UK VDSL user looking for SQM/Cake setup

This was indeed with a 'stack' of:
PPPoE
NAT
firewall
SQM
WiFi

So without e.g. SQM things would be less CPU starved, but in my testing I liked the responsiveness with SQM considerably better even if it cost me a few Mbps. However I still had a more powerful OpenWrt router sitting around and hence did not have to accept that throughput loss but relegated the HH5A to act as bridged-modem.

Not fully sure, I would assume all modems marketed as capable for VDSL2 vectoring to be fast enough. I have a zyxel vmg1312-b30a as backup, which on the dsl side has no problems of syncing at 116.7 Mbps, but only has fast ethernet ports so will already mildly slow down 100 Mbps dsl links (it was/is marketed as vectoring capable).

Over here in Germany one observation was that broadcom dsl modems harmonize slightly better with the almost exclusively broadcom line cards, but on a normal link, lantiq/intel/maxlinear modems also work OK. (For super vectoring profile 35b vrx518 lantiq modems appear just as good as broadcom).

Hi Everyone, in the absence of any HH5a flashed available, having read about for modems for a while the below in bridge mode seem pretty reasonable for the cost I would probably go with these.
Zyxel VMG3925, Zyxel VMG1312.

But for a wireless router is proving confusing. Some talked about options I'm seeing are the Belkin RT3200, Netgear R7800, Linksys MR8300. I was looking for something "safe" that as I beginner, if I buy the device new from amazon and mess it up or it doesn't work I could flash to stock and return it.

The Belkin seems I would have to flash using a complex method involving UBI or something as the normal install partition size isn't big enough? Seems complex/likely I might do something wrong as a beginner. The MR8300 I've read about some people having connection problems.

The R7800 seems to have good reviews and the openwrt wiki indicates it has an "easy" recovery method?

Honestly just open to any thoughts or suggestions so I can go away and continue to research my options.

I originally used a BT HH 5A for an all in one unit. This worked fine for SQM but started to run out of RAM when running adblock with medium sized block lists.

Afterwards I changed to Zyxel in bridge mode with a Checkpoint L50 wired router and a Meraki MR42 access point. These can be bought for good prices since the companies stop supporting them. However they require a bit more effort to convert to OpenWrt. (opening the cases and using serial connection etc).

Nowadays I still use a Zyxel in bridge mode but with a Cyberoam CR25ING router and a Dynalink DL-WRX36 as wifi 6 access point. This is complete overkill for my ISP connection however the router runs a lot more stuff than before. It does SQM, adblocking, ksmb network share, syncthing backup of my phone and radicale calendar to name a few.

There currently is a refurbished RT-AX53U on ebay UK which would be a good choice in combination with vdsl modem / router in bridge mode. This is perfectly capable to run SQM on your connection and has the latest wifi 6. If you decide that you need something more powerful in the future you can just add a wired x86 router and turn the RT-AX53U into an access point. The OpenWrt install does require SSH and is therefor a bit more work than just uploading the firmware but it does not require opening the case or any soldering.

The install process for this looking at the wiki looks much shorter than other routers, seems like it could be a good option

Hi Adam

I don't know if you're still about but I'll give my experience and thoughts. I was in a similar situation to you. I wanted better coverage than the BT Smart Hub could give (to cover some outside wireless blink cameras) and eliminate the bufferbloat. I wanted something relatively simple.

Although not anything to do with openwrt, I did like eero for its sqm. Not only was coverage now great and simple, I could get rid of bufferbloat... or so I thought. I got a 6 model and the sqm on that is horrendous for our cases. Either it doesn't work with such low download/upload rates or as people have said the fq_codel implementation is buggy. I'm considering returning it and getting a previous model which apparently has cake for the sqm. However that doesn't have PPPoE support so I'm debating if having to double NAT (use smart hub as PPPoE authenticator) is worth the headache of having cake. They are £58 rn so not too bad.

I did read that you can hack a HG612 modem and do something called like PPPoE DHCP spoofing to eliminate double NAT but I couldn't find concrete evidence or guide on that. Plus apparently people were saying the HG612 modems now are likely going to have capacitor issues with how old they are.

TP-Link do Deco products with modem built in. The X50-DSL is being released this week. It doesn't have sqm but simple QoS. I can't find any waveform bufferbloat results of these devices but there are some deco units with openwrt support so maybe in the future.

Out of your price range but I was really considering the DSL-AX82U. Problem was I didn't want to shell out £210 on one router and just have range be an issue again. I do however know it has merlin firmware support via a fork from official and that runs cake really well. I got a dude to give me bufferbloat waveform screenshots and his worst result was a spike of +6ms giving him an A over an A+ (also FTTC)

Don't get me wrong I like the configurability of the Asus and similar but when it starts talking about stuff like WAN packet overhead I get abit lost that I kind of want to embrace the Eero. If that thing can even give me close to what the DSL-AX82U does I'll be happy even if I am in this double nat stuff. Seems like thousands of uk eero users do that and are generally ok with voip and online gaming so worth a shot. Plus at £58 can't really say no. Can get a 2 pack for just over £100.

Hey,

I have just managed to obtain a HH5A flashed from a super nice guy on reddit. I've set up SQM and have improved my scores to an A, download active is about +5 and upload active about +2.

Cake definitely works but I'm still looking for ways to improve this, I'll probably search the forum some more and maybe start a new thread!

Thanks for your comment, if I really like this I may upgrade to a better router and flash it myself.

That's great. To improve you want to make sure your wan packet overhead is at 34 for BT FTTC. I got help here in confirming: Cake WAN packet overhead for VDSL2 PPPoE PTM with VLAN Tagging

You should be able to get A+ after making this change.

Thank you. I've had it set at 34 from following this guide: https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm-details

Even with really aggressive bandwidth limits I'm getting pretty poor performance I feel (its better than +300/600 download/upload active I was getting before)

There must be some other improvements I can make. I've just made a couple new threads in the installing and using openwrt forum to try and iron these issues out!

This is not really bad, is it? Cake by default will allow a standing queue of 5ms even under saturating loads (and since cake only engages if the standing queue actually exceeds 5ms, the true queueing delay under sustained saturating loads typically is higher than 5ms), so +10/+4 is not all that bad.
Especially at your achieved rates a single full MTU (~1534bytes effective size) will take a considerable time to be transmitted:
Estimate the gross rate from your measured goodput (assuming IPv6 as the test seems to prefer IPv6):

11.2 * ((1534)/(1500-40-20)) = 11.9311111111 -> 12 Mbps
5.15 * ((1534)/(1500-40-20)) = 5.48618055556 -> 5.5 Mbps

Estimate the transmission time per full-MTU packet in Milliseconds:

1000/((12 * 1000^2) / (1534*8)) = 1.02 ms 
1000/((5.5 * 1000^2) / (1534*8)) = 2.23 ms

so on average you expect your latency probes to show at least +0.5ms for download and +1.2ms just by the fact that on a saturated link there likelihood is high that on enqueueing there is already a packet being transmitted. And typically you will have a few flows with queued packets and cake will round-robin through these so I think your measured delay is pretty fine for such slow a link.