the machine is usually ~20 feet away from the router, through multiple
walls and a floor. i moved it last night so it was 6 feet away with LOS,
and the results from that are very interesting:

Linux brix 4.15.0-36-generic #39~16.04.1-Ubuntu SMP Tue Sep 25 08:59:23 UTC 
2018 x86_64 x86_64 x86_64 GNU/Linux
          Current Frequency:2.452 GHz (Channel 9)
          Link Quality=70/70  Signal level=-30 dBm  
Sat 20-Oct-18 17:23
    429,840,384 100%   10.91MB/s    0:00:37 (xfr#1, to-chk=0/1)
sent 429,945,420 bytes  received 35 bytes  11,167,414.42 bytes/sec

and under the same ideal conditions:

Linux brix 4.15.0-32-generic #35~16.04.1-Ubuntu SMP Fri Aug 10 21:54:34 UTC 
2018 x86_64 x86_64 x86_64 GNU/Linux
          Current Frequency:2.452 GHz (Channel 9)
          Link Quality=70/70  Signal level=-32 dBm  
Sat 20-Oct-18 17:29
    429,840,384 100%    9.10MB/s    0:00:45 (xfr#1, to-chk=0/1)
sent 429,945,420 bytes  received 35 bytes  9,449,350.66 bytes/sec

so the newer kernel is MUCH better in that situation, whereas the old kernel 
essentially performs exactly the same.
the machine has a fairly weak cpu, and basically has one core pegged with 
iowait during these transfers, so optimisations (either in general for meltdown 
etc, or the stack, or in the driver specifically) could certainly account for 
that sort of speedup.

however, that only further highlights just how bad this regression is,
because the driver is now nearly 20% faster in the abstract but still
massively slower at range despite all that improvement.

it also explains why nobody would notice the regression during
development.

i'm happy to test proposed fixes or provide more information, but i
don't think there's anything more i can do at my end until somebody else
steps in.

@joseph - do you have a tracking reference for upstream yet?

TIA

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux in Ubuntu.
https://bugs.launchpad.net/bugs/1795116

Title:
  large performance regression (~20-40%) in wifi with 4.15.0-33 and
  later

Status in linux package in Ubuntu:
  Confirmed

Bug description:
  16.04 install using the HWE stack. after several weeks of uptime on
  -32, an update to -34 showed a major drop in wifi throughput,
  dependent solely on the kernel chosen:

  $  uname -a && ./wifibench.sh 
  Linux brix 4.15.0-32-generic #35~16.04.1-Ubuntu SMP Fri Aug 10 21:54:34 UTC 
2018 x86_64 x86_64 x86_64 GNU/Linux
  Tue 25-Sep-18 06:07
  PlatformSDK-SVR2003R2.iso
      429,840,384 100%    8.41MB/s    0:00:48 (xfr#1, to-chk=0/1)
  sent 429,945,420 bytes  received 35 bytes  8,685,766.77 bytes/sec

  $  uname -a && ./wifibench.sh 
  Linux brix 4.15.0-34-generic #37~16.04.1-Ubuntu SMP Tue Aug 28 10:44:06 UTC 
2018 x86_64 x86_64 x86_64 GNU/Linux
  Tue 25-Sep-18 06:14
  PlatformSDK-SVR2003R2.iso
      429,840,384 100%    4.83MB/s    0:01:24 (xfr#1, to-chk=0/1)
  sent 429,945,420 bytes  received 35 bytes  5,028,601.81 bytes/sec

  (the script is a simple rsync from a NAS. nothing in the NAS, or the
  router, or the physical positions of the devices or etc etc has
  changed, let alone in those 7 minutes).

  there is no interference, no other devices connected, etc.
  prior to the transfer, the link quality is 62-64 and the signal similarly 
weaker because of power saving, but once packets are in flight it looks perfect:

  $  iwconfig 
  wlan0     IEEE 802.11  ESSID:"****"
            Mode:Managed  Frequency:2.452 GHz  Access Point: ****   
            Bit Rate=150 Mb/s   Tx-Power=20 dBm   
            Retry short limit:7   RTS thr=2347 B   Fragment thr:off
            Power Management:on
            Link Quality=70/70  Signal level=10 dBm  
            Rx invalid nwid:0  Rx invalid crypt:0  Rx invalid frag:0
            Tx excessive retries:0  Invalid misc:15   Missed beacon:0

  the client is a J1900-based (baytrail) machine with a RTL8723BE wifi module. 
average performance while on -32 was consistently 70-80 Mb/s, over a period of 
several weeks. with -33 and later, it's generally 50-65.
  (that machine has no access to launchpad. i'll try apport-cli over the 
weekend).

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1795116/+subscriptions

-- 
Mailing list: https://launchpad.net/~kernel-packages
Post to     : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help   : https://help.launchpad.net/ListHelp

Reply via email to