernel.org \
--cc=linux-nfs@vger.kernel.org \
--cc=trond.myklebust@hammerspace.com \
/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