[dpdk-dev] [RFC 0/6] New sync modes for ring

Konstantin Ananyev konstantin.ananyev at intel.com
Mon Feb 24 12:35:09 CET 2020

Upfront note - that RFC is not a complete patch.
It introduces an ABI breakage, plus it doesn't update ring_elem
code properly, etc.
I plan to deal with all these things in later versions.
Right now I seek an initial feedback about proposed ideas.
Would also ask people to repeat performance tests (see below)
on their platforms to confirm the impact.

More and more customers use(/try to use) DPDK based apps within
overcommitted systems (multiple acttive threads over same pysical cores):
VM, container deployments, etc.
One quite common problem they hit: Lock-Holder-Preemption with rte_ring.
LHP is quite a common problem for spin-based sync primitives
(spin-locks, etc.) on overcommitted systems.
The situation gets much worse when some sort of
fair-locking technique is used (ticket-lock, etc.).
As now not only lock-owner but also lock-waiters scheduling
order matters a lot.
This is a well-known problem for kernel within VMs:
The problem with rte_ring is that while head accusion is sort of
un-fair locking, waiting on tail is very similar to ticket lock schema -
tail has to be updated in particular order.
That makes current rte_ring implementation to perform
really pure on some overcommited scenarios.
While it is probably not possible to completely resolve this problem in
userspace only (without some kernel communication/intervention),
removing fairness in tail update can mitigate it significantly.
So this RFC proposes two new optional ring synchronization modes:
1) Head/Tail Sync (HTS) mode
In that mode enqueue/dequeue operation is fully serialized:
    only one thread at a time is allowed to perform given op.
    As another enhancement provide ability to split enqueue/dequeue
    operation into two phases:
      - enqueue/dequeue start
      - enqueue/dequeue finish
    That allows user to inspect objects in the ring without removing
    them from it (aka MT safe peek).
2) Relaxed Tail Sync (RTS)
The main difference from original MP/MC algorithm is that
tail value is increased not by every thread that finished enqueue/dequeue,
but only by the last one.
That allows threads to avoid spinning on ring tail value,
leaving actual tail value change to the last thread in the update queue.

Test results on IA (see below) show significant improvements
for average enqueue/dequeue op times on overcommitted systems.
For 'classic' DPDK deployments (one thread per core) original MP/MC
algorithm still shows best numbers, though for 64-bit target
RTS numbers are not that far away.
Numbers were produced by ring_stress_*autotest
(first patch in these series).

X86_64 @ Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz
DEQ+ENQ average cycles/obj

                                                MP/MC      HTS     RTS
1thread at 1core(--lcores=6-7)                     8.00       8.15    8.99
2thread at 2core(--lcores=6-8)                     19.14      19.61   20.35
4thread at 4core(--lcores=6-10)                    29.43      29.79   31.82
8thread at 8core(--lcores=6-14)                    110.59     192.81  119.50
16thread at 16core(--lcores=6-22)                  461.03     813.12  495.59
32thread/@32core(--lcores='6-22,55-70')         982.90     1972.38 1160.51

2thread at 1core(--lcores='6,(10-11)@7'            20140.50   23.58   25.14
4thread at 2core(--lcores='6,(10-11)@7,(20-21)@8'  153680.60  76.88   80.05
8thread at 2core(--lcores='6,(10-13)@7,(20-23)@8'  280314.32  294.72  318.79
16thread at 2core(--lcores='6,(10-17)@7,(20-27)@8' 643176.59  1144.02 1175.14
32thread at 2core(--lcores='6,(10-25)@7,(30-45)@8' 4264238.80 4627.48 4892.68

8thread at 2core(--lcores='6,(10-17)@(7,8))'       321085.98  298.59  307.47
16thread at 4core(--lcores='6,(20-35)@(7-10))'     1900705.61 575.35  678.29
32thread at 4core(--lcores='6,(20-51)@(7-10))'     5510445.85 2164.36 2714.12

i686 @ Intel(R) Xeon(R) Platinum 8160 CPU @ 2.10GHz
DEQ+ENQ average cycles/obj

                                                MP/MC      HTS     RTS
1thread at 1core(--lcores=6-7)                     7.85       12.13   11.31
2thread at 2core(--lcores=6-8)                     17.89      24.52   21.86
8thread at 8core(--lcores=6-14)                    32.58      354.20  54.58
32thread/@32core(--lcores='6-22,55-70')         813.77     6072.41 2169.91

2thread at 1core(--lcores='6,(10-11)@7'            16095.00   36.06   34.74
8thread at 2core(--lcores='6,(10-13)@7,(20-23)@8'  1140354.54 346.61  361.57
16thread at 2core(--lcores='6,(10-17)@7,(20-27)@8' 1920417.86 1314.90 1416.65

8thread at 2core(--lcores='6,(10-17)@(7,8))'       594358.61  332.70  357.74
32thread at 4core(--lcores='6,(20-51)@(7-10))'     5319896.86 2836.44 3028.87

Konstantin Ananyev (6):
  test/ring: add contention stress test
  ring: rework ring layout to allow new sync schemes
  ring: introduce RTS ring mode
  test/ring: add contention stress test for RTS ring
  ring: introduce HTS ring mode
  test/ring: add contention stress test for HTS ring

 app/test/Makefile                      |   3 +
 app/test/meson.build                   |   3 +
 app/test/test_pdump.c                  |   6 +-
 app/test/test_ring_hts_stress.c        |  28 ++
 app/test/test_ring_rts_stress.c        |  28 ++
 app/test/test_ring_stress.c            |  27 ++
 app/test/test_ring_stress.h            | 477 +++++++++++++++++++
 lib/librte_pdump/rte_pdump.c           |   2 +-
 lib/librte_port/rte_port_ring.c        |  12 +-
 lib/librte_ring/Makefile               |   4 +-
 lib/librte_ring/meson.build            |   4 +-
 lib/librte_ring/rte_ring.c             |  84 +++-
 lib/librte_ring/rte_ring.h             | 619 +++++++++++++++++++++++--
 lib/librte_ring/rte_ring_elem.h        |   8 +-
 lib/librte_ring/rte_ring_hts_generic.h | 228 +++++++++
 lib/librte_ring/rte_ring_rts_generic.h | 240 ++++++++++
 16 files changed, 1721 insertions(+), 52 deletions(-)
 create mode 100644 app/test/test_ring_hts_stress.c
 create mode 100644 app/test/test_ring_rts_stress.c
 create mode 100644 app/test/test_ring_stress.c
 create mode 100644 app/test/test_ring_stress.h
 create mode 100644 lib/librte_ring/rte_ring_hts_generic.h
 create mode 100644 lib/librte_ring/rte_ring_rts_generic.h


More information about the dev mailing list