r7800 wifi settings problem Slow Performance

Post new topic   Reply to topic    DD-WRT Forum Index -> Atheros WiSOC based Hardware
Goto page Previous  1, 2
Author Message
tatsuya46
DD-WRT Guru


Joined: 03 Jan 2010
Posts: 7568
Location: YWG, Canada

PostPosted: Fri Dec 11, 2020 20:21    Post subject: Reply with quote
ho1Aetoo wrote:
with the receive queue >1000 the R7800 manages almost 1Gbit

the only question is why the dd-wrt default is set to 120

for most linux distributions 1000 is the default value

I can speak of luck that I have looked in egc's guide
alone I wouldn't have had the idea

now the thread creator would have to test again


i set it to 2000 across all my routers, seems to help need to test more.. do u know how long its been like this?

_________________
LATEST FIRMWARE(S)

BrainSlayer wrote:
we just do it since we do not like any restrictions enforced by stupid cocaine snorting managers

[x86_64] Haswell i3-4150/QCA9984/QCA9882 ------> r55797 std
[QUALCOMM] DIR-862L --------------------------------> r55797 std
▲ ACTIVE / INACTIVE ▼
[QUALCOMM] WNDR4300 v1 --------------------------> r50485 std
[BROADCOM] DIR-860L A1 ----------------------------> r50485 std


Sigh.. why do i exist anyway.. | I love you Anthony.. never forget that.. my other 99% that ill never see again..

Sponsor
blkt
DD-WRT Guru


Joined: 20 Jan 2019
Posts: 5700

PostPosted: Sat Dec 12, 2020 2:44    Post subject: Reply with quote
tatsuya46 wrote:
i set it to 2000 across all my routers, seems to help need to test more.. do u know how long its been like this?

https://svn.dd-wrt.com/changeset/21921
ho1Aetoo
DD-WRT Guru


Joined: 19 Feb 2019
Posts: 3004
Location: Germany

PostPosted: Sat Dec 12, 2020 9:55    Post subject: Reply with quote
tatsuya46 wrote:
ho1Aetoo wrote:
with the receive queue >1000 the R7800 manages almost 1Gbit

the only question is why the dd-wrt default is set to 120

for most linux distributions 1000 is the default value

I can speak of luck that I have looked in egc's guide
alone I wouldn't have had the idea

now the thread creator would have to test again


i set it to 2000 across all my routers, seems to help need to test more.. do u know how long its been like this?


so for me it fixed my VHT80 4x4 and additionally the 100.000 Errors in the WebIF are gone.

after ~16h speed test with ~900Mbit

Code:
ath0      Link encap:Ethernet  HWaddr   
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:466240003 errors:0 dropped:221 overruns:0 frame:0
          TX packets:4495677887 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:30819612005 (28.7 GiB)  TX bytes:6896270105449 (6.2 TiB)


Last edited by ho1Aetoo on Sat Dec 12, 2020 10:03; edited 1 time in total
egc
DD-WRT Guru


Joined: 18 Mar 2014
Posts: 12917
Location: Netherlands

PostPosted: Sat Dec 12, 2020 10:03    Post subject: Reply with quote
That is an excellent result Smile

I have build 44980 for grabs which should have the netdev_max_backlog set to 2000 see:

https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=327477

_________________
Routers:Netgear R7000, R6400v1, R6400v2, EA6900 (XvortexCFE), E2000, E1200v1, WRT54GS v1.
Install guide R6400v2, R6700v3,XR300:https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=316399
Install guide R7800/XR500: https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=320614
Forum Guide Lines (important read):https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=324087
ho1Aetoo
DD-WRT Guru


Joined: 19 Feb 2019
Posts: 3004
Location: Germany

PostPosted: Sun Dec 13, 2020 9:56    Post subject: Reply with quote
v3.0-r44980

VHT160 2x2

R7800 (QCA9984) <---> x86 PC (QCA9984)

same Radio Firmware on both 10.4-ddwrt-9984-fW-13-5881

Code:
echo 120 > /proc/sys/net/core/netdev_max_backlog


1 Stream Download

[ 5] 402.00-403.00 sec 69.4 MBytes 582 Mbits/sec

Code:
echo 2048 > /proc/sys/net/core/netdev_max_backlog


1 Stream Download

[ 5] 34.00-35.00 sec 98.3 MBytes 824 Mbits/sec


2 Stream Download

[ 5] 0.00-60.07 sec 2.78 GBytes 398 Mbits/sec
[ 7] 0.00-60.07 sec 3.06 GBytes 437 Mbits/sec
[SUM] 0.00-60.07 sec 5.84 GBytes 835 Mbits/sec

1 Stream Upload

[ 5] 0.00-30.00 sec 3.20 GBytes 916 Mbits/sec

2 Stream Upload

[ 5] 0.00-60.00 sec 3.29 GBytes 471 Mbits/sec
[ 7] 0.00-60.00 sec 3.28 GBytes 470 Mbits/sec
[SUM] 0.00-60.00 sec 6.57 GBytes 940 Mbits/sec

so the change also speeds up VHT160 a bit...
when VHT160 works correctly as an exception Evil or Very Mad

with a mix of the Kvalo firmware on the PC and the "CT"? Firmware on the router I have not so good results
ho1Aetoo
DD-WRT Guru


Joined: 19 Feb 2019
Posts: 3004
Location: Germany

PostPosted: Sun Jan 10, 2021 10:14    Post subject: Reply with quote
Hello,

so i have tested extensively.

> latency not advantageous. performance yes

So I describe the test setup....

I run an iperf3 test .....

Raspberry 4 8GB --> LAN ---> R7800 --> WLAN --> PC

Code:
[SUM] 69.00-70.00 sec 99.0 MBytes 830 Mbits/sec


Ping on PC (ping 8.8.8.8 -i 0,2)

Code:
64 bytes from 8.8.8.8: icmp_seq=4 ttl=119 time=22.7 ms
64 bytes from 8.8.8.8: icmp_seq=5 ttl=119 time=21.0 ms
64 bytes from 8.8.8.8: icmp_seq=6 ttl=119 time=21.2 ms


Ping on 2nd LAN device (ping 8.8.8.8 )

Code:
64 bytes from 8.8.8.8: seq=748 ttl=119 time=18.807 ms
64 bytes from 8.8.8.8: seq=749 ttl=119 time=16.595 ms
64 bytes from 8.8.8.8: seq=750 ttl=119 time=18.521 ms


If I now give additional load on the WAN (100Mbit download) QoS is enabled

Ping on the PC (ping 8.8.8.8 -i 0,2)

Code:
4 bytes from 8.8.8.8: icmp_seq=1097 ttl=119 time=127 ms
64 bytes from 8.8.8.8: icmp_seq=1098 ttl=119 time=106 ms
64 bytes from 8.8.8.8: icmp_seq=1099 ttl=119 time=132 ms
64 bytes from 8.8.8.8: icmp_seq=1100 ttl=119 time=102 ms
64 bytes from 8.8.8.8: icmp_seq=1101 ttl=119 time=113 ms
64 bytes from 8.8.8.8: icmp_seq=1102 ttl=119 time=152 ms


Ping on 2nd LAN device (ping 8.8.8.8 )

Code:
64 bytes from 8.8.8.8: seq=904 ttl=119 time=100.212 ms
64 bytes from 8.8.8.8: seq=905 ttl=119 time=113.761 ms
64 bytes from 8.8.8.8: seq=906 ttl=119 time=111.623 ms
64 bytes from 8.8.8.8: seq=907 ttl=119 time=110.001 ms
64 bytes from 8.8.8.8: seq=908 ttl=119 time=119.100 ms
64 bytes from 8.8.8.8: seq=909 ttl=119 time=112.685 ms
64 bytes from 8.8.8.8: seq=910 ttl=119 time=128.994 ms


http://www.dslreports.com/speedtest/66927743

http://www.dslreports.com/speedtest/66927755

so the latency doesn't look good at all .....


Code:
echo 128 > /proc/sys/net/core/netdev_max_backlog


Raspberry 4 8GB --> LAN ---> R7800 --> WLAN --> PC

Code:
[SUM] 285.00-286.00 sec 83.8 MBytes 703 Mbits/sec


Ping on PC (ping 8.8.8.8 -i 0,2)

Code:
64 bytes from 8.8.8.8: icmp_seq=5075 ttl=119 time=18.1 ms
64 bytes from 8.8.8.8: icmp_seq=5076 ttl=119 time=19.8 ms
64 bytes from 8.8.8.8: icmp_seq=5077 ttl=119 time=18.0 ms
64 bytes from 8.8.8.8: icmp_seq=5078 ttl=119 time=21.5 ms


Ping on 2nd LAN device (ping 8.8.8.8 )

Code:
64 bytes from 8.8.8.8: seq=1811 ttl=119 time=16.552 ms
64 bytes from 8.8.8.8: seq=1812 ttl=119 time=15.664 ms
64 bytes from 8.8.8.8: seq=1813 ttl=119 time=19.921 ms


If I now give additional load on the WAN (100Mbit download) QoS is enabled

Ping on the PC (ping 8.8.8.8 -i 0,2)

Code:
64 bytes from 8.8.8.8: icmp_seq=5721 ttl=119 time=25.7 ms
64 bytes from 8.8.8.8: icmp_seq=5722 ttl=119 time=23.0 ms
64 bytes from 8.8.8.8: icmp_seq=5723 ttl=119 time=25.2 ms


Ping on 2nd LAN device (ping 8.8.8.8 )

Code:
64 bytes from 8.8.8.8: seq=1959 ttl=119 time=20.904 ms
64 bytes from 8.8.8.8: seq=1960 ttl=119 time=22.529 ms
64 bytes from 8.8.8.8: seq=1961 ttl=119 time=19.710 ms


http://www.dslreports.com/speedtest/66927785

top bufferbloat ....

so the bufferbloat tests are from the PC via WLAN while parallel via WLAN the iperf3 test is running on the LAN

i also tested some more...

netdev_max_backlog = 128 (low latency like with cable)

netdev_max_backlog = 256 (good compromise between latency and throughput)

Code:
[SUM] 941.00-942.00 sec 95.9 MBytes 804 Mbits/sec


http://www.dslreports.com/speedtest/66929214
http://www.dslreports.com/speedtest/66929263
http://www.dslreports.com/speedtest/66929242

netdev_max_backlog = 2048 (well you see above the latency under full load is ~10x higher)


so if you want a low latency the low values are better
for more throughput higher values are better

(now you can set everything in WebIF)


Last edited by ho1Aetoo on Sun Jan 10, 2021 13:17; edited 2 times in total
egc
DD-WRT Guru


Joined: 18 Mar 2014
Posts: 12917
Location: Netherlands

PostPosted: Sun Jan 10, 2021 12:23    Post subject: Reply with quote
Very interesting, excellent work, thanks!
_________________
Routers:Netgear R7000, R6400v1, R6400v2, EA6900 (XvortexCFE), E2000, E1200v1, WRT54GS v1.
Install guide R6400v2, R6700v3,XR300:https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=316399
Install guide R7800/XR500: https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=320614
Forum Guide Lines (important read):https://forum.dd-wrt.com/phpBB2/viewtopic.php?t=324087
tatsuya46
DD-WRT Guru


Joined: 03 Jan 2010
Posts: 7568
Location: YWG, Canada

PostPosted: Tue Jan 12, 2021 2:06    Post subject: Reply with quote
cpu usages? current 2048 isnt affecting my latency on anything wan or lan, but this is on x86.. it just increases throughput and stops packet drops seemingly for "free".
_________________
LATEST FIRMWARE(S)

BrainSlayer wrote:
we just do it since we do not like any restrictions enforced by stupid cocaine snorting managers

[x86_64] Haswell i3-4150/QCA9984/QCA9882 ------> r55797 std
[QUALCOMM] DIR-862L --------------------------------> r55797 std
▲ ACTIVE / INACTIVE ▼
[QUALCOMM] WNDR4300 v1 --------------------------> r50485 std
[BROADCOM] DIR-860L A1 ----------------------------> r50485 std


Sigh.. why do i exist anyway.. | I love you Anthony.. never forget that.. my other 99% that ill never see again..

ho1Aetoo
DD-WRT Guru


Joined: 19 Feb 2019
Posts: 3004
Location: Germany

PostPosted: Tue Jan 12, 2021 7:56    Post subject: Reply with quote
well the cpu load is there about 100% (is not difficult with the r7800 Smile )
netdev_max_backlog is a buffer that fills up when packets arrive faster than the kernel can process them

but I can compare it later exactly
ho1Aetoo
DD-WRT Guru


Joined: 19 Feb 2019
Posts: 3004
Location: Germany

PostPosted: Tue Jan 12, 2021 13:45    Post subject: Reply with quote
tatsuya46

Code:
cat /proc/sys/net/core/netdev_max_backlog
128


iperf3 + fast.com

Code:
CPU:  0.3% usr 21.7% sys  0.0% nic  2.6% idle  0.0% io  0.0% irq 75.1% sirq
Load average: 3.49 2.42 1.12 5/93 21811


load is ~97-98%

Code:
cat /proc/sys/net/core/netdev_max_backlog
2048


iperf3 + fast.com

Code:
CPU:  0.0% usr 23.8% sys  0.0% nic  0.0% idle  0.0% io  0.0% irq 76.0% sirq


load ~100%

the machine clearly has too little cpu power Very Happy

under full load the buffer fills up and the packets stay a little longer in the cache before they are processed, this increases the latency

a trick is to keep the buffer as small as possible, then no long queues are produced, but then masses of packages are dropped

let's see when my x86 will be ready
i know that the r7800 has too little CPU power
with pure WLAN/LAN transfer I already manage nearly 100% CPU load (there is no load on the WAN)
tatsuya46
DD-WRT Guru


Joined: 03 Jan 2010
Posts: 7568
Location: YWG, Canada

PostPosted: Tue Jan 12, 2021 20:12    Post subject: Reply with quote
probably for cpu limited systems lower may be better, but when cpu isnt saturated, the higher default works better with no latency effect
_________________
LATEST FIRMWARE(S)

BrainSlayer wrote:
we just do it since we do not like any restrictions enforced by stupid cocaine snorting managers

[x86_64] Haswell i3-4150/QCA9984/QCA9882 ------> r55797 std
[QUALCOMM] DIR-862L --------------------------------> r55797 std
▲ ACTIVE / INACTIVE ▼
[QUALCOMM] WNDR4300 v1 --------------------------> r50485 std
[BROADCOM] DIR-860L A1 ----------------------------> r50485 std


Sigh.. why do i exist anyway.. | I love you Anthony.. never forget that.. my other 99% that ill never see again..

a2issa
DD-WRT Novice


Joined: 24 May 2023
Posts: 1

PostPosted: Fri May 26, 2023 22:50    Post subject: Reply with quote
tatsuya46 wrote:
probably for cpu limited systems lower may be better, but when cpu isnt saturated, the higher default works better with no latency effect


Hi

So How do you achieve the 900MB download speed on the Netgear R7800? could you please list all the steps required for that>

Many thanks to all
Goto page Previous  1, 2 Display posts from previous:    Page 2 of 2
Post new topic   Reply to topic    DD-WRT Forum Index -> Atheros WiSOC based Hardware All times are GMT

Navigation

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You cannot download files in this forum