Router that can do full-speed Gigabit qos?

[danjayh]

I have a highly asymmetrical comcast plan that runs at about 920 down and 25 up. I need QoS to manage my upstream bandwidth since I work from home (and have several other home PCs doing things like online backups, cloud uploads, etc.), and I'd like to use QoS to prioritize my work PC. Unfortunately, my aging RT-AC68U can't handle my downstream bandwidth if I turn QoS on because it disables CTF and FA, so I currently have to choose between giving up 2/3 of my downstream bandwidth or having a horrible upstream experience.

Just dropping a note here to see if there's anyone here that's aware of a router that can do 1000mbit+ with QoS on? I've been thinking of upgrading to 1.2gig internet so that I can get the next upstream speed tier (35mbps), so it'd be nice if it had at least two 2.5g Ethernet ports (one for the modem, one for the home network). I like Asus routers, but since I'm using a wired backhaul for my mesh, I'm pretty much brand agnostic (I can just set 'em all up as APs with the same SSID and password). I've considered the GT-AX6000 since it checks all of my boxes, but I'm not sure if even it is fast enough to do QoS at gigabit speeds. Anyone have any experience with finding something fast enough to get this working?

 

[Tech9]

GT-AX6000 may get close, but you need to use Adaptive QoS only. Traditional, Cake and Bandwidth Limiter are NAT acceleration incompatible and will cut down WAN-LAN throughput to about 350Mbps. No home router can do Gigabit with NAT acceleration disabled.

 

[danjayh]

GT-AX6000 may get close, but you need to use Adaptive QoS only. Traditional, Cake and Bandwidth Limiter are NAT acceleration incompatible and will cut down WAN-LAN throughput to about 350Mbps. No home router can do Gigabit with NAT acceleration disabled.

Are there any affordable options that can do it that aren't targeted at home users? I've considered setting up a PC to do it, but I'm a little hesitant to do it due to the increase power usage vs. a dedicated router... I was looking some of the Ubiquity EdgeRouters ... people claim that they can get close, but I have a hard time believing it since their CPUs are even slower than some of newer mass market routers.

 

[Paliv]

I can tell you the AX86U is good for about 750-850 with Adaptive QoS when I tested it last. I don't have issues that require QoS so I leave it off, that testing was a couple firmwares ago. So the AX6000 may get you close and has 2 2.5 Gb ports.

 

[eibgrad]

As I've mentioned previously, you're probably barking up the wrong tree is you expect *any* of these consumer-grade wireless routers to cut the mustard when it comes to fully exploiting anything in the gigabit (or better) world.

These OEMS are trying to "stretch the soup" as much as possible w/ their underpowered hardware and various hacks (CTF/SFE/FA), rather than VASTLY improve it. They want to avoid the costs involved for as long as possible. But that strategy is only going to last for so long before ppl wake up and look elsewhere.

More likely, you're going to need to consider MUCH better hardware from the likes of prosumer providers like Ubiquiti. Or else even the x86 desktop platform (e.g., pfSense).

We've simply reached the limited of what this consumer-grade hardware can do, and it's only going to get worse as ISP bandwidth offerings grow even better.

 

[Tech9]

Are there any affordable options that can do it that aren't targeted at home users?

x86 platform + power efficient + multi-Gigabit ports = expensive for home use. You have to sacrifice something to bring the price down. Home router is underpowered. Home built PC router is power inefficient. Ready to use appliances pre-loaded with pfSense, OPNsense or Untangle start from about $800. You need PoE switch and APs on top. You don’t want to know the price of managed multi-port multi-Gigabit PoE+ switches and high-end APs.

 

[danjayh]

x86 platform + power efficient + multi-Gigabit ports = expensive for home use. You have to sacrifice something to bring the price down. Home router is underpowered. Home built PC router is power inefficient. Ready to use appliances pre-loaded with pfSense, OPNsense or Untangle start from about $800. You need PoE switch and APs on top. You don’t want to know the price of managed multi-port multi-Gigabit PoE+ switches and high-end APs.

I was thinking a 3 or 4 gen old laptop with pfsense and a couple of 2.5g usb 3.0 adapters could possibly hit the mark with a budget of $250 or so... not sure what i need cpu wise ... 8250u based stuff is what i was looking at

 

[dave14305]

I like to refer to this post on another forum:

So you have 500Mbps-1Gbps fiber and need a router READ THIS FIRST

It's such a frequently asked question over the last few years that we really need a post we can point people to. So your cable company rolled out 500Mbps or 1Gbps download speeds, or you have 1Gbps symmetric fiber (GPON/EPON) from ATT or Centurylink or Orange or whoever. You realize your old...

forum.openwrt.org

 

[eibgrad]

BTW, if you like third-party firmware, you might want to consider DD-WRT for x86.

The reason I bring it up is because something like DD-WRT is going to be more familiar to many users than pfSense, Untangle, etc. The learning curve is probably going to be a lot less, and many developers (myself included) routinely support DD-WRT, FreshTomato, and Merlin w/ similar scripting. That's one of the reasons I'm a bit reluctant to dive into the likes of pfSense; I'm not quite sure just how configurable it will be outside the GUI, something I've come to expect for my personal customizations.

Like any firmware, DD-WRT has its pluses and minuses. For example, it only supports a single OpenVPN client and server in the GUI. But its support for PBR (policy based routing) is very powerful, including support for port-based routing (a little awkward to use, but it's there). And it's WireGuard support is second to none.

Or perhaps just consider it as a stop-gap solution on your way to the eventually full adoption of something like pfSense. That's what I'm likely to do myself, esp. since I have a boatload of scripts developed for the platform over on PasteBin. Full scale, deep end of the pool, adoption of something like pfSense doesn't quite appeal to me. Not yet anyway.

 

[Tech9]

I was thinking a 3 or 4 gen old laptop with pfsense and a couple of 2.5g usb 3.0 adapters

There are some issues with this idea:

1) The usb controller doesn’t know nor follow your interface assignments in pfSense. On reboot you may find your ports switched.

2) It works well with the built-in NIC as LAN and USB-to-Ethernet as WAN. You want the more reliable interface on your LAN side.

3) Not all USB-to-Ethernet adapters will work with pfSense. For Gigabit adapters I found ASIX based working better than Realtek. Amazon Basics marketed one is good.

4) Not a good idea in general in terms of reliability. I had a setup like this for about 6 months and it never failed, but my ISP back then was 300Mbps and I never hit high load on the NICs.

 

[danjayh]

I like to refer to this post on another forum:

So you have 500Mbps-1Gbps fiber and need a router READ THIS FIRST

It's such a frequently asked question over the last few years that we really need a post we can point people to. So your cable company rolled out 500Mbps or 1Gbps download speeds, or you have 1Gbps symmetric fiber (GPON/EPON) from ATT or Centurylink or Orange or whoever. You realize your old...

forum.openwrt.org

I guess I should clarify that my idea of an affordable router that can do gigabit is the $300 to $400 range. I fully understand that an average $100 consumer piece of junk isn't going to do it, I just like to avoid spending $1,200. I just find it frustrating that in top and routers we are still seeing a53 cores instead of a77 cores. If a quad core a53 at 2 GHz can't do the trick (as implemented in the gt-ax6000 and rt-ax89x), maybe a quad core a77 could. It seems that if laptop manufacturers can pack a fast dual band radio, an i5 cpu, ethernet, and a screen for under $500, router manufacturers ought to be able to make a similarly powerful router in a similar price range.

 

[dave14305]

I just find it frustrating that in top and routers we are still seeing a53 cores instead of a77 cores. If a quad core a53 at 2 GHz can't do the trick (as implemented in the gt-ax6000 and rt-ax89x), maybe a quad core a77 could.

The issue in many cases is that many QoS solutions are single threaded (like CAKE). So extra cores don’t translate to extra performance all the time.

I’m happy with my Raspberry Pi and my 600 Mbit connection. Very inexpensive back in 2020. Hard to find in 2022.

 

[danjayh]

So I didn't realize this until someone in a reddit thread pointed it out, but apparently there are now tons of mini pcs with 4x 2.5G NICs and Intel CPUs out there now, all in the 200-400 price range. Guess the next step is to figure out how much CPU it takes.

EDIT - so it looks like most of the x86-based router firmwares can do QoS on upload only (unlike Asus, Netgear, etc.), so even something as anemic as a j4125 w/ 4x 2.5G NICs ought to work in my case. This can presently be had on Amazon with 4gb RAM and a 64gb SSD for $299. Has anybody had any experience with these fanless mini-pc routers?

 

[Tech9]

It seems that if laptop manufacturers can pack a fast dual band radio, an i5 cpu, ethernet, and a screen for under $500

Consumer router manufacturers will give you just enough of what you need today. They expect you to upgrade in few years. Also, in AIO routers usually the AP part advertisement sells the device. Most customers don’t know or care about CPU or RAM.

 

[degrub]

So I didn't realize this until someone in a reddit thread pointed it out, but apparently there are now tons of mini pcs with 4x 2.5G NICs and Intel CPUs out there now, all in the 200-400 price range. Guess the next step is to figure out how much CPU it takes.

EDIT - so it looks like most of the x86-based router firmwares can do QoS on upload only (unlike Asus, Netgear, etc.), so even something as anemic as a j4125 w/ 4x 2.5G NICs ought to work in my case. This can presently be had on Amazon with 4gb RAM and a 64gb SSD for $299. Has anybody had any experience with these fanless mini-pc routers?

Check
Servethehome.com

 

[Tech9]

it looks like most of the x86-based router firmwares can do QoS on upload only

You don’t have much control of what the ISP is sending to you anyway. Most of the time simple Bandwidth Limiter is enough to keep the latency low. The issue with home routers is NAT acceleration is incompatible with BL and if you set it on upload it hurts the download traffic as well.

 

[Christos]

You only need QoS on the upload path (25Mbit).
Any router can do that.

 

[RMerlin]

It would be interesting to someday have a definite answer as to how much throughput your typical low power Intel CPU (Celeron or i3) could achieve. We all know it will be faster than any of the current ARM-based routers, but I never saw any actual numbers.

We need to coerce @thiggins out of reviewing retirement to do some benchmarking review of an x86 DIY platform

 

[sfx2000]

It would be interesting to someday have a definite answer as to how much throughput your typical low power Intel CPU (Celeron or i3) could achieve. We all know it will be faster than any of the current ARM-based routers, but I never saw any actual numbers.

Not as much as one would think...

Broadcom's HDN and Qualcomm's QSDK, in current SoC's, are amazingly performant

 

[RMerlin]

Broadcom's HDN and Qualcomm's QSDK, in current SoC's, are amazingly performant

Provided you use Flow Cache/CTF/SFE, indeed - the throughput can be impressive. These low-powered CPUs can then break the 1 Gbps barrier. But unless you have a special baked in QoS implementation like TrendMicro's bwdpi, you can't use regular QoS or even a modern SQM implementation like Cake.

We'd need Broadcom/QCA to either provide some form of API to allow a packet scheduler (like Cake) get access to flows, or for them to start integrating support for SQM into their NAT acceleration engine. Otherwise, the only alternative is going DIY with a higher performance x86 CPU. That's where the actual capable throughput remains a big unknown, beyond "it will be faster than ARM".

This might be one thing where IPv6 might greatly benefit performance-wise, as it doesn't require any NAT.