[dpdk-dev] [PATCH v9 00/18] distributor lib performance enhancements

David Hunt david.hunt at intel.com
Mon Mar 6 10:10:15 CET 2017

This patch aims to improve the throughput of the distributor library.

It uses a similar handshake mechanism to the previous version of
the library, in that bits are used to indicate when packets are ready
to be sent to a worker and ready to be returned from a worker. One main
difference is that instead of sending one packet in a cache line, it makes
use of the 7 free spaces in the same cache line in order to send up to
8 packets at a time to/from a worker.

The flow matching algorithm has had significant re-work, and now keeps an
array of inflight flows and an array of backlog flows, and matches incoming
flows to the inflight/backlog flows of all workers so that flow pinning to
workers can be maintained.

The Flow Match algorithm has both scalar and a vector versions, and a
function pointer is used to select the post appropriate function at run time,
depending on the presence of the SSE2 cpu flag. On non-x86 platforms, the
the scalar match function is selected, which should still gives a good boost
in performance over the non-burst API.

v9 changes:
   * fixed symbol versioning so it will compile on CentOS and RedHat

v8 changes:
   * Changed the patch set to have a more logical order order of
     the changes, but the end result is basically the same.
   * Fixed broken shared library build.
   * Split down the updates to example app more
   * No longer changes the test app and sample app to use a temporary
   * No longer temporarily re-names the functions in the
     version.map file.

v7 changes:
   * Reorganised patch so there's a more natural progression in the
     changes, and divided them down into easier to review chunks.
   * Previous versions of this patch set were effectively two APIs.
     We now have a single API. Legacy functionality can
     be used by by using the rte_distributor_create API call with the
     RTE_DISTRIBUTOR_SINGLE flag when creating a distributor instance.
   * Added symbol versioning for old API so that ABI is preserved.

v6 changes:
   * Fixed intermittent segfault where num pkts not divisible
     by BURST_SIZE
   * Cleanup due to review comments on mailing list
   * Renamed _priv.h to _private.h.

v5 changes:
   * Removed some un-needed code around retries in worker API calls
   * Cleanup due to review comments on mailing list
   * Cleanup of non-x86 platform compilation, fallback to scalar match

v4 changes:
   * fixed issue building shared libraries

v3 changes:
  * Addressed mailing list review comments
  * Test code removal
  * Split out SSE match into separate file to facilitate NEON addition
  * Cleaned up conditional compilation flags for SSE2
  * Addressed c99 style compilation errors
  * rebased on latest head (Jan 2 2017, Happy New Year to all)

v2 changes:
  * Created a common distributor_priv.h header file with common
    definitions and structures.
  * Added a scalar version so it can be built and used on machines without
    sse2 instruction set
  * Added unit autotests
  * Added perf autotest

   Apps must now work in bursts, as up to 8 are given to a worker at a time
   For performance in matching, Flow ID's are 15-bits
   If 32 bits Flow IDs are required, use the packet-at-a-time (SINGLE)

Performance Gains
   2.2GHz Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz
   2 x XL710 40GbE NICS to 2 x 40Gbps traffic generator channels 64b packets
   separate cores for rx, tx, distributor
    1 worker  - up to 4.8x
    4 workers - up to 2.9x
    8 workers - up to 1.8x
   12 workers - up to 2.1x
   16 workers - up to 1.8x

[01/18] lib: rename legacy distributor lib files
[02/18] lib: create private header file
[03/18] lib: add new burst oriented distributor structs
[04/18] lib: add new distributor code
[05/18] lib: add SIMD flow matching to distributor
[06/18] test/distributor: extra params for autotests
[07/18] lib: switch distributor over to new API
[08/18] lib: make v20 header file private
[09/18] lib: add symbol versioning to distributor
[10/18] test: test single and burst distributor API
[11/18] test: add perf test for distributor burst mode
[12/18] examples/distributor: allow for extra stats
[13/18] sample: distributor: wait for ports to come up
[14/18] examples/distributor: give distributor a core
[15/18] examples/distributor: limit number of Tx rings
[16/18] examples/distributor: give Rx thread a core
[17/18] doc: distributor library changes for new burst API
[18/18] maintainers: add to distributor lib maintainers

More information about the dev mailing list