Vdsl 100mbit wdr3600 and sqm?

hey,
i get soon a dsl upgrade from 16 to 100mbit and want to know if my router still powerful enough to handle sqm, with cake?

It's able to cope with 100 MBit/s, but with only little headroom (system load in the top quarter when you stress it, with some peaks down to 0% idle; the tl-wdr3600 is already supported in ath79, which would give you access to software flow-offloading as well).

SQM on top most probably won't work at that speed.

what does this mean? do i need to install snapshots for that or is it in the latest 18.06.0 release? if so, do i have to enable it, or is it on by default?

ath79 is not part of openwrt-18.06.

At this moment ath79 is still marked as source-only in the master branch, which means it doesn't get built on the buildbots (--> there are no downloadable images, you'd have to build it form source yourself). The reason for this is it being rather new code (work in progress, quite actively) and not every device that has some initial porting done is working at this moment (e.g. tl-wr1043ndv1/ wndr3800 exist, but are still broken), the tl-wdr3600/ tl-wdr4300 images however are working nicely.

The plan is to enable building ath79 snapshot images for master (snapshots) 'soon' - and to remove ar71xx images for ath79-supported devices soon after.

You could overclock your cpu from 560 MHz to 700MHz or more to squeeze out a bit more horsepower...

Checkout pepe2k's great u-boot mod:

He supports only a highest clock of 650 MHz, if you want more i've added also 700 MHz and 750 MHz to the setclk u-boot command in my fork.

You could soft brick your device, but if you have a backup of all mtd partitions and a hw programmer (~5 bucks) you could always recover you wdr3600.

According to ath79, i've using it on most of my ar71xx based devices and shared my builds with the community...
Checkout the community build thread if you want to use it without compiling the images yourself:

The tl-wdr3600 is fast enough (for 100 MBit/s, but without SQM) to just wait and see what happens.

Consider upgrading to something with much more CPU power. Espressobin would be more or less the minimum. Mini X86 pc would do well.

100 MBit/s ("i get soon a dsl upgrade from 16 to 100mbit") in this context most likely means 100/40 MBit/s VDSL2 with vectoring (G.993.5, dsl profile 17a), with -perhaps- a potential for 250/40 MBit/s (G.993.2, dsl profile 35b) down the line. The former is easily done with modern consumer router SOCs (ipq40xx, ipq806x, mt7621 (perhaps a bit borderline for SQM), mvebu), including SQM, the later should still be comfortable for mvebu and ipq806x, including SQM. ar71xx, like the tl-wdr3600 mentioned here, just is a bit borderline and showing its age, that doesn't mean there wouldn't be plenty faster options in the 70-200 EUR range that can cope with the expected loads easily.

x86 is definitely an option and becomes more attractive if you're a business or need to run additional services, but it's overkill for normal home uses (and VDSL2+vectoring with 50/10 MBit/s or 100/40 MBit/s is actively marketed to consumers, wherever possible - for basically the same prices as ADSL2+ with 16/1 MBit/s). The reason for VDSL 100/40 being actively marketed in, e.g., germany is that the ISPs want to get rid of ISDN+xDSL (BRAS) and PSTN+xDSL (the later at the latest at the telco switchboard) until the end of 2018 in favour of a common BNG platform with VoIP/ SIP. To accomplish this, and in order to profit from government grants, there has been a real frenzy of building outdoor DSLAMs (FTTC) for the last 1.5-2 years (including areas that had no prospect of improvement for the last 10 years), with aggressive marketing (VDSL 50/10 MBit/s costs the same as (up to, but often significantly less than) 16/1 MBit/s ADSL2+, and VDSL 100/40 a fiver more). Super-vectoring is appearing on the market within the next couple of months, for yet another 10 bucks more and probably only reaches enthusiasts for now (the upgrade to 50/1 MBit/s is effectively 'free', 100/40 just a bargain - but 250/40 MBit/s probably feels more expensive for most 'normal' home users). A further speed increase is not in sight any time soon for many, without changing technology (cable (probably not that different in terms of effective throughput) or FTTH) - and the competition has a difficult position against the existing copper wires for the next couple of years.

Disclaimer: I don't claim that concentrating on VDSL+vectoring wouldn't be short-sighted, but it's what the companies are pushing and what's 'cheap' and 'good enough' for most home users for the next couple of years. This short-sighted'ness will hurt in the future, but neither companies nor ordinary users will look that far.

Edit: as mentioned, the existing tl-wdr3600 (with SQM disabled) is -at its limits, but fast enough for a start- for 100/40 MBit/s, so mezo can relax for now and just wait and see if it's good enough for the envisioned tasks, before starting to shop in a hurry. (yes, I did test a tl-wdr3600 on such a 100/40 MBit/s VDSL2 connection).

Edit: I would currently put the cut-off between a high-end consumer router with OpenWrt and a x86 system with OpenWrt at somewhere between 300-500 MBit/s (symmetric, or close to~), the goal posts for this category of high-end consumer routers will change with SOCs improving in the future though.

The thing is there are plenty of mini PC's that reach below the high end router prices... $150 has been a doable price for some mini PCs. At this point the high end consumer router is really not that economically viable since the PC will perform circles around it. Trade-offs vary of course by region and market conditions etc. But the mini PC if you can get one at $150-250 is a good bargain compared to high end consumer routers

1 Like

I agree with that (I just put the cut-off point for x86 to be viable a bit higher), but prices starting around 150-200 EUR don't include concurrent dual-band wlan - adding that means another 50-60 EUR, either in wlan cards/ pigtails to add to your mini PC (and the selection of boards with at least two 1 GBit/s ethernet ports and two mini-PCIe slots is very limited (and more expensive), that's even before adding a dedicated managed switch) or by adding a dedicated plastic-router to use as AP (40-70 EUR at least).

On the other hand you will find high-end ipq806x and mvebu routers with concurrent dual-band wlan, which are fast enough for 300 (ipq806x) to almost 1000 (top-end mvebu) MBit/s, for 160-200 EUR (delivered to your doorstep from a reputable same-country shop, including taxes).

Yes, if you have highend expectations, you're more likely to prefer decoupling the individual functionalities into dedicated devices, but that's not quite the case for most home users (who'd even like to put modem, pbx/ ATA/ DECT into the same device, if that were supported by OpenWrt).

But we're still talking about 'just' 100/40 MBit/s here; while that is a lot, compared to the days of yore, it's just a bit more than mezo's existing router can deliver - and not putting more modern plastic gear into sweat.

...and btw., I don't really think you even need to go up to high-end ipq806x/ mvebu here - ipq40xx (AVM Fritz!Box 4040, ZyXEL NBG6617) for ~70 EUR or low-end mvebu (Linksys WRT1200AC) for ~120 EUR should cope with 100/40 MBit/s easily, including SQM (if you don't have enterprise requirements/ if you don't share your flat with a handful hardcore gamers^w^wstudents or p2p supernodes).

this is exactly the router i had in mind to switch to... i read here and there about some problems with openwrt on this router, but they seem all fixed... can you recommend this device?

I have no personal experience with ipq40xx devices (they weren't supported early last year, when I was in the market for a new router), only with ipq8065/ nbg6817, but I've heard rather positive feedback from Fritz!Box 4040 users (although they only used it as AP, rather than as router). Hardware wise, Fritz!Box 4040 and nbg6617 are very similar, enough to consider them being equivalent in terms of performance and the general level of support (aside from the initial factory installation). While I'm rather confident that ipq40xx should be able to cope with your networking load, I can't vouch for it (even less for SQM usage, as I don't need it personally).

hey, i got the nbg6617 like 2 months ago and im super unhappy with this device. i face so many wifi problems, that im thinking about to get another device.

  • wifi adds like 3-4ms extra ping
  • random wifi firmware crashes
  • not possible to stream 2 1080p streams parallel over wifi, without alot of buffers
  • after "longer" uptime wifi gets unstable

are there wlan ac devices with stable wifi and good sqm performance for 100/40mbit?

Loot at the Archer C2600. Im using it myself on my heavily used 100/40, and another one behind 2 thick walls and a stair as a great wlan-bridge (wds). Second hand it cost around 50-70€. If you get one i can help you about more specific vdsl2 sqm settings, and irqbalance. This device might get even more powerful in the near future because the chipset family is still under a lot of development (like NSS drivers for the 2x NSS 800mhz cores that can be used one day to offload network/sqm/firewall cpu usage).

The Fritzbox 4040 seem to be a great choice on the second hand market too (starting at 40€).

For myself: i stay at ath9k and ath10k wlan chipsets if possible. Even those are not perfect, i just had too much trouble with other brands. Those chipset are supported by the wifi bufferbloat project: https://www.bufferbloat.net/projects/make-wifi-fast/wiki/

By the way, if you live in Germany, think about using the old wdr3600 for Freifunk :wink:

imho you should get some mini-pc with intel network cards and use your wdr as a dumb ap. easy to swap the wifi device if need arises.

+1, except for 100/40 an ubiquity edgerouter X would be a decent no-wifi prmary sqm-router (see https://forum.openwrt.org/t/ubiquiti-edgerouter-x-loading-openwrt-and-performance-numbers/27470 for installation instructions)

isnt the fb 4040 basically the same devices as my zyxel? both are ipq40xx based.

i was looking for some all in one solution, because i have already the modem not included and somehow i dont want to have 3 devices running, what in theory one device could do.

i feel you
but truth is, none of the devices you own can handle all the tasks alone.
one that would, is probably either not cheap or does not run openwrt.

the aera of 10ish mbit/s networking is mostly gone but the typical openwrt embeded devices only scale in the ballpark of 10x while connection speeds vary as far as 100x (10mbit-1gbit), hence the recommendation for pc-class.