Joined: 21 Jan 2017 Posts: 1783 Location: Illinois Moderator
Posted: Tue Feb 11, 2020 16:33 Post subject:
It sounds like NVRAM got bunked and you need to reset and start over. Use the settings in my signature below. Those settings will assure you have everything set up the right way and should yield the best/consistent results. _________________ FORUM RULES
Also is there a problem with achieving throughput with QoS enabled with this router? I thought I'd be able to get around my connection speed (460mbps). With QoS off, I get full speed, QoS+SFE I get like 160, QoS+SFE Disabled I get around 260mbps.
Looking at CPU loads on the status page it doesn't look like anything is maxing out.
Joined: 21 Jan 2017 Posts: 1783 Location: Illinois Moderator
Posted: Tue Feb 11, 2020 17:35 Post subject:
SFE does not work with Q.o.S. turned on.
The R7800 should be able to hit 600mbps just fine.
Are you sure your Q.o.S. settings are set the right way?
Again, maybe you need to do a reset first and then try again. _________________ FORUM RULES
Wonder if my version has some pretty major bugs. Basically had the same issues after doing a NVRAM reset. I'm running r42287.
Even setting HT80, AC only, it doesn't seem to work properly. I was getting 400mbps after enabling short preamble, however after trying to enable TurboQAM it dropped me down to 173 and no amount of tweaking is bringing it back to 400mbps.
I was misreading things. I thought because Voxel and Stock supported 80/40/20 (like my 86u), DDWRT would too. It appears you can only do 20/40 as per the option without explicitly setting it to 80. So 400mbps is the max unless I want to kick some older AC devices off my network. That being said I'm still not getting it.
QoS seems to now have proper throughput, getting close to max speeds when I set a really high max.
Not sure how you determine SFE is off? With QoS on, the radial button is still enabled on the setup page.
Joined: 03 Jan 2010 Posts: 7568 Location: YWG, Canada
Posted: Wed Feb 12, 2020 7:22 Post subject:
Bensam123 wrote:
Not sure how you determine SFE is off? With QoS on, the radial button is still enabled on the setup page.
its overridden and force disabled. if u override the override and force enable sfe, qos will only half work, being on upload only not download. at least that how it was when i last looked at it. _________________ LATEST FIRMWARE(S)
BrainSlayer wrote:
we just do it since we do not like any restrictions enforced by stupid cocaine snorting managers
no idea what you mean exactly
a correct VHT80 configuration delivers 20+40+80mhz wide channels.
you should post your complete WLAN configuration
Did not know that, when I selected VHT80 my laptop refused to connect and assumed it didn't support 20/40 (as it wasn't explicitly mentioned).
Still refuses to connect, however my cell connects at 866mpbs (which is what originally happened). Both cell and laptop work fine with 86u and stock/voxel.
Short preamble, single beamforming, multi beamforming, airtime fairness, and turboqam (in the 2.4ghz options) are the only ones I've been really messing around with. Channel width is either on 40 or 80 (since 80 doesn't work with my laptop it's been on 40).
Got it. It was doing something silly in the background while it was on auto, which is why I couldn't nail it down. Any of the channels pretty much fixes the issue.
Swear I tried 161 the other night, I must've had channel width set to something else while doing it.
So, I'm back... Curiously is QoS broken in DDWRT? I don't remember it being so easy to just outright wreck it. The router performs better without QoS on. When selecting HTB or HFSC it kinda works, but if you actually try to 'battle' different hosts on the network against the router the router outright explodes. If HFSC is on I can't seemingly add one of the computers without it causing the WAN connection to just disconnect outright.
(Figured out the above after testing. If HFSC is enabled and any computer is set to standard or below it causes the WAN to drop out.)
I first noticed a issue when attempting to stream and it didn't seem to make a whole lot of a difference whether it's on/off (compared to a 86u). While downloads throttle and it looks like it's working, if you fire something up that competes for bandwidth it just chokes. I tried running a torrent client with a few big torrents on it, over 600 connections, hitting ~300mpbs (under 460 my connection provides). With either HFSC or HTB on tons of packetloss happens, extremely erratic ping times on a maximum priority system compared to a 'bulk' that's doing the torrenting. If I try to run a speedtest while the torrent is running, it squeezes a measly 90mbps out of the connection while the torrent machine runs full tilt. Video streams are interrupted on the maximum priority machine and of course includes anything latency sensitive.
Manual speed limitations work, but that's extremely rudimentary QoS.
Turning off QoS, the packetloss almost completely goes away, however all the fun of having a fully unhinged network happens. The Torrent machine is basically a worst case test scenario, but it seems as though somethings broken. I used to use DDWRT on my 68u and with HTB/HFSC FQ_Codel it didn't have these sorts of issues. This isn't remotely close to the performance of my 86u on Merlin.
Doing some benchmarking, packetloss starts happening when the torrent machine hits around 220-250 connections.
Comparing HTB to HFSC, HFSC limits the torrent machine to around 120mbps, however it doesn't seem to matter (QoS limit is 420/19, under my 460/22 cap). It results in roughly the same amount of packetloss, maximum priority machine still has dropped streams, voip is still broken.
Watching stats, CPU utilization is still around 40-50%, free memory is around 75% for HFSC. HTB CPU sits at around 30-40% with higher speeds and better pings
I heard good things about cake, which actually seems inferior to FQ_codel based on testing with pingplotter and various loads (not specifically torrents), but this seems ridiculous.
If I remember right I think I was using Kongs build on the 68u, does anyone have access to his last version for this router or is there a repository?
This was bothering me, to answer my own question here, I found a bootleg version of Kongs older firmware. Works fantastic with the above testing parameters. A lot less knobs to tweak, but extremely solid. Currently outperforming my 86u with HTB+FQ_Codel on ping times. Only occasional packet loss showed up after 490+ connections. Even then it only seems to really happen when I battle against the torrent machine with a speed test on the high priority machine.
The maximum priority machine will steal almost all the bandwidth from the bulk torrent machine, respecting what is listed in the wiki.
Packetloss is only around 4.7% on HTB compared to 8% with HTB and 10% with HFSC on BS.
Removing ACK from high priority removed a lot of the PL.
Kong I'm getting better average ping and less jitter then the 86u with Merlin under the full torrent testload. However the 86u Merlin I don't have any PL.
Better CPU utilization under both HTB and HFSC compared to BS, however it wasn't maxed out in the first place.
Despite packetloss video stream on the high priority machine never buffered or lost connection.
Going to further add, I thought it looked like the cores were throttling and dropping down to 800mhz. Turns out it's the scaling governor (old problem). Setting the cores to their maximum frequency removed all packetloss on primary targets and further improved latency and jitter by a slight bit.
Code:
for CPUFREQ in /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do [ -f $CPUFREQ ] || continue; echo -n performance > $CPUFREQ; done
echo 2 > /proc/irq/255/smp_affinity
After Kong and tweaking it looks superior to my 86u Merlin router.
yes the ondemand govenor and frequency scaling causes package drops
and i think the performance govenor is standard in the BS builds
QoS also only runs on one processor core and the processor is definitely too weak for 500mbit lines and 500+ connections .. there are many many small packages that have to be processed and in principle only powerful x86 can do this
