| Message ID | 1622684880-39895-1-git-send-email-linyunsheng@huawei.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Some optimization for lockless qdisc | expand |
On Thu, 3 Jun 2021 09:47:57 +0800 Yunsheng Lin wrote: > Patch 1: remove unnecessary seqcount operation. > Patch 2: implement TCQ_F_CAN_BYPASS. > Patch 3: remove qdisc->empty. > > Performance data for pktgen in queue_xmit mode + dummy netdev > with pfifo_fast: > > threads unpatched patched delta > 1 2.60Mpps 3.21Mpps +23% > 2 3.84Mpps 5.56Mpps +44% > 4 5.52Mpps 5.58Mpps +1% > 8 2.77Mpps 2.76Mpps -0.3% > 16 2.24Mpps 2.23Mpps +0.4% > > Performance for IP forward testing: 1.05Mpps increases to > 1.16Mpps, about 10% improvement. Acked-by: Jakub Kicinski <kuba@kernel.org>
On Thu, Jun 03, 2021 at 11:35:48AM -0700, Jakub Kicinski wrote: > On Thu, 3 Jun 2021 09:47:57 +0800 Yunsheng Lin wrote: > > Patch 1: remove unnecessary seqcount operation. > > Patch 2: implement TCQ_F_CAN_BYPASS. > > Patch 3: remove qdisc->empty. > > > > Performance data for pktgen in queue_xmit mode + dummy netdev > > with pfifo_fast: > > > > threads unpatched patched delta > > 1 2.60Mpps 3.21Mpps +23% > > 2 3.84Mpps 5.56Mpps +44% > > 4 5.52Mpps 5.58Mpps +1% > > 8 2.77Mpps 2.76Mpps -0.3% > > 16 2.24Mpps 2.23Mpps +0.4% > > > > Performance for IP forward testing: 1.05Mpps increases to > > 1.16Mpps, about 10% improvement. > > Acked-by: Jakub Kicinski <kuba@kernel.org> Any idea why these patches are deferred in patchwork? https://patchwork.kernel.org/project/netdevbpf/cover/1622684880-39895-1-git-send-email-linyunsheng@huawei.com/
On 2021/6/8 20:53, Vladimir Oltean wrote: > On Thu, Jun 03, 2021 at 11:35:48AM -0700, Jakub Kicinski wrote: >> On Thu, 3 Jun 2021 09:47:57 +0800 Yunsheng Lin wrote: >>> Patch 1: remove unnecessary seqcount operation. >>> Patch 2: implement TCQ_F_CAN_BYPASS. >>> Patch 3: remove qdisc->empty. >>> >>> Performance data for pktgen in queue_xmit mode + dummy netdev >>> with pfifo_fast: >>> >>> threads unpatched patched delta >>> 1 2.60Mpps 3.21Mpps +23% >>> 2 3.84Mpps 5.56Mpps +44% >>> 4 5.52Mpps 5.58Mpps +1% >>> 8 2.77Mpps 2.76Mpps -0.3% >>> 16 2.24Mpps 2.23Mpps +0.4% >>> >>> Performance for IP forward testing: 1.05Mpps increases to >>> 1.16Mpps, about 10% improvement. >> >> Acked-by: Jakub Kicinski <kuba@kernel.org> > > Any idea why these patches are deferred in patchwork? > https://patchwork.kernel.org/project/netdevbpf/cover/1622684880-39895-1-git-send-email-linyunsheng@huawei.com/ I suppose it is a controversial change, which need more time hanging to be reviewed and tested. By the way, I did not pick up your "Tested-by" from previous RFC version because there is some change between those version that deserves a retesting. So it would be good to have a "Tested-by" from you after confirming no out of order happening for this version, thanks.
On Wed, 9 Jun 2021 09:31:39 +0800 Yunsheng Lin wrote: > On 2021/6/8 20:53, Vladimir Oltean wrote: > > On Thu, Jun 03, 2021 at 11:35:48AM -0700, Jakub Kicinski wrote: > >> On Thu, 3 Jun 2021 09:47:57 +0800 Yunsheng Lin wrote: > >>> Patch 1: remove unnecessary seqcount operation. > >>> Patch 2: implement TCQ_F_CAN_BYPASS. > >>> Patch 3: remove qdisc->empty. > >>> > >>> Performance data for pktgen in queue_xmit mode + dummy netdev > >>> with pfifo_fast: > >>> > >>> threads unpatched patched delta > >>> 1 2.60Mpps 3.21Mpps +23% > >>> 2 3.84Mpps 5.56Mpps +44% > >>> 4 5.52Mpps 5.58Mpps +1% > >>> 8 2.77Mpps 2.76Mpps -0.3% > >>> 16 2.24Mpps 2.23Mpps +0.4% > >>> > >>> Performance for IP forward testing: 1.05Mpps increases to > >>> 1.16Mpps, about 10% improvement. > >> > >> Acked-by: Jakub Kicinski <kuba@kernel.org> > > > > Any idea why these patches are deferred in patchwork? > > https://patchwork.kernel.org/project/netdevbpf/cover/1622684880-39895-1-git-send-email-linyunsheng@huawei.com/ > > I suppose it is a controversial change, which need more time > hanging to be reviewed and tested. That'd be my guess also. A review from area experts would be great, perhaps from Cong, John, Michal.. If the review doesn't come by Friday - I'd repost.
On Wed, Jun 09, 2021 at 09:31:39AM +0800, Yunsheng Lin wrote: > By the way, I did not pick up your "Tested-by" from previous > RFC version because there is some change between those version > that deserves a retesting. So it would be good to have a > "Tested-by" from you after confirming no out of order happening > for this version, thanks. Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # flexcan
Patch 1: remove unnecessary seqcount operation. Patch 2: implement TCQ_F_CAN_BYPASS. Patch 3: remove qdisc->empty. Performance data for pktgen in queue_xmit mode + dummy netdev with pfifo_fast: threads unpatched patched delta 1 2.60Mpps 3.21Mpps +23% 2 3.84Mpps 5.56Mpps +44% 4 5.52Mpps 5.58Mpps +1% 8 2.77Mpps 2.76Mpps -0.3% 16 2.24Mpps 2.23Mpps +0.4% Performance for IP forward testing: 1.05Mpps increases to 1.16Mpps, about 10% improvement. V2: Adjust the comment and commit log according to discussion in V1. V1: Drop RFC tag, Add nolock_qdisc_is_empty() and do the qdisc empty checking without the protection of qdisc->seqlock to aviod doing unnecessary spin_trylock() for contention case. RFC v4: Use STATE_MISSED and STATE_DRAINING to indicate non-empty qdisc, and add patch 1 and 3. Yunsheng Lin (3): net: sched: avoid unnecessary seqcount operation for lockless qdisc net: sched: implement TCQ_F_CAN_BYPASS for lockless qdisc net: sched: remove qdisc->empty for lockless qdisc include/net/sch_generic.h | 31 ++++++++++++++++++------------- net/core/dev.c | 27 +++++++++++++++++++++++++-- net/sched/sch_generic.c | 23 ++++++++++++++++------- 3 files changed, 59 insertions(+), 22 deletions(-)