mbox series

[v3,00/10] NFS: Add support for the v4.2 READ_PLUS operation

Message ID 20200803170013.1348350-1-Anna.Schumaker@Netapp.com (mailing list archive)
Headers show
Series NFS: Add support for the v4.2 READ_PLUS operation | expand

Message

Anna Schumaker Aug. 3, 2020, 5 p.m. UTC
From: Anna Schumaker <Anna.Schumaker@Netapp.com>

These patches add client support for the READ_PLUS operation, which
breaks read requests into several "data" and "hole" segments when
replying to the client.

Here are the results of some performance tests I ran on my own virtual
machines, which does still have the slow SEEK_HOLE/SEEK_DATA behavior.
A recent minimum-compiler-version patch makes installing newer kernels
difficult on the lab machines I have access to, but I'll still try to
collect that data since it seemed promising during my last posting
(https://www.spinics.net/lists/linux-nfs/msg76488.html).

I tested by reading various 2G files from a few different underlying
filesystems and across several NFS versions. I used the `vmtouch` utility
to make sure files were only cached when we wanted them to be. In addition
to 100% data and 100% hole cases, I also tested with files that alternate
between data and hole segments. These files have either 4K, 8K, 16K, or 32K
segment sizes and start with either data or hole segments. So the file
mixed-4d has a 4K segment size beginning with a data segment, but mixed-32h
has 32K segments beginning with a hole. The units are in seconds, with the
first number for each NFS version being the uncached read time and the second
number is for when the file is cached on the server.


          |         v3        |        v4.0       |        v4.1       |        v4.2       |
ext4      | uncached:  cached | uncached:  cached | uncached:  cached | uncached:  cached |
----------|-------------------|-------------------|-------------------|-------------------|
data      |   2.697 :   1.365 |   2.539 :   1.482 |   2.246 :   1.595 |   2.516 :   1.716 |
hole      |   2.389 :   1.480 |   1.692 :   1.444 |   1.904 :   1.539 |   1.042 :   1.005 |
mixed-4d  |   2.426 :   1.448 |   2.468 :   1.480 |   2.258 :   1.731 |   2.408 :   1.674 |
mixed-8d  |   2.446 :   1.592 |   2.497 :   1.523 |   2.240 :   1.736 |   2.179 :   1.547 |
mixed-16d |   2.608 :   1.575 |   2.446 :   1.582 |   2.330 :   1.670 |   2.192 :   1.488 |
mixed-32d |   2.544 :   1.587 |   2.164 :   1.499 |   2.076 :   1.452 |   1.982 :   1.346 |
mixed-4h  |   2.591 :   1.486 |   2.233 :   1.503 |   2.190 :   1.520 |   3.679 :   1.815 |
mixed-8h  |   2.498 :   1.547 |   2.168 :   1.510 |   2.098 :   1.511 |   2.815 :   1.484 |
mixed-16h |   2.480 :   1.664 |   2.152 :   1.597 |   2.096 :   1.537 |   2.288 :   1.580 |
mixed-32h |   2.520 :   1.467 |   2.171 :   1.496 |   2.246 :   1.593 |   2.066 :   1.389 |

          |         v3        |        v4.0       |        v4.1       |        v4.2       |
xfs       | uncached:  cached | uncached:  cached | uncached:  cached | uncached:  cached |
----------|-------------------|-------------------|-------------------|-------------------|
data      |   2.074 :   1.353 |   2.129 :   1.489 |   2.198 :   1.564 |   2.579 :   1.719 |
hole      |   1.714 :   1.430 |   1.647 :   1.440 |   1.748 :   1.464 |   1.019 :   1.028 |
mixed-4d  |   2.699 :   1.561 |   2.782 :   1.657 |   2.800 :   1.619 |   2.848 :   2.166 |
mixed-8d  |   2.204 :   1.540 |   2.346 :   1.595 |   2.356 :   1.589 |   2.335 :   1.809 |
mixed-16d |   2.034 :   1.445 |   2.212 :   1.561 |   2.172 :   1.546 |   2.127 :   1.658 |
mixed-32d |   1.982 :   1.480 |   2.135 :   1.544 |   2.136 :   1.565 |   2.024 :   1.555 |
mixed-4h  |   2.600 :   1.549 |   2.790 :   1.660 |   2.745 :   1.634 |   8.949 :   2.529 |
mixed-8h  |   2.283 :   1.555 |   2.414 :   1.605 |   2.373 :   1.607 |   5.626 :   2.015 |
mixed-16h |   2.115 :   1.512 |   2.247 :   1.593 |   2.217 :   1.608 |   3.740 :   1.803 |
mixed-32h |   2.069 :   1.499 |   2.212 :   1.582 |   2.235 :   1.631 |   2.819 :   1.542 |

          |         v3        |        v4.0       |        v4.1       |        v4.2       |
btrfs     | uncached:  cached | uncached:  cached | uncached:  cached | uncached:  cached |
----------|-------------------|-------------------|-------------------|-------------------|
data      |   2.417 :   1.317 |   2.095 :   1.445 |   2.145 :   1.523 |   2.615 :   1.713 |
hole      |   2.107 :   1.483 |   2.121 :   1.496 |   2.106 :   1.461 |   1.011 :   1.061 |
mixed-4d  |   2.348 :   1.471 |   2.370 :   1.523 |   2.379 :   1.499 |   3.028 : 225.812 |
mixed-8d  |   2.227 :   1.476 |   2.231 :   1.467 |   2.272 :   1.529 |   2.723 :  36.179 |
mixed-16d |   2.175 :   1.460 |   2.208 :   1.457 |   2.200 :   1.464 |   2.526 :  10.371 |
mixed-32d |   2.148 :   1.501 |   2.191 :   1.468 |   2.167 :   1.471 |   2.455 :   3.367 |
mixed-4h  |   2.362 :   1.561 |   2.387 :   1.513 |   2.352 :   1.536 |   5.935 :  41.494 |
mixed-8h  |   2.241 :   1.477 |   2.251 :   1.503 |   2.256 :   1.496 |   3.672 :  10.261 |
mixed-16h |   2.238 :   1.477 |   2.188 :   1.496 |   2.183 :   1.503 |   2.955 :   3.809 |
mixed-32h |   2.146 :   1.490 |   2.135 :   1.523 |   2.157 :   1.557 |   2.327 :   2.088 |


Anna Schumaker (10):
  SUNRPC: Split out a function for setting current page
  SUNRPC: Implement a xdr_page_pos() function
  NFS: Use xdr_page_pos() in NFSv4 decode_getacl()
  NFS: Add READ_PLUS data segment support
  SUNRPC: Split out xdr_realign_pages() from xdr_align_pages()
  SUNRPC: Split out _shift_data_right_tail()
  SUNRPC: Add the ability to expand holes in data pages
  NFS: Add READ_PLUS hole segment decoding
  SUNRPC: Add an xdr_align_data() function
  NFS: Decode a full READ_PLUS reply

 fs/nfs/nfs42xdr.c          | 163 ++++++++++++++++++++
 fs/nfs/nfs4proc.c          |  43 +++++-
 fs/nfs/nfs4xdr.c           |   7 +-
 include/linux/nfs4.h       |   2 +-
 include/linux/nfs_fs_sb.h  |   1 +
 include/linux/nfs_xdr.h    |   2 +-
 include/linux/sunrpc/xdr.h |   5 +
 net/sunrpc/xdr.c           | 308 ++++++++++++++++++++++++++++++++-----
 8 files changed, 482 insertions(+), 49 deletions(-)