Cheap router for 600mbits/s sqm?

It seems promising.

1 Like

https://forum.openwrt.org/t/nanopi-r2s-is-a-great-openwrt-device/65374

It's what you want. I know it doesn't reach 600 Mbps in download with SQM, but the build is at initial state, it should be improved over the time, and anyway, you have to take off 10-15% bandwidth for SQM, so you will have to set SQM to something like 500-540. I don't think it's anything better for you, it's going to be cheaper and you don't have to mesh with the ethernet adapter of the RPI.

The price difference is minimal, and the second ethernet port on the NanoPi is just a permanently wired USB ethernet adapter. It doesn't have any benefits over using a pluggable USB adapter with the same NIC chipset. You still need all the same drivers in OpenWRT and have to set it up in just the same way, but if it ever goes wrong you can't easily replace it.

1 Like

RPI 4: 40 €
Heatsink/case: 7€
USB to gigabit ethernet adapter: 8€

Total: 55€

Nanopi R2S from official store plus case+heatsink: 22.30€
Shipping: 5.35 €

Total: 28€

And it looks better as you don't have the usb to ethernet adapter. Plus it has AES acceleration, so if you use openvpn it's going to make things faster.

1 Like

When you consider the improved specs of the RPi4 (including that benchmarks done by others seem to indicate it can shape traffic at gigabit speeds) then the price difference is minimal, especially if you end up upgrading your service to speeds the R2S can't cope with.

That's pretty subjective. I much prefer the look of my RPi4 case, even with the USB adapter, than the lurid yellow monstrosity that is the R2S. But others may have different views.

I will concede that currently performance on openvpn will be better. Whether that continues to be the case after the next major release which will let you use the chacha20-poly1305 cipher (as used by wireguard) to encrypt data remains to be seen. But then again why not just use wireguard instead...

Half the price. The title of the post say cheap router.

Nanopi is able to route full gigabit speeds, didn't know raspberry could handle better speeds than gigabit.

Me too, but good looking case for Raspberry are expensives. Again, we are looking for a cheap solution.

AES encryption is not used only for openvpn, there are other applications types that use AES.

I use both, but you can't use TCP with wireguard.
You can't use layer 2 for your vpn tunnel as openvpn.
You can't use proxies to establish the tunnel.

I know some of this things i say are specific to my use, and i know the RPI 4 is more powerful than the nanopi (i've got a RPI 3, a RPI 4 and two nanopi R2S), but the nanopi has enough power to be a good router that can handle gigabit speeds without problems and more. That's why i recommend the nanopi over RPI if the nanopi meets your needs. And i don't know how much RPI 4 can handle with SQM, but i dont think its not going to be a big difference to what nanopi can handle.

in australia... and elsewhere likely... compared to a rpi4-2G... you cannot buy these for less than 85% of the cost... accessories are also hard to source and quality is generally lower... returns on anything would instanltly make the purchase more costly than an rpi4... the code base / soc / support is not as well supported...

sure... if i could get one locally (or prevalently from overseas ) for the cost you assert... it's a nice little board. but simply having AES doesn't push the cons over the pro's for most people. openvpn at max gigabit speeds is not so relevant realworld... unless your doing business site to site or VPS...

is it a good option... yes... but all the other factors need to be balanced appropriately.

Totally agree. In europe you can buy the nanopi for the price i say before, i even bought two of them and the shipment cost were 7€, so each one cost me like 26€. The good thing of the nanopi is that you dont need any accesory, and even that openwrt for this board is an alpha, i tested it for a week and its rock solid without any crash or problem.

You have to evaluate and decide, what I was referring to is that it isnt a bad board although you compare it to the rpi.

1 Like

kinda strange point...(shows how local factors influence choice);

  • if I was installing 10 at customer sites... i'd choose the nanoR2S
  • if I was advising 10 friends... i'd tell them to get the RPI4

lol...

1 Like

So the RPi4 can be a decent router if need be? More powerful than a lot of consumer models like the R7800 etc?

Yes. Go to the performance results thread linked a few posts above, and read on.

With the correct USB/LAN adapter for the second port, the RPI4 is a powerhouse. It effortlessly handles gigabit throughput with shaping and many bells and whistles. If you add a VLAN capable switch, it's an awesome wired router. You can then add APs or repurpose whatever existing devices you already have to AP duty.

1 Like

Alright thanks all for the replies. I see you've been recommending the same options I had in mind. BTW I guess I could use an USB AP along with the RPI4 (or the nanopi), can't I?

Someone recently posted that they hit 700Mbit+ with SQM cake with the WRT3200ACM / WRT32X (same internal hardware):

You might want to message him to confirm. It's rock solid over 300Mbit for me on WRT32X with kernel 5.4 getting A+ bufferbloat ratings too. SQM cake, software flow offload, 97% bandwidth set. CPU utilization is low enough I know it can go much higher if my internet was faster.

I set my WRT32X to 97% of my max dl/ul bandwidth with SQM cake. I get A+ ratings for both quality and bufferbloat on dslreports.com/speedtest. Just saying because you can set it much more than losing 10-15% bandwidth like you mentioned.

The nanopi only has a USB 2.0, on the RPI 4 you would have another USB 3.0, don't know the speeds you want to get, but with USB 2.0 you could get something like 250-300 Mbps.

Can you post the dsl report? Yes, you can put more bandwidth but the bufferbloat will be worst. Try to make the test when downloading something with torrent and many peers, you will see that the bufferbloat isn't so good now. Dslreports just say that you have an A+ bufferbloat if your ping doesn't fluctuate so much, but that doesn't mean that settings the bandwidth to something like 85% it's not going to improve it.

Try to ping for example google DNS, and make an speedtest that saturate your connection, you will see that the ping to google DNS is increased, not by much. If you set the speed to something like 85% you won't notice this things.

The nanopi r2s has two gigabit ports, one of which goes through usb 3.0. I saw someone get 475 ingress 700mb egress with sqm, so I guess thats not speedy enough for me.

Mmmh, so the opinion that bufferbloat only matters on slow links is one often brought up. Personally I am not sure that I concurr, but I agree that on faster links the occurrence of conditions in which SQM would help is lower than on slow links. Whether that means a link is more usable with somewhat more bandwidth (700) and no SQM or with lower bandwidth (400) and SQM is something worth testing.
Personally I am on a nominal 100/40 Mbps link but shape it down to 49/36 Mbps, because even at that steep bandwidth loss the link is better usable for concurrent interactive applications from multiple users with SQM and my current router simply can not shape at a higher aggregate rate than around 80 Mbps.
I am not saying that that is the only viable option, but really that it is worth, even at higher bandwidth plans, to test whether SQM is worth the required cost in bandwidth.

Yes, I am not doing this because SQM needs that scarifice, but because my current router can not shape above 80 Mbps and I always configure for the bi-directionally saturating case, as I do not want my internet to flake out under load, so with ~35 Upload that leaves around 50Mbps for the downlink. With a more powerful CPU I would try something like 100/36 (I sync at ~110/387) with cake's ingress keyword for the downlink side. My point is that I happily throw away half my downlink since even at 50 Mbps SQM makes the link much better than a non-SQM 100 Mbps link. This was just intended as an example to demonstrate that bandwidth is not the only thing that makes a link useful :wink: it was not intended to show one needs to throw away half of the bandwidth to use SQM.

Well, unless your link is so fast you can never saturate it I disagree. But as I said, I would recommend for each user/network to simply test it out. Some people have such tolerant requirements that they do not see a meaningful improvement by SQM even at ADSL speeds, so it really is a policy question each network administrator should decide on. And I agree at 600 Mbps quite a lot will be happy without SQM. I try not to give blanket recommendations for or against using traffic shaping as a function of access speed as I believe this to be really hard to asses correctly from the outside.

1000/1000 I am jealous :wink:

1 Like

Indeed, shaping a gigabit is necessary for my satisfaction. I use VOIP for small business calls and now with COVID there are the zoom meetings for my wife and jitsi or slack for me.

My audio and video is always rock solid. Others not so much. It's not that hard to saturate a gigabit fiber. Two kids each watching Netflix or youtube, and a download or upload of a large file to Dropbox or Google drive while my wife is in a 120 person dept wide zoom meeting are all routine here

1 Like

I am building a high-end device for R&D purposes and the hardware I eventually settled on was PC Engine's APU4.
https://pcengines.ch/apu2.htm
Swiss company with fantastic support and several models to chose from.
The available anodized aluminium cases are nice as well. You can also get a rack-mount enclosure (that holds two modules) though that's pricey (~$200).

I unfortunately do not have SQM performance numbers.
I push a lot of data through Wireguard and it doesn't break a sweat.
This thing almost needs a 2/5/10 Gbps uplink - it does have USB 3.0 as well so that is not impossible.

If you buy one of the new WiFi 6 modules you can (finally!) get a dual-radio in a standard mini-PCIe package. These are pretty new so I'm not sure were support is at. Last time I checked a number of months ago there was only proprietary drivers.

We have a pile of features integrated and working with the 3-module device which is one-true PCIe and then two modem slots. You can get 2 to 4 jacks on the front for routing or switching as well.