Archer C7 WAN/LAN throughput decreased 30% since LEDE 17.01

For Developers
1

I upgraded My old TP-Link Archer C7 v2 release by release, measuring WAN/LAN and LAN/WAN wired
throughput for each release. Results are below.

I did the same exercise on TP-Link WDR3600 a while ago: WAN LAN routing performance from 17.01 to 21.02 and the odd thing is that while WDR3600 throughput has increased over releases the Archer C7 throughput has decreased! Why is that?

                         Pct  WtoL  LtoW  WtoL     LtoW
Release                                   refresh  refresh
--------------------------------------------------------------------
LEDE    17.01.5          100   451   591   423      542
OpenWRT 18.06.9           75   339   417   224      265
OpenWRT 19.07.7        <<< failed to build due to: libubus20191227 >>>
OpenWRT 21.02.0           73   328   289   228      189
21.02 + SW Flow offload  158   714   931   522      681

Device was tested with factory default settings. All numbers are in Mbps (1e6 bits per second).

Columns are:

  • Pct: WAN to LAN throughput percentual ratio to LEDE 17.01
  • WtoL: WAN to LAN throughput
  • LtoW: WAN to LAN throughput
  • WtoL refresh: WAN to LAN throughput (on realtime graph traffic web page with refresh)
  • LtoW refresh: WAN to LAN throughput (on realtime graph traffic web page with refresh)

I wasn't able to build OpenWRT 19.07.7 because the image-builder failed with
ERROR: Cannot satisfy the following dependencies for luci: libubus20191227

Furthermore, computing realtime traffic graph throughput impact (from WtoL / WtoL refresh),
shows that the OpenWRT realtime traffic graph heavily impacts the performance:
dropping to approx. 70% of max in all releases except LEDE 17.01
which only drops to 94% of max.

I used the python script mmeisner/sshopenwrt: SSH OpenWRT Control Automatron
to setup and execute the throughput test.
Details of the test are here: Archer-C7-v2-thruput.txt

2 Likes
2

It'll get faster with the next version. A transition to a new switch driver is planned, which increases speed. More info: https://github.com/openwrt/openwrt/pull/4622

3

Try removing ipv6 by adding to /etc/sysctl.conf

net.ipv6.conf.all.disable_ipv6=1
net.ipv6.conf.default.disable_ipv6=1
net.ipv6.conf.lo.disable_ipv6=1

also:

opkg remove ip6tables
opkg remove kmod-ip6tables
opkg remove kmod-nf-ipt6
opkg remove kmod-nf-reject6
opkg remove odhcp6c
opkg remove odhcpd-ipv6only

And replace the ath10k-ct driver and firmware with non-ct ones:

opkg remove ath10k-firmware-qca988x-ct kmod-ath10k-ct
opkg update && opkg install ath10k-firmware-qca988x kmod-ath10k

Hope it helps.

With Metta

4

Yeah, the new DSA driver looks promising but that hardly explains the speed decrease from LEDE 17.01 to today. And why has WDR3600 speed increased?

5

Thanks for your suggestions
Although it's not so much that I am in desperate need of speed (I have fiber and thus using RPi4 router and this old Archer as an AP) but I was just wondering WHY LEDE was faster.
LEDE also had IPv6
What are -ct and non-ct drivers? Sounds WIFI related which was not part of the test at all.

6

Hello,

here are the reference topics for further reading:

With Metta

1 Like
7

Uh.... he said these are wired tests. No wifi.

8

So, this is concerning, and worth looking into more. It needs to be verified, first off. I'm very interested, since I am a C7 owner.

Sadly, my working C7 is doing AP duty for the household internet, not a lot of opportunity to take it off line for serious testing. I'll look at possibly sneaking some in. Anyone else, with the hardware, and willingness to swap FW versions around and run tests? Also, what would be good is trying these tests on different routers, to see if it's a wider issue or not.

9

pretty sure these issues are common among all ath79 devices.

Testing different firmware versions is fairly pointless. If you want to test something: https://github.com/openwrt/openwrt/pull/4622 . That's going to be the future. HNAT is probably going to come in the future with that.

10

But it does seem clear that there is a large speed difference between at least 17.x.x and 21.x.x era versions.
What explains that?

And, since it changes in a negative direction, it should be looked into. IMHO.