[GCC 7.2 BUILD] Optimized TP-Link Archer C7 V2 AC1750 LEDE Firmware

2.4 GHz stays off and 5 GHz is not usable... will give them a try in a few days....

If you "upgrade" from a working system everything seems to be fine :slight_smile:

@grap & @fRUTTiFrESH could you try again? I just pushed a new build that includes the full version of wpad, which should support ieee80211w.

Thanks! I will try again - I will have access at 9-10PM eastern time.

Seems to work and the Dog had to wait to go out for a walk for 5 minutes.... Thanks

[ ]( )

Will tune in for more tests - Setup worked for me from scratch with performing a full reset!

Thanks

Time: ~ 3 Minutes

Awesome, thanks for giving it another shot so quickly. Now I only need to figure out why the SHA-1 and SHA-256 performance went down compared to the stock build, which might take a while - need to rebuild every setting from scratch and test in between to narrow it down to a particular setting that's causing this. Will let you guys know when that's solved.

You can convert your C5 to Archer C7. I did it and it works great.

http://www.hagensieker.com/archerc5toc7/index.php

Just flashed the latest version of your firmware and so far so good, but there are a couple things maybe can help others.

Is it possible to add the following packages:

  • ca-certificates
  • ca-bundle

And some strange messages appear on about ath10k, I'm not sure if it is the ath10k crashing.

Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1851.984897] ath10k_pci 0000:01:00.0: ath10k_pci     ATH10K_DBG_BUFFER:
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1851.991223] ath10k: [0000]: 6F031D00 0128FC17 070B1071 03000A01 03300106 FEFF0000 FF000000 6F031D00
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.000442] ath10k: [0008]: 0128FC17 80881071 08000000 A42A4400 35009600 F00F0000 6F031D00 0128FC17
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.009623] ath10k: [0016]: 80881071 08000100 A42A4400 35009600 FF000000 6F031D00 0128FC17 80881071
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.018812] ath10k: [0024]: 08000200 A42A4400 35009600 FF000000 6F031D00 0128FC17 80881071 09000300
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.027998] ath10k: [0032]: A42A4400 35009600 FF010000 6F031D00 0128FC17 80881071 0A000400 A42A4400
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.037183] ath10k: [0040]: 35009600 FF030000 6F031D00 0128FC17 80881071 0A000500 A42A4400 35009600
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.046363] ath10k: [0048]: FF030000    
Thu Feb 16 22:27:09 2017 kern.debug kernel: [ 1852.049999] ath10k_pci 0000:01:00.0: ATH10K_END

Those syslog entries are normal and just very verbosive.

Could you elaborate on why you need CAs?

Hi.

I use the CAs to perform tasks like DDNS, over SSL.

Another thing I'm struggling is with a OpenVPN message that appears after recreating the previous configuration.
IP packet with unknown IP version=15 seen
But even with that messages the VPN is working, I'll continue investigating.

Nice to see the alternative, optimized version out there...

I'm also interested in the speed increase potentials... but for the SQM capability.

I recently went from a 50/5mbit to a 300/30mbit connection. Looks like a C7 starts to saturate when running Cake and Piece of Cake, at about 120-130mbit DL. It will do 220 or so, but it's pegged out on the IRQ's long before that and probably not doing the SQM very well. I've seen mine hit 350mbit DL with no SQM. Not sure if compiler optimizations would help. Maybe I should be asking about OC'ing the CPU instead? :wink:

Doubt it will ever do the full 300, but would be nice to get better than half... Anyone else playing around with this on a faster link?

Thanks for sharing this. I was about to buy WR1043NDv4 with almost identical CPU (750Mhz compered to 720Mhz in Archer C7) for my 120/10Mb connection and I think I will put his on hold.

I don't think this CPU can be OC''ed.

Pretty unrealistic...
Did you maybe think about buying EdgeRouter X and putting Archer C7 as AP and switch? The ER-X costs about 50$.

  • Both fq_codel and Cake top out for me somewhere between 120 and 140 Mbit symmetric
  • Both fq_codel and Cake are capable of 20 Mbit upload, 200 download on this hardware, but Cake's results appear "better" (significantly higher average throughput)

It's still not 300Mb but close. If you'd want more you would have to probably invest in something multicore, like TP-Link Archer C2600 (can do 800Mb without SQM) or Netgear R7800 or just buy APU2 or APU3 from PC Engines that can do 1Gb with NAT and can do 500Mb with SQM with 15% (or 30%, I'm not sure) CPU utilisation on all 4 cores so it should handle 1Gb with easy.

@JonP

Have You tried different devices ?

@all

Any Idea where I can order the STORYLiNK SAP-G3200U3 ? Specs looks nice....

Seems to be available here: http://item2.gmarket.co.kr/English/detailview/item.aspx?goodscode=882097586

There's also this, which seems similar but with more flash: http://item2.gmarket.co.kr/English/detailview/item.aspx?goodscode=653463439

Hi Adrien...

If the one your'e looking at does have equal performance to the C7, I'd
think it would be OK at 120mbit, with Cake. I read that Cake is supposed
to be lighter on the resources than fc_codel, a while back. Right now, I
set mine to limit at 140-145, and get abt 125-135mbit speeds on my
DSLReports testing. I'd guess you'd run a limit of 100-115mbit, and have a
bit more headroom. That's thru the 5Ghz wifi. With no SQM, Ive seen
250-280mbit on 5Ghz, and even as much as 350 direct thru the eth hub (no
wifi) my cable willing and the traffic low... with the sirq only getting
to 70%

I found you have to run top on the terminal to really see how loaded it
is. Load numbers don't show it. Even the CPU% number in top dosent,
though you will see 80-90% on ksoftirqd during a DL. The sirq will rise to
90-100% when its running up to its limit. I'm not sure whether the
ksoftirqd CPU or the sirq % is the one to pay the most attention to. At
140mbit limit, my C7 runs mid to high 80's on the CPU, and low to high 90's
sirq %. Hope that's helpful info...

Thanks for the Edgerouter tip.... might try that if/when my cable co gets
their s#!t together and I don't just decide to go down to 100mbit out of
frustration.
I went up to C7 from a very old Netgear, mostly due to bang for buck, good
wifi performance reviews, and its one of the dev platforms the
Make-wifi-fast guys were using. Haven't used anything else.

nvm: I found a work around

Another quick note on C7 performance at the high end... I've been playing with a suggestion from JMoller, to run 0 as the DL speed, to only limit the UL. Running 0/28mbits, I see pretty much the full speed, i.e. up to 340-360mbit in, while Cake limits my UL to below 30. I see ksoftirqd running 60-80% CPU, and 90-98% SIRQ with the download peaking that fast. Very even and low latency on both sides.

So, it looks like a C7 can handle (partially) an eth 300/30 connection with Cake after all. I'll try this with the wifi later. The wifi probably will take that down some.

Caviat... my cable service, modem and router seem to be pretty bloat free already, I rarely see bloat now even WITHOUT sqm applied, so maybe I'm not as hard of a test case. Wish they could cure the packet loss and latency spikes I've been having for months, though...

If I get adventurous, I might try this firmware to see if there's a noticeable bump in SQM speed, but kind of busy now. Any problems sysupdating to this and then back to LEDE?
Anyone else up for some benchmarking?

What do you mean? This build is based off of the LEDE branch - main differences is that its preconfigured with additional utilities

Wasn't sure how far out on the branch he's climbing... :wink: thought possibly with all the kernel compile changes and specializing, there might be a chance of issues, and also not sure the amount of difference between sysupgrade and full upgrades for changing things.

Hey, last time I spent a lot of time playing around with Linux, the kernel had just gone to 2.0.x.x...

I personally tried enabling SQM only on upload but I wasn't happy with the results. I saw bufferbloat here and there while downloading on multiple machines, at full connection speed.

In my case the bufferbloat was almost non-existent when using WiFi, but on cable it was happening. Does you modem from the ISP is working as modem or router? Do you get public IP on Archers C7 WAN interface? I was told that cable internet providers have already build in some anti-bufferbloat solutions but they are working only when the device is working as router.