so iv been messing with routers and bufferbloat and researching for some time now. im a gamer i also stream on twitch i mostly play valorant so low bufferloat is a must for me. im looking for a router/hardware to help me with that. here is what i have right now.
i have xfinity 1200 down 40 up paired with a arris s33 modem.
i also have multi routers in my home 3 floor house 8 ppl live in my house and about 30 devices connected.
my routers are....
asus GT-AX6000 as main router with merlin installed using flex qos with this router i get 1400 download speeds and my pc has a 2.6gig card
then i have about 4 AP routers around my house also asus for easy mesh all hard wired aswell
RT-AC5300 RT-AC68U and 2 RT-AX92U
flex qos works good but not perfect for me i set the Bandwidth like this 22.7 upload 600 download when im downloading a huge file qos kicks in and my GT-AX6000 cpu core 1 gets to 99 % i feel like if this keeps happning my router may die sooner. im looking for a new hardware that can help me with this. and started to look into openwrt and see if this can help me with bufferbloat i just started researching raspberry pi 4 with open wrt but took a break from my head hurting from researching so much lol i own a raspberry pi 4 B with 8 gig of ram also have 2 usb lan ports NS-pu98635 and a 2.5g usb-c2500 would the raspberry pi work well or should i look into some better? now my GT-AX6000 is under warranty from bestbuy so i can exchange it for something els from them if need to be.
Depends a bit on what future speed ups are possible given the infrastructure, but I'd say medium (contemporary[0]) i3 to low end i5 (or any ryzen) - I'd just skip the Atom based systems here (even if they might just barely cope) or the AMD Jaguar SBCs (pcengines APU) (which would be totally out of their league).
For reference, my ivy-bridge celeron 1037u can do sqm at 1 GBit/s symmetric (synthetic benchmark done locally) easily and extrapolating its (CPU-) performance suggests that ~2.5 GBit/s (with according network cards) should be possible as well.
The discontinued (Atom based-) ODROID-H2+ or the upcoming (rk3568b2 based) NanoPi R5S would be a bit too marginal for my taste, especially with avid gaming for 8 persons in mind.
--
[0] contemporary more refers to sane power consumption/ noise levels, than strictly performance related.
>1 GBit/s ethernet cards are pricy (and we aren't even talking switches yet), that's what you pay for most. Mainstream mainboards were just supposed to gain 2.5GBASE-T cards, when the supply chain problems kicked in and deferred this goal. As addon (PCIe-) cards you can still tally ~100 USD/ EUR per card…
Modern Atom isn't bad (at all), but still an in-order architecture, while i3 would be out-of-order. Yes, you are comparing 2013 vintage haswell against 2021's jasper-lake atom - and networking is more in-order, than out-of-order load, but it's also inherently single-threaded, which is why I'd still be more at ease with haswell than jasper-lake.
I would assume that the N6005 would cope with that load, but without a hands-on comparison, I'd err on the side on non-Atom for beyond-1-GBit/s tasks with sqm.
This kind of asymmetry is a real performance killer. Typically the ACK volume on 1200Mbps is going to be pretty much 100% of your upload speed. This means if you're downloading at full throttle you can't do anything else at all.
I'd suggest stop looking at multi gigabit gear and just set your download speed to say 800Mbps and upload around 40 or a little less. You will be happier and be able to do this work with RPi4 or cheaper x86 gear. Seriously the above 1000Mbps speeds just aren't worth it for a home user yet.
i didnt really care for the download speed much, i had to get those speeds for the upload speed. thank you for the info going to keep looking around the fourm and google.
As a rule of thumb TCP Reno will send one ACK packet for at least every two MSS sized segments received, with an MTU ~1500 Byte and Ethernet this results in reverse ACK traffic taking ~1/40 of the forward data traffic, so in the OP's case 1200/40 = 30, which is the lion's share of his upload.
However, newer TCPs and offload techniques tend to send considerably fewer ACK packets, and (some) cable ISP employ ACK filters that also thin out the ACK stream of more chatty TCPs. But still this is a rather user-unfriendly asymmetry... so using a traffic shaper to push this down to something saner like 600/40 makes a lot of sense!
According to a number of posts in this forum a raspberry pi 4B (with an additional TP-Link UE300 gigabit ethernet dongle, and an AP with switch) should be good for up to 1 Gbps internet links with enough CPU cycles to spare to actually do something at that rate, like traffic shaping. While I have not tested that myself (my pi400 got immediately repurposed as light browsing computer for one of the kids during tele-teaching covid days, and the 4B immediately went to do duty as GPS-disciplined NTP time source) I fully trust @dlakelan and all the other's that have tried it.
It will work fine at the kinds of speeds we're suggesting. It isn't even idling at 40Mbps upload, it can handle 900Mbps download for sure. I use it on a 1Gbps symmetric line with good results. You want to enable packet steering and install irqbalance.