Quick question.
Can this CPU handle 250/20 Mb/s with SQM and cake? Can any of you maybe test this?
How bad really is the 2.4Ghz network? I have a good price on this router and I'm thinking hard about buying it but most of my hardware is 2.4Ghz so it's important to me to be good enough.
It can easily handle 250/20 with cake. I did some benchmarking with my dir-860l a few months ago with iperf3 WAN <> LAN.
It did gigabit speeds without shaping (930mbit in both directions).
450-550 Mbit in both directions with cake enabled.
600-700 Mbit in both directions with fq_codel enabled (yes fq_codel performed better for me).
It does 350-400 Mbit on my 500/500 connection with shaping enabled over a PPPoE connection (I hate my ISP for using PPPoE. Extra CPU load).
It handles the full 500/500 Mbit over my PPPoE connection with shaping disabled.
I hope these figures are helpful for you 
It does. If it can do those figures then great!
Can you tell me more about WiFi? It's not a complete disaster if I end up disabling 2.4Ghz entirely and using my current router for this but I would like to know how it works in practise.
I also read somewhere that the most popular router around here is Archer C7 and just behind it is D-Link DIR-860L. Is this really true?
What is actually the state of SQM on DIR-860L. You guys are saying it's all working great now, while here are reports that it actually doesn't https://bugs.lede-project.org/index.php?do=details&task_id=764.
@r43k3n Wifi works fine for my usage, but I don't use heavy traffic over wifi. Only my android phone experiences some weird issues related to wifi, but it experienced weird issues with my previous archer c7 router as well, so I suspect the phone rather than router. Speeds are a bit lower on DIR-860L wifi compared to my previous Archer c7, but for all other devices it's been rock solid stable, which is most important to me.
As you can read here, @drbrains has successfully applied fastpat to a mt7621 device. He is using a ZBT WE1326 for that. It might be a nice feature to explore for your builds as well 
The commit that added the functionality applied cleanly and successfully compiled without any additional modifications. This is the commit in question: https://github.com/dissent1/r7800/commit/93ae487b9e685bd73e6ad301d708009e46f22886
I am a bit short on time, but I will see whether I can apply this commit to the latest master branch once I have time to play around with this
That happens after a while but in my latest build (r4633) opkg can still install packages. Just tried installing vim and it worked like a charm.
Using iperf3 with the following switches "-n 512M" on the 2.4 GHz band in a semi-crowded environment (5 other AP's) I get the following results:
- Android phone (Asus Zenfone 2) connecting at 72 Mbps:
[ ID] Interval Transfer Bandwidth
[ 5] 0.00-99.95 sec 0.00 Bytes 0.00 bits/sec sender
[ 5] 0.00-99.95 sec 512 MBytes 43.0 Mbits/sec receiver
- Windows laptop connecting at 144 Mbps:
[ ID] Interval Transfer Bandwidth
[ 5] 0.00-76.48 sec 0.00 Bytes 0.00 bits/sec sender
[ 5] 0.00-76.48 sec 512 MBytes 56.2 Mbits/sec receiver
Decent enough imo but like @Mushoz I am not a heavy user of the 2.4 GHz band. Devices that use a lot of bandwith are connected with an ethernet cable.
I'll try applying the patch tonight. I have been waiting for it to be in an easier to use form 
1 Like
OP updated with a build which contains the Qualcomm shortcut-fe module. Happy testing!
No new builds for some time because I am going on vacation for two weeks.
Edit: Derp, I selected the fast classifier module. New build on the way with the shortcut-fe module!
Anything we have to do to enable shortcut-fe?
EDIT: Seems to break SQM somehow... had large bufferbloat in download, but not upload?
I am not sure whether "breaking" is the right term, I believe it avoids using the part of the linux kernel networking stack where ingress trafic shaping is happening (in SQM's case we use an IFB and the tc action redirect to pass the incoming packets through a virtual egress interface to which we can attach a traffic shaper). I have not yet an understanding how shortcut actually works, but the fact that working around the kernel's network stack will forego some of that stack's features does not surprise me. (My assumption is that the shortcut module is simply faster by doing less work then the kernel would usually do, so leaving some features unsupported seems fair; except it would be nice to simply not go into fast-path if "slow-path" features are requested and configured)...
Actually he's right.
SQM on download it broken when used with Fast Path (fast-classifier). It requires some kernel modification to make it work and it's not done yet. However it is reported that shortcut-fe is working fine with SQM. You may want to try it out.
How do you switch between fast-path and shortcut-fe?
I've asked that a few times and didn't get an answer.
Apparently if you build an image with both the shortcut-fe will take precedence and will be the default one. Haven't tested this myself, just something I read.
@All, I incidently selected fast-classifier instead of the the shortcut-fe. Made a new build with just the shortcut-fe module and uploaded it to the OP. Be awera that this build has the same revision number and that it is untested!
You could load/unload kernel modules but I haven't played around with it yet due to time. If both are selected and loaded it looks like, as @r43k3n posted, fast-classifier has precendence over shortcut-fe.
Please try a later build since r3285 is kinda old. The wireless driver has seen a few commits which improve it's performance.
Thanks @Bartvz for fixing this build even though you're on vacation!
I believe the shortcut-fe build got SQM working again:
How do I check and see if shortcut-fe is running?
You could try grepping for shortcut-fe while using lsmod 
check it by yourself, latest build laptop asus gl553vw 866Mbps wifi adapter ,connected to my router 4 m away, the transfer is from laptop to gigabit wired connected to router PC....pretty dissapointing!!!!
Latest build is untested by me but with the previous build I was able to get more throughout.
What is your network configuration and your network environment (amount of surrounding AP's with their bands, etc)?
the only 5Ghz access point in my area its mine, all the others are 2.4Ghz (not a city area) at the same position with an archer c7 v1 factory FW i get about 50-70 MB/sec, dir-860l b1 with factory FW v2.03b3 i was getting 40-50MB/s a dir-818lw b1 factory FW too is giving about20-25MB/s and a 941ND (2.4Ghz) v3 with ddwrt latest, about 11.2MB/sec, all the above same scenario, transfer of a single big file from my laptop (866Mbps intel adapter) to my gigabit lan cable connected to router pc.
What throughout you had with previous build? to be honest i am expecting from this router a better throughout than The Draft AC Archer c7 v1 , i mean something between 70-80MB/sec because of the better hardware D-link has and also the more powerfull CPU with more cores.but till now even with factory FW its impossible to reach it!
both 5ghz and 2.4ghz performance has never been great on my dir860b1 (good enough to saturate my isp speed though at least for 5ghz)
I just did some iperf testing with Reboot (SNAPSHOT, r3764-afc1675833) loaded.
VHT20 -> 110 Mbits/sec
VHT80 -> 170 Mbits/sec
Using my tm ac1900 (asus ac68u) at the same location:
VHT20 -> ~110 Mbits/sec
VHT80 -> ~350 Mbits/sec
Server: x1 carbon 1gbps Ethernet connection, client: galaxy note 5
Iperf in the other direction, which is usually the more important direction (client downloading from router), the 860l seems to do a little better
server: galaxy note 5, client: x1 carbon 1gbps ethernet
860l:
VHT80 -> 210 Mbits/sec
AC-68U:
VHT80 -> 250 Mbits/sec
The performance drop for ac68u from 350 to 250 is quite surprising to me...
But I suppose a real test for the wireless capability would be using a pair of 860l with one configured in client mode, and have ethernet hooked up to iperf server and router on both ends.
170 bps i am getting with my 1043nd v3 @ VHT40, i think D-link dir-860l has a big way ahead to obtain a decent build! at least this here is way better than official lede, but even fcactory FW in this router lacks the performance, this hardware should have!! With 2 cores @ 880 and 866Mbps AC should not been lower than 70MB/s, the 20MB/s is good for 2.4Ghz old N atheros chip@400 with 1 core and gigabit lan!!