18.06.1 loading down Archer C7, slow network speeds?

Anyone have and idea what's going on here? I just bumped my C7 from 17.01.4 to 18.06.1, and not only am I seeing much slower SQM speeds, I'm not even able to do the speeds that I used to without SQM .

I'm also seeing higher background loads, 10 or more processes constantly at 1% or so where most were only occasionally bumping to 1%, the router seems to be doing a lot more work than before. I see higher peak load numbers, though they do settle down when things are quiet.

I used to be able to see 300-340mbit ingress, with no SQM and on eth on my 300/30 rated cable, without running out of idle %. Used to be able to see 200-240mbit or so over the AC wifi.

Now, I only hit 150-170mbit max, with the idle at 0%, over eth and with SQM off. Obviously, being hardware limited now. But why?

I'm only on DSL, so line limits my throughput, but I did notice the higher load values and CPU usage when looking at top.
When running luci-bwc for the traffic graphs in the UI, top shows very high loads, much higher than 17.x.

Was there a change to how load is calculated in more recent kernels?
It looks like non-interuptible process rate a higher load than previously.

Hoping someone out there can shed some light on this. I might expect a bit more load with later code, but this seems so extreme, its like something has to be wrong/broken/misconfigured somewhere.

fwiw, have you enabled Flow Offloading in 18.06.1 ?

I havent done anything to do that, unsure whether it is enabled or not out of the box. I usually run SQM, and think that they aren't compatable, though theres a lot of confusion out there on that. Even with SQM off, I'm seeing the big performance hit.

In the short time I've moved to 18.06, I havent found info on checking or how to turn it on and off. Know its basic Linux config, so shoot me, gurus. :wink: Been too busy to dig in deep, lately.

I have followed closely the SQM vs FastPath / flow offloading, and they are mutually exclusive. The SQM needs to see / manage all the packets, as allowing bypassing the usual path for extended periods (like a big transfer in one flow) can impact latencies.

I run with SQM on, and like you, see similar loads even without it on. Of course, at my line rate, it's not a limiter.

Recently I moved my DHCP Server from my Internet Router (TL-WR1043ND v3) to my 2.4/5 GHz Access Point (Archer C7 v2). With 17.01.4 there was no problem entering the "DHCP and DNS" Page with 150+ DHCP Entries, within seconds all fields filled in and modifying entries proceeded correctly. After upgrading to 18.06.1 (jumped over 18.06.0) that same "DHCP and DNS" Page is now unworkable: Taking a long long time before all fields are filled in (during which status says "Calculating"). Then once all fields have been filled in, problems/delays while attempting modifications are so pronounced that it unusable! Additionally I noticed while 18.06.1 is being accessed through the LuCI Web Interface, CPU Load remains high, even while WiFi Traffic Load is Light.

I find the unusable "DHCP and DNS" Page with 18.06.1 unacceptable and I have returned to 17.01.5 on the Router and all Access Point (except the TL-WR1043N v5 Access Point, which is NOT supported by the 17.01.5 Release Version)!

Curious to know whether there are other 18.06.1 Users having similar issues with the LuCI Web Interface and more specifically the "DHCP and DNS" Page, when there are a large number of DHCP Entries?

OK, looking back at this again. Have taken some speed and load info from my box running 17.01.4, and the other C7 running 18.06.1.

Some notes:

I'm not seeing many processes running at 1%, at least somewhat after the first bootup. My bad on seeing a new column in top (?) VSZ% and thinking it CPU.

The 18.06.1 C7 does seem busier than the 17.01.4 one.

Seeing a bit less of the slowdown in Luci operations than the last time, but still noticeably slower in things like the realtime graphs, or any other page that has a lot of data to gather and display.

The networking slowdown is still there, I did test runs with ethernet and the 5Ghz AC WiFi, on both firmware versions and no SQM, Results below:

17.01.4 C7 v3 router

Router - ethernet - PC 250-270mbit DL, speeding up to 340mbit, UL typical odd 20's to 40's, with typical for my cable huge latency spikes
Router workload: ksoftirqd 40-60% CPU idle 5-30% (mid teens to 35% when below 300mbit, single digits when hitting 320-340mbit)

Router - 5ghz Wifi - PC 180-240mbit, avg 220's, similar UL speeds
Router workload: ksoftirqd 60-70% CPU idle 5-15%

18.06.1 C7 v2 router

Router - ethernet - PC 130-240mbit, avg 150's, a few moments peak 280, not 340's
Router workload: idle 5-40's%

Router - 5ghz Wifi - PC 130-170mbit, avg 150's, similar UL speeds
Router workload: ksoftirqd 40-80% CPU idle mostly hard 0%, occasional rise to 5%, sirq98%

As you see from the above, there seems to be a significant drop in speeds, with an increase in the loading of the system, though not entirely a clear cut one. A large drop in eth speed, but the idle time isn't hitting 0%, while you see a similar drop in the wifi speed, and idle hitting 0% for much of the time.

Later down the weekend, I will try to cross check between my two boxes, though I'm fairly certain that the difference between the C7 v2 and v3 (US one, not the Asian market one) is only in the wireless outboard radio and antennas, not in the processor. I might try 17.04.5 as well.

I'm wondering if there's a better system level tool or logger than top, to better capture what's eating the resources?