Repository navigation

Optimisations (and cleanups) of Linux backend - #68

Merged

bors-servo merged 33 commits into

servo:masterfrom

antrik:optimise-buffers

May 4, 2016

Commits on May 4, 2016
- Linux: Rename `get_maximum_send_size()` -> `get_system_sendbuf_size()`
  antrik
  committed
  e6d57cc
- Add some more tests for sizes around the maximum fragment size
  antrik
  committed
  e4f695e
- Linux: Cleanup: Remove casts between `usize` and `size_t`
  antrik
  committed
  1438628
- Linux: Cleanup: Return `usize` instead of `ssize_t` from `UnixCmsg.recv()`
  antrik
  committed
  46e9f8f
- Linux: Refactor: Move FD array assembly out of `construct_header()`
  antrik
  committed
  d0c3938
- Linux: Cleanup: Use mem::size_of_val() instead of manual calculation
  antrik
  committed
  ca14b46
- Linux: Don't deduct auxiliary data size from main buffer size
  antrik
  committed
  3a81025
- Linux: Cleanup: Turn magic value into `RESERVED_SIZE` symbolic constant
  antrik
  committed
  65d964e
- Linux: Don't over-allocate followup fragment receive buffers
  antrik
  committed
  6d94eed
- Linux: Only deduct 32 bytes from maximum send size reported by kernel
  antrik
  committed
  ef74247
- Linux: Cleanup: Improve size recalculation on ENOBUFS
  antrik
  committed
  04ac943
- Linux: Refactor: Introduce `downsize()` function
  antrik
  committed
  f2b1a02
- Linux: Refactor: Split out `UnixSender::fragment_size()`
  antrik
  committed
  37d9bb7
- Linux: Cleanup: Provide a `get_max_fragment_size()` method
  antrik
  committed
  1e9ac09
- Linux: Cleanup: Make `msghdr.iovec` a `*const`
  antrik
  committed
  e554671
- Linux: Cleanup: Move header construction right before use
  antrik
  committed
  a43aefd
- Linux: Refactor: Turn `construct_header()` into `send_first_fragment()`
  antrik
  committed
  a7a1373
- Linux: Refactor: Split out new `send_followup_fragment()` function
  antrik
  committed
  08dcd10
- Linux: Cleanup: Remove no longer needed `unsafe` block
  antrik
  committed
  06ed0c3
- Add benchmark tests for low-level transfers
  antrik
  committed
  8951014
- Linux: Drop fragment headers in favour of a total length header
  antrik
  committed
  1769906
- Linux: Use separate send buffers for header and main data
  antrik
  committed
  31235f6
- Linux: Don't initialise memory of receive buffer
  antrik
  committed
  4050b26
- Linux: Avoid copying data when receiving followup fragments
  antrik
  committed
  4ef8256
- Linux: Use separate receive buffers for header and main message
  antrik
  committed
  f2536df
- Linux: Don't check maximum send size individually for each channel
  antrik
  committed
  07fbcff
- Linux: Only send a control message if necessary
  antrik
  committed
  6cf99d9
- Linux: Align fragment size
  antrik
  committed
  224a652
- Linux: Drop use of `byteorder`
  antrik
  committed
  02902dd
- Linux: Precise calculation of followup fragment size
  antrik
  committed
  af0223d
- Linux: Check maximum send size up front
  antrik
  committed
  5f27e85
- Linux: Don't over-allocate first fragment receive buffer
  antrik
  committed
  0bda93a
- Linux: Cleanup: Trim `unsafe` blocks in `recv()`
  antrik
  committed
  7c2466e

Navigation Menu

Repository navigation

Optimisations (and cleanups) of Linux backend - #68

Optimisations (and cleanups) of Linux backend#68
bors-servo merged 33 commits into
servo:masterfrom
antrik:optimise-buffers

Commits on May 4, 2016

Linux: Rename `get_maximum_send_size()` -> `get_system_sendbuf_size()`

Add some more tests for sizes around the maximum fragment size

Linux: Cleanup: Remove casts between `usize` and `size_t`

Linux: Cleanup: Return `usize` instead of `ssize_t` from `UnixCmsg.recv()`

Linux: Refactor: Move FD array assembly out of `construct_header()`

Linux: Cleanup: Use mem::size_of_val() instead of manual calculation

Linux: Don't deduct auxiliary data size from main buffer size

Linux: Cleanup: Turn magic value into `RESERVED_SIZE` symbolic constant

Linux: Don't over-allocate followup fragment receive buffers

Linux: Only deduct 32 bytes from maximum send size reported by kernel

Linux: Cleanup: Improve size recalculation on ENOBUFS

Linux: Refactor: Introduce `downsize()` function

Linux: Refactor: Split out `UnixSender::fragment_size()`

Linux: Cleanup: Provide a `get_max_fragment_size()` method

Linux: Cleanup: Make `msghdr.iovec` a `*const`

Linux: Cleanup: Move header construction right before use

Linux: Refactor: Turn `construct_header()` into `send_first_fragment()`

Linux: Refactor: Split out new `send_followup_fragment()` function

Linux: Cleanup: Remove no longer needed `unsafe` block

Add benchmark tests for low-level transfers

Linux: Drop fragment headers in favour of a total length header

Linux: Use separate send buffers for header and main data

Linux: Don't initialise memory of receive buffer

Linux: Avoid copying data when receiving followup fragments

Linux: Use separate receive buffers for header and main message

Linux: Don't check maximum send size individually for each channel

Linux: Only send a control message if necessary

Linux: Align fragment size

Linux: Drop use of `byteorder`

Linux: Precise calculation of followup fragment size

Linux: Check maximum send size up front

Linux: Don't over-allocate first fragment receive buffer

Linux: Cleanup: Trim `unsafe` blocks in `recv()`