SQM is something you configure yourself. With 16/128 this doesn't look like a very future-proof device.
I hope that's not USD or as least no in US, or it would be seriously overpriced.
It's not.
The MT7621DAT chipset shouldn't achieve more than ~150 mbps with cake. Not future proof at all unless you don't have more than 150mbps speed. I wouldn't go lower than Filogic 820 (e.g. OpenWrt One) peformance these days, better Filogic 830 or 880 at higher speeds (500mbps+) if you want a All-in-one Router solution (minus modem) with addtional things running (OpenVPN, Wireguard, Adblock aso.).
@miels This router is intriguing because it would fill out a lower-price tier (US$69 on Amazon today). I would definitely like to hear how it performs with SQM to mitigate bufferbloat.
If you have time, please run these tests:
- Run a Waveform bufferbloat and speed test in the router's current state (without SQM).
- Save the "Share Your Results:" link that appears at the bottom of the page
- Install
luci-app-sqmand tune it as described in the SQM How To - Run the Waveform test again
- Post both "share your results" links here so we can see how it handles the bufferbloat algorithm
- Thanks!
Glad I posted something. This sure is interesting stuff 
I figured it was ticking all the boxes, except the layer added "future proof" apparently. I've just swapped my (over a decade) old Netgear WNDR3700 v2 for this one, and it feels like it's making a real difference!
Anyway, ran the SQM experiment right away!
I'm on a 400 Mbit up and down link via fiber (running adblock, packet steering is enabled).
For this test I've hooked up via UTP with a short cable, and disabled all radios. Connection should have been mine alone
Waveform results before
Waveform results after
I only did the "regular" SQM setup mentioned on the page you linked to (download speed 332000, upload speed 404000). Haven't done the " a little more tuning" part.
So enabling SQM seems to halve my speed, which seems like a bad thing.
Guess that's what you all mean by "not future proof"?
Seems like I should stick with SQM disabled, right?
[Update]
I reenabled the wifi and went back to my usual desk (with 30m cable and switch in between), ran waveform again without changing anything else. See results here. Download speed is now 198 instead of the previous 152
[/update]
1 Like
What great research! Thanks! Here are my observations:
- The Before test results are quite good. (368/449 down/up, 23/0 msec, on Ethernet, with a grade of A)
- The After test indicates that the SQM processing was likely consuming a lot of processor, and that slowed the data rate.(151/262 down/up, 9/15msec, on Ethernet, with grade of A)
- The Third test is a little confusing - I wasn't clear whether you were using Ethernet or WiFi in this third test... (198/257, 12/25 msec, still Grade A)
And my "conclusions"...
- The latency of bufferbloat is only important if it affects you. It doesn't seem to be biting you.
- I would be tempted to leave SQM turned off - you have a pretty good situation. If you're happy, I'm happy.
- As for the search for a good newcomer router, I don't think the AX-23 makes the cut, since it tops out at 150-200mbps with SQM enabled.
I really appreciate the careful research you did - it helps characterize the lower-price routers. Thanks again
PS I loved the WNDR3700v2's (mine were purchased in 2012 - a dozen years ago). That's where the original fq_codel algorithm was developed and tested.
These results (151/262 Mbps) are kind of puzzling as they seem to be slower than what a non-overclocked Archer C7 v2 with SQM could do:
- 288/495 Mbps with fq_codel + simplest_tbf, and
- 266/337 Mbps with cake + piece_of_cake
which is a much older device. Could it be that you didn't turn on Software flow offloading under Network > Firewall?
Not sure software flow offloading with nftables is still compatible with sqm... at least for some wan encapsulations nftables steals the packets before sqm's qdiscs see them...
mt7621a can still be a viable low-cost choice for an AP or as router for a ~VDSL (like) connection (so roughly speaking 100/40 MBit/s or maybe 250/40 MBit/s if sqm is not needed; thanks to hardware flow-offloading it can even exceed that for simple routing (as long as it can be offloaded, big if - sqm means no, it can't, PPPoE raises a question)), if prices are the major concern. But (imho) 69 USD would be way too much for such a device in 2025 (e.g. dap-x1860/ covr-x1860 (at least these two are easy to flash/ push-button recovery webinterface, albeit a little picky towards the used browser), wsm20 sold for ~15-20 EUR delivered 1.5-2 years ago - while that was a particularly good price, it still sets expectations).
Above 25 EUR, I would really look for something better - that may entail a more difficult (initial) flashing procedure (from OEM), but the devices are there. E.g. the cheapest I've seen, was ~27 EUR for a Xiaomi ax3000t (filogic 820) delivered, including customs and 19% VAT during a sale from Jack Ma's market place; given the recent'ish hardware variances (and Xiaomi closing the exploits needed for flashing OpenWrt) I would no longer recommend this device (especially not to a beginner), but again it does add reference points to the pricing. Likewise Walmart was selling off their Belkin rt3200 (mt7622bv) stock for ~40 USD ~1.5 years(?) ago, and the ZyXEL T-56 (filogic 830) is something from a completely different range (if you are within shipping range) - again, neither of these are easy to flash (and therefore fail the bar on this requirement), but capable.
Yes, ease of flashing and recovery options (including unobstructed serial console access, looking at TP-Link here, who like to cut the serial traces or pull them to ground) are worth a markup - the extent of which is a highly personal decision.
EDIT: I fully realize that my market rundown above is biased, in the sense that I'm remembering the lowest recorded prices -partially from sales- and not necessarily the regular going prices. And I'm obviously only following the listings that are tempting me personally (and I'm willing to accept difficult flashing, if the hardware warrants that), which is the reason why I don't have suggestions for easy-to-flash (but technically inferior or more expensive) options. Additionally the router market is surprisingly regional, in the sense that both availability and pricing varies wildly between countries (and especially the US and Europe, with Asia and Australia/ New Zealand being polar opposites), which makes general suggestions really hard.
I agree.
They sell Belkin RT1800s on US eBay for $40 (open box), was $30 a couple of weeks ago, still better than $69.
If you're willing to pay $69, you might as well grab the WRX36 off Amazon, it's $60.
You can still buy new E8450 dual-packs (called E8452) on eBay for ~$60, assuming you need more than one unit.
Then there's the MX4300 on w00t, where they're sold for as low as $18, currently supported by snapshots only though.
regarding my previous third test: it was via ethernet, but I turned the radio's of the router back on. So that might mean there is other traffic of IoT etc flowing.
And indeed, I didn't turn Software Flow Offloading on (didn't know I should)
Just enabled it, and went for another round of tests
Without sqm 345 / 398
Now enabled sqm:
test run 1 141 - 297
test run 2: 171 - 296
test run 3 192 -304
test run 4: 169 - 295
Changed SQM settings to fq_codel + simplest_tbf
run 1 294 - 377
run 2 311 - 400
run 3 264 388
run 4 299 - 387
Following the discussion closely!
I found out that the switch I have near my desk is still a 16 yr old one, 10/100 Mbit. So I was thinking that instead of buying a new switch I could use this TP-Link as an AP, and buy a better router that is more future proof.
1 Like
The MT7621 is good as a 5 port Ethernet switch. It will switch by hardware (no CPU involvement) between the ports at full gigabit line speed.
2 Likes
Link, please, a google search does not show that at all
Were, not are.
https://sellout.woot.com/offers/linksys-ln1301-wifi-6-router-open-box, doesn't show the actual price though.
Yep, that's the same thing I saw when I googled for it - sold out.
There's some cheap ones still on Ebay, though. But not THAT cheap.
What I have found is as a general rule, inexpensive in quantity <> Easy to Flash.
I frankly think it's a huge mistake to make Easy to Flash exclude "plug a 3.3v serial cable into the serial header"
The DD-WRT community is hugely into this business of flashing routers without serial ports, they only talk about using serial ports in context of debricking. Part of this is because they do so much of their stuff with Broadcom and virtually all Broadcom-based routers I've seen have the serial port holes filled with solder. If you buy a very tiny drill bit it's easy to clear those holes and then solder on a serial header, but if you try using solder wick or something to clear them almost always you end up ruining the board. I don't know if this is something Broadcom corp insists on router vendors doing when they use their chips but I wouldn't put it past them.
But I think encouraging newbies by showing just how easy it is to solder on a serial header then use the proper 3.3v serial cable is better, since that gives them access to a huge variety of super-cheap routers, and it's a skill they need for debricking also. It's a lot less painful to pick up a cheap router that requires soldering, and screw it up while learning, then pick up an expensive router that does not require soldering then screw it up while learning.
I am aware that there is a real focus on mediatek soc's duo to the good support for it. Was just wondering: how is qualcomm support these days? Are there qualcomm ax wifi devices that are being supported by openwrt? I seem to remember that these devices use ath11k?
Yes, ipq807x supports a quite large number of devices and ipq60xx is slowly joining the family in snapshots (ipq50xx is also in its early PR stages, but that will take quite a while - and we're also seeing the first steps of ipq95xx support for wifi7). These devices are well supported and work fine as AP, but given their inherent SOC design they rely to a large extent on proprietary NSS hardware offloading, which isn't available on OpenWrt (and most likely never will be), this isn't much of a problem for AP usage - but it does limit the maximum routing throughput and how far you can drive sqm at speed.
Again, ipq807x is well supported and works fine (thanks to robimarko and Ansuel) - but QCA is making it difficult (and the mainline support status is very spotty/ incomplete and largely untested by them), ethernet/ switchdev drivers are extremely basic/ unaccelerated and ath11k needs huge amounts of RAM for internal operations.
The major reason why filogic 830 is the current fan favourite, is because it's easier to support it (Mediatek is working mainline and with FOSS developers) and has mainline supported hardware flow-offloading and highly optimized/ offloaded ethernet/ DSA (switch) drivers, which makes them easier to support and break the 1 GBit/s barrier at full wirespeed (and mt76 needs much less RAM, leaving more of it to your needs). Quite a few of these devices offer solid specifications for quite good prices.
Just to add some perspective, (non-NSS) ipq8071a (this is the downclocked sibling of ipq8072a/ ipq8074a) will give you a routing throughput of ~600 MBit/s and 180-200 MBit/s with sqm/ cake - filogic 830 can route at 2.5 GBit/s wirespeed and sqm/cake should be well over 1 GBit/s. As AP, both can do ~700-800 MBit/s (HE80).
In general, QCA tends to be earlier to market and offers higher-end (and accordingly expensive) hardware (e.g. there are more ipq8074a device with 6 GHz radios and up to 10GBASE-T ethernet ports on the market than filogic 830 ones), but if you can't get full wirespeed out of it using OpenWrt…
What usually helps OpenWrt support is:
- global availability
- interesting features
- affordable(!)
developing support for a new device involves taking it apart (which usually isn't easy and often implies breaking quite a few plastic brackets), soldering on the board (serial console, flash), probing and working with the naked PCB for weeks to months, before it gets to a state where it gets into shape for potential merging. It usually doesn't come out neatly out of this process, while it may- or may not be broken afterwards, it is a sacrifice to support this model.
It's easier to do this with a cheap device, than a 400+ buck one. - sufficient hardware specifications
- FOSS driver support
- mainline support
- a chance to exploit all hardware capabilities to the max (at least sufficiently, this is where NSS is a major drawback - the capability is so close, but yet so far and will never be within reach)
It is always good to have multiple options, the current focus on Mediatek is natural - but not necessarily healthy in the long term and ideally we'd see a third competitor getting decent support (Realtek? SOC support is getting potentially easier, since lexra is history, wireless support has quite some way to go (but it is moving forward), but very low-end specs do spoil the fun; and QCA really needs to improve its hardware design and mainline support, to remain viable!) as well.
13 Likes
Thank you. One thing i (as a novice) never understood why it is/was neccesary to have separate builds for NSS support? For example, i'm using a couple of tp-link C2600 routers as AP. There are seperate NSS enabled builds available for it (on this forum). With NSS enabled, these are pretty powerful devices.
But now in your reply, you state that NSS support is a big problem? Again, i'm no expert, for me this is hard to understand when there are NSS enabled builds already available for ipq806x. Why can't there be nss enabled builds for ipq8074? How is qualcomm participating in offering opensource support for there AX products? I can remember that this was already an issue with ipq8064 / ath9k (with that special candela tech firmware).
Hopefully these IPQ95xx and IPQ53xx targets will provide some more device options to select from in the future.
(per https://github.com/openwrt/openwrt/pull/17725
and https://lore.kernel.org/linux-hardening/20250108-qcom_ipq_ppe-v2-0-7394dbda7199@quicinc.com/)
It's also critical to understand that it's NOT 100% the SoC that is the problem.
For example I have a Linksys EA7500v2. Mediatek, supposedly excellent support, etc.
For months I had 23.05.05 running on it just fine. Then one day I decided I'd flash 24.10 on it for beta testing. After all, VERY supported chipset, Blessed by the OpenWRT Gods, the whole thing - right?
So I did. After flashing it, rebooted the router. No problem. Comes up just fine. Second reboot - comes up fine. 3rd power cycle a few days later - SEEMS to come up - the blinkin lights working - but I/O is dead as a doornail.
Did a pin-reset AKA Factory Reset - nothing. Same as before. Does not come up on 192.168.1.1 as it should. So now, I have a brick and I'll have to take it apart and solder in a serial header and all that in order to see what's going on.
So much for using the "blessed SoC"