mbox series

[net-next,V2,00/11] vhost_net TX batching

Message ID 20180912031709.14112-1-jasowang@redhat.com
Headers show
Series vhost_net TX batching | expand

Message

Jason Wang Sept. 12, 2018, 3:16 a.m. UTC
Hi all:

This series tries to batch submitting packets to underlayer socket
through msg_control during sendmsg(). This is done by:

1) Doing userspace copy inside vhost_net
2) Build XDP buff
3) Batch at most 64 (VHOST_NET_BATCH) XDP buffs and submit them once
   through msg_control during sendmsg().
4) Underlayer sockets can use XDP buffs directly when XDP is enalbed,
   or build skb based on XDP buff.

For the packet that can not be built easily with XDP or for the case
that batch submission is hard (e.g sndbuf is limited). We will go for
the previous slow path, passing iov iterator to underlayer socket
through sendmsg() once per packet.

This can help to improve cache utilization and avoid lots of indirect
calls with sendmsg(). It can also co-operate with the batching support
of the underlayer sockets (e.g the case of XDP redirection through
maps).

Testpmd(txonly) in guest shows obvious improvements:

Test                /+pps%
XDP_DROP on TAP     /+44.8%
XDP_REDIRECT on TAP /+29%
macvtap (skb)       /+26%

Netperf TCP_STREAM TX from guest shows obvious improvements on small
packet:

    size/session/+thu%/+normalize%
       64/     1/   +2%/    0%
       64/     2/   +3%/   +1%
       64/     4/   +7%/   +5%
       64/     8/   +8%/   +6%
      256/     1/   +3%/    0%
      256/     2/  +10%/   +7%
      256/     4/  +26%/  +22%
      256/     8/  +27%/  +23%
      512/     1/   +3%/   +2%
      512/     2/  +19%/  +14%
      512/     4/  +43%/  +40%
      512/     8/  +45%/  +41%
     1024/     1/   +4%/    0%
     1024/     2/  +27%/  +21%
     1024/     4/  +38%/  +73%
     1024/     8/  +15%/  +24%
     2048/     1/  +10%/   +7%
     2048/     2/  +16%/  +12%
     2048/     4/    0%/   +2%
     2048/     8/    0%/   +2%
     4096/     1/  +36%/  +60%
     4096/     2/  -11%/  -26%
     4096/     4/    0%/  +14%
     4096/     8/    0%/   +4%
    16384/     1/   -1%/   +5%
    16384/     2/    0%/   +2%
    16384/     4/    0%/   -3%
    16384/     8/    0%/   +4%
    65535/     1/    0%/  +10%
    65535/     2/    0%/   +8%
    65535/     4/    0%/   +1%
    65535/     8/    0%/   +3%

Please review.

Thanks

Changes from V1:

- don't hold page refcnt for XDP_DROP
- release page if build_skb() fails
- introduce a helper to build skb for the path of no batching mode
- rename tun_do_xdp() as tun_xdp_act() and use negative for reporting
  errors instead of a pointer to integer
- introduce a num field in tun_msg_ctl
- introduce a new structure for storing metadata in the head of XDP
  packet
- do not store batche XDP inside vhoet_net_virtqueue

Jason Wang (11):
  net: sock: introduce SOCK_XDP
  tuntap: switch to use XDP_PACKET_HEADROOM
  tuntap: enable bh early during processing XDP
  tuntap: simplify error handling in tun_build_skb()
  tuntap: tweak on the path of skb XDP case in tun_build_skb()
  tuntap: split out XDP logic
  tuntap: move XDP flushing out of tun_do_xdp()
  tun: switch to new type of msg_control
  tuntap: accept an array of XDP buffs through sendmsg()
  tap: accept an array of XDP buffs through sendmsg()
  vhost_net: batch submitting XDP buffers to underlayer sockets

 drivers/net/tap.c      |  88 +++++++++++++-
 drivers/net/tun.c      | 267 ++++++++++++++++++++++++++++++++---------
 drivers/vhost/net.c    | 181 +++++++++++++++++++++++++---
 include/linux/if_tun.h |  14 +++
 include/net/sock.h     |   1 +
 5 files changed, 471 insertions(+), 80 deletions(-)

Comments

David Miller Sept. 13, 2018, 4:28 p.m. UTC | #1
From: Jason Wang <jasowang@redhat.com>
Date: Wed, 12 Sep 2018 11:16:58 +0800

> This series tries to batch submitting packets to underlayer socket
> through msg_control during sendmsg(). This is done by:
 ...

Series applied, thanks Jason.
Michael S. Tsirkin Sept. 13, 2018, 6:02 p.m. UTC | #2
On Thu, Sep 13, 2018 at 09:28:19AM -0700, David Miller wrote:
> From: Jason Wang <jasowang@redhat.com>
> Date: Wed, 12 Sep 2018 11:16:58 +0800
> 
> > This series tries to batch submitting packets to underlayer socket
> > through msg_control during sendmsg(). This is done by:
>  ...
> 
> Series applied, thanks Jason.

Going over it now I don't see a lot to complain about, but I'd
appreciate a bit more time for review in the future. 3 days ok?
David Miller Sept. 13, 2018, 6:05 p.m. UTC | #3
From: "Michael S. Tsirkin" <mst@redhat.com>
Date: Thu, 13 Sep 2018 14:02:13 -0400

> On Thu, Sep 13, 2018 at 09:28:19AM -0700, David Miller wrote:
>> From: Jason Wang <jasowang@redhat.com>
>> Date: Wed, 12 Sep 2018 11:16:58 +0800
>> 
>> > This series tries to batch submitting packets to underlayer socket
>> > through msg_control during sendmsg(). This is done by:
>>  ...
>> 
>> Series applied, thanks Jason.
> 
> Going over it now I don't see a lot to complain about, but I'd
> appreciate a bit more time for review in the future. 3 days ok?

I try to handle every patchwork entry within 48 hours, more
specifically 2 business days.

Given the rate at which patches flow through networking, I think
this is both reasonable and necessary.

One always have the option to send a quick reply to the list saying
"Please wait for my review, I'll get to it in the next day or two."
which I will certainly honor.

Thank you.