Recent content by fink.nottle

I've noticed this too after the android P update. It happens sometimes randomly. I have an ac68U. I think this is related to android P changes. It started after that.

Will it be easy to build the firmware with the older sdk/drivers ? Is it a simple config change, or will it need major surgery in the code base ? I commented in the other beta thread that I couldn't quite get the performance with new drivers even after trying a few tweaks here and there. I...

I'm quite convinced that this is entirely a client side oddity, and the internet is replete with performance issues on these wifi cards and iwlwifi drivers. I did try windows on the same laptop, and it did fine. I will try out the other settings you mentioned, and report back. Other than that, I...

My regulatory db file on xenial is pretty recent. It's from 2015. I did try with a factory reset too, and rebooted my laptop as well. The throughput was somewhat better (in the 70 Mbps range), but the original firmware still performs better. I can do more repeated tests and average out the...

I'm pretty sure that range will be fine, and even windows will perform alright. Android did in my case. The issue is more about the client's peculiarities on linux. I didn't try the firmware reset, but I can try it. If anyone has a sandybride/ivybridge era thinkpad, you could confirm my...

Does this affect the wireless on older (regular) AC68U ? This makes the wifi performance worse for my client (Intel 6300N on linux). This is a terrible wifi card with all sorts of performance problems, but the fork manages to make it hit 100 Mbps on linux, whereas the main merlin version...

@john9527 I managed to solve most of my bufferbloat woes by using one AC68U as a router running lede, and another one running your fork as an access point. Everything performs pretty well now. One minor issue that I've noticed is that if I enable ipv6 on the router, the wifi clients connected...

@ColinTaylor , @john9527 So I had a look at the detailed results (it's a clickable link in my previous post), and it sort of confirms the observations. i.e. Pings are affected by download saturation, whereas upload saturation is a lot better. It looks like the QoS takes a while to react to...

@ColinTaylor With QoS disabled and the settings you recommended: size: 42467 bytes (23069 left) qos_ack=on qos_addr_err=4 qos_burst0= qos_burst1= qos_bw_rulelist= qos_default=3 qos_enable=0 qos_fin=off qos_ibw=81920 qos_ibwopt=1 qos_icmp=on qos_inuse=9 qos_irates=100,100,100,100,100,0,0,0,0,0...

Yes, I changed the units to Mbps. I also know that tomato works much better for bufferbloat for my case. Does tomato have some sort of default settings that can be transferred over to the QoS here (assuming the fq_codel implementations are similar enough) ?

Base values are 120/12 (comcast). This is the baseline run without any QoS (on wifi). Wifi subtracts about 10 Mbps for my tests. I get full 120 in ethernet tests. http://www.dslreports.com/speedtest/8715410 There are the results after enabling QoS, and setting limits as 100/10...

Yes. The download/upload bandwidth caps seem to have an effect, but not on bufferbloat.

@john9527 , just wanted to get your attention.

So I flashed the latest stable release today, and did a factory reset. I don't see any difference in bufferbloat with the default QoS settings (fq_codel) in automatic mode. The default rules have Web Surf (0~1536) Highest, HTTPS (0~1536) Highest, File Transfer 80 (1536~) Low, File Transfer 443...

That reminds me. During the initial setup, you can't copy paste the admin password in the fork. Kinda annoying if you use a password manager. Can this be changed ?