bufferbloat differs between bands

Report wireless and/or network connectivity problems in this forum.

Moderator: Moderators

Post Reply
jayray
Posts: 17
Joined: Fri Apr 14, 2017 7:02 pm

bufferbloat differs between bands

Post by jayray »

Hi, new guy here. I've got a whole backlog of networking nags filling up the ol' brainbuffer to the point where the wetware is clogged and needs cleansing. I'll try to keep focused on one at a time, but you know how these things can be.

Gargoyle version 1.9.X (Built 20170323-1952 git@85255a6) is installed on my Linksys WRT1900ACSv2. I came over to gargoyle from DD-WRT because in the four builds I tried, the wireless network on both bands would inevitably degrade over time. They'd start out great but after a few days what was once a nice, stable maxed out transfer rate (~100 mb/s) would end up real ugly and spiky and slow, frequently down into the kb/s range.

Here on gargoyle I'm experiencing similar although not identical symptoms. As per other fw installs, everything starts out smooth, stable and clean.

2.4 ghz set to:
-b/g/n
-ch 11, 40mhz, lower
5 ghz set to:
-a,n,ac
-ch 36, 80mhz, upper implied

QoS is different here from DD-WRT, so I left them at the defaults for now except two. I turned on ACC and set throughput limits on tx and rx at a very conservative 80%.

So the router ran so great. I mean just really, really great. Like i said above, speeds on both bands were smooth, stable and fast. (Upload was a little wobbly compared to all the DD-WRT variants but nothing to stress about just yet.)

Then I woke up this morning to find that during bloat tests at dslreports, the 2.4 ghz band was bufferbloated pretty badly. Speeds were spiky, and much slower than the night before. Not as bad as where the dd-wrt fw took me but certainly troublesome. Lag is present from the moment the test starts polling servers. Assuming the test even begins, lag on download ranges from roughly 400-1000+ ms during each test.

The 5ghz ran just fine. Download speeds topped out at 85mb/s right about where I put them. Lag increases under +40. Upload speeds are a little wobbly like I said, but still average out more or less to the limit and that's good enough for now. And no bufferbloat issues there either.

Rebooting solved nothing on the 2.4 band.

So I turned of QoS completely.

Things get a little weird here. Unexpected results arose.

The 2.4 ghz band now shows stable download speeds. Depending on.. things.. this could be anywhere around 50 - 70 mbs but per test its pretty stable. Lag rises to around 200-220 or so.

Upload is just as wobbly and laggy as before.

The 5ghz network, sans QoS, is better than ever. Download speeds rose from ~85 mb/s to ~120 mb/s. Everything nice and level on the speed graph. And the lag is almost non existant, identical to the result with QoS on.

Uploads on the 5ghz band are about twice as high, but as usual a little wobbly and with lag hovering around 400.

So I should probably wrap this up.

What the heck is going on with the 2.4 network? My channel is clean. I see two potential sources of interference from my wifi analyzer, but those routers, when they pop up at all, never rise higher than -90 dbm vs my routers ~35. I've tried this also on channel 1 to the same effect.

I'm testing when the network is quiet. Router temps don't seem excessive. But somehow the 2.4 ghz band is worse off with ACC on and with a throughput limit than with neither.


Since without ACC active the 5ghz band shows no bufferbloat, it seems to me that the 2.4 band should put even less stress on the router queues. And with ACC on, the 2.4 band gets even worse.

I can't crack this one by myself. Why does it develop these symptoms after awhile instead of showing them right away? Am I likely to see the kind of degredation i saw with DD-WRT? Also on both bands? What could be going on here and can it be fixed?
WRT1900acsv2 / alas, back to DD-WRT
WRT54G /DD-WRT v24-sp2 (08/12/10) mini

ispyisail
Moderator
Posts: 5185
Joined: Mon Apr 06, 2009 3:15 am
Location: New Zealand

Re: bufferbloat differs between bands

Post by ispyisail »

Gargoyle is just a front end for OpenWRT / LEDE

Install OpenWRT or LEDE and re-run your tests

If the problem goes away then the problem is with Gargoyle.....

Please report any results

tapper
Moderator
Posts: 1076
Joined: Sun Oct 13, 2013 5:49 pm
Location: Stoke-on-trent UK

Re: bufferbloat differs between bands

Post by tapper »

Hi the bufferbloat you are seeing mite be in the wifi drivers.
You can read more about wifi and bufferbloat in the wifi stack here:
https://www.bufferbloat.net/projects/make-wifi-fast/
What router are you using? All the work on make wifi fast so far has bin put in to LEDE 17.1 I have it flashed to a C7 I have running as a AP and it is not as laggy as OpenWRT / Gargoyle. When Gargoyle is built ontop of LEDE all the wifi fixs will come to. We still don't know when that will be yet as Openwrt and LEDE are still talking about merging again. All the work on bufferbloat in OpenWRT CC 15.1 was dun on the wan and is most apparent with wired connections. Hope this helps.
Linksys WRT3200ACM
NETGEAR Nighthawk R7800
NETGEAR R6260

jayray
Posts: 17
Joined: Fri Apr 14, 2017 7:02 pm

Re: bufferbloat differs between bands

Post by jayray »

Thanks for checking in on this post.

Tapper, I'm using the Linksys wrt1900acs v2. It's one of those versions where the box says v1, but the sticker on the router says v2. The consensus around the web seems to be that the box is wrong, and the sticker is correct.

Also, I've been reading about bufferbloat quite a bit lately, including I've just about everything on bufferbloat . net and similar sites. Thanks for the reading suggestion. Unfortunately, I'm not certain what, if any, of that information explains what's going on with my 2.4 band. Granted, much of it is more technical than I can properly absorb. But what I've read so far suggests that the router<>wan interface is the primary bottleneck that edge networks have any real control over.

BTW, these open source FW packages are fantastic, considerably improved over the default FW and make that control possible. So my deep gratitude goes out to the coders, tweakers, troubleshooters, theorists, technicians and interested laymen who take the plunge.

Where was I? Oh, yes. Since both wifi bands have to go through the potentially bottlenecked router<>wan interface, i would expect that if the 5ghz network can max out download speeds with no appreciable latency--even with QoS turned OFF--then the 2.4 band should do the same. Am I wrong about this?

When my network is unpopulated, I'll give openwrt/LEDE another go. I first came to gargoyle after failed attempts to set up them up. _Supplying internet access via wifi on LEDE/openwrt is a more convoluted process than I anticipated. I'll have another go at it in five or six days and let you know once its set up.

Meanwhile I'll keep gargoyle going and hope the system doesn't degrade any further (it seems not to be) and that maybe someone has something practical i can try until I can get LEDE up and working.
WRT1900acsv2 / alas, back to DD-WRT
WRT54G /DD-WRT v24-sp2 (08/12/10) mini

Post Reply