What's new

Best Router for Packet Priortization

  • SNBForums Code of Conduct

    SNBForums is a community for everyone, no matter what their level of experience.

    Please be tolerant and patient of others, especially newcomers. We are all here to share and learn!

    The rules are simple: Be patient, be nice, be helpful or be gone!

@Trip - congrats for reaching a post milestone...
*devil smiley* :)

You may have a point but with todays faster routers your point is melting away as technology gets better and faster. [...] On top of that I use a super fast Cisco layer3 switch which is definitely not under powered for a home residence. Way faster than any consumer router.
I would propose my point still holds true in the majority of cases, sadly, but I would also say that I hope you're right, as we hopefully see ARM chips, or whatever platform, becoming capable of routing in-software at line-rate ISP speeds, for the majority of embedded devices, hopefully in the next few years. My supposition, however, is that they will continue to lag behind, per previous trends. We shall see.

As for running a Cisco L3 switch at home, I think most would agree that a fair bit above the typical SOHO user's pay grade, albeit a very cool and capable bit of tech to have. Likely no bottlenecks in your LAN. :)
 
This. Times 1,000. Can't believe packet queuing didn't come into play earlier in this thread.

@OP, IMHO it's really queue management that you're after, not legacy classing/shaping. Find a platform that at the very least offers fq_codel, if not CAKE, as a qdisc, and your problems should all but disappear (at least for small, flat topologies, which is what we're dealing with 95% of the time on SNB).

As for "just buy more bandwidth": that doesn't solve the issue of a gateway's inability to egress all packet streams at acceptably low enough latency, jitter and packet loss. If you've got an under-powered engine and/or poor packet queuing platform (ie. most consumer routers out there), there's a high chance that buying more pipe will not solve latency/jitter to the endpoint on its own.

Yup, as I described in my opening post AC68u and 86u are supposed to have queue management, it doesn't seem to work properly... or at all. Only the QoS features do.

I've used DDWRT, which seemed to work pretty well with port and mac prioritization once you set it up, but DDWRT doesn't work with a 86u. This seems an awful lot like firmware implementations just aren't designed quite right, regardless of how much money you spend or the quality of the hardware, they just all around are 'meh'. The whole idea of building a router around latency sensitivity doesn't seem to be something companies are actually doing. They say they are (as it is easy to say a lot of things and slap some LEDs on it), but all they're offering is more of what we already have in a bug ridden state. I haven't tried DUMAOS as it seems like a giant marketing gimmick, but it's another specifically designed firmware and might help with the problems I'm having.

Surprised there isn't something that holds packets in queue and allows whatever XYZ rules are setup to bypass the queue almost completely. Maybe there is something that does this, but I haven't seen anything like this or it hasn't worked properly (using QoS and packet prioritization features). Almost everything is based around bandwidth, not time.

I really could care less about wifi performance.
 
Last edited:
There's no magic bullet for QoS/classification/delay. Don't forget that you only truly control half of the traffic - you have very little control over traffic being sent to you as it leaves your ISP and enters your network.
 
Yup, I understand agree with that, but as long as you aren't hitting your cap something isn't quite right with the consistency of delivery both ways. It always seems based on a first come, first serve basis. And depending on the type of traffic or where it's coming from, it definitely shouldn't be served first. This might be a very in depth problem that needs a hardware solution and maybe something that's supported on both ends (ISP and home router). At the most basic level while I'm no where close to my cap, I still run into ping problems just by multiple clients on the same network all try to use the connection. Where as if I have one computer using roughly the same amount of bandwidth I wont have nearly as much variance in a ping plot for instance.

It's like just the act of distributing multiple packets to multiple different clients causes thrashing. You can look at the router and see there is almost no load on it, so it has to be something at the firmware level or hardware. DDWRT with a worse router (AC68u) was better at handling it then my current 86u with Merlin, even though the hardware is a lot better.
 
Just to for something different you could use a layer 3 switch for QOS. There are lots of queues in my Cisco layer 3 switches. My guess is they would be faster than any small router but there is no NAT. Let the layer 3 switch order the packets and the small router just opens and closes the internet door based on the predefined order. I don't do this but it is available in the switch. I think the switches were built with VOIP in mind.
 
You could grab a cheap multi-port server network card, especially anything like an Intel i350-T4v2 or Intel Pro 1000 or HP NC364T (also Intel) and stick it in one of your "many computers". Fire up any router OS, like pfSense/opnsense/Mikrotik CHR/whatever and test to see if you can get the performance you want. They should also have improved visibility and tools for the type of network stats and metrics you want to measure, e.g. latency variance across routing interfaces.
 
I'm bumping this up again. I tried pfSense and it's a bit too complicated for me, it's more along the lines of using nix and without in depth knowledge becomes very archaic very fast. Out of the box install it performed worse ping wise then my 86u and has more jitter. Untangle I thought was something that might be what I'm looking for and I became very excited... Then I found out it doesn't have a MAC address clone option(?!?!!?) and I need that for my ISP.

Currently I'm once again looking for something that does this. Does anyone know if CAKE fixes these problems? I might just end up pulling the trigger on a NetDUMA router as it sounds like it's QoS is much better then everything else, but thats just word of mouth and is never backed up with any meaningful data. They're also insanely expensive and most don't even have 802.11ax.
 
Currently I'm once again looking for something that does this. Does anyone know if CAKE fixes these problems? I might just end up pulling the trigger on a NetDUMA router as it sounds like it's QoS is much better then everything else, but thats just word of mouth and is never backed up with any meaningful data. They're also insanely expensive and most don't even have 802.11ax.
Sorry for the almost 2-month old bump, but curious as to which way you ended up going for a router?

Also, from the looks of it, the Netduma R1 is nothing more than a MikroTik RB951G-2HnD flashed with OpenWRT and whatever packages (CAKE, etc.), plus a custom web GUI, as opposed to stock LuCi. The 600Mhz MIPS chip in there isn't going to be able to drive more than ~100Mb/s of CAKE-governed throughput, so if you want to utilize anywhere near your full 460/22 line, you'd either have to enable CAKE on upload only (which helps, albeit not completely), or you'll have to look elsewhere. A good compromise might be a Ubiquiti ER-4 or ER-12. They'll do ~400Mb of fq_codel+HTB "smart queue" and I believe manual MAC "cloning" is possible in EdgeOS (Vyatta). For much more throughput than that, you'll either have to pay for a pre-built business-grade firewall or source your own x86-based hardware, like a Protectli embedded box or a SFF refurb PC with a multi-NIC card in it. Any way you slice it, some skill and learning will be required if you want to route everything in-software and do so at line-rate speed.

One last point - when looking to max out the wired side of a network, it's commonplace to decouple wireless to discrete hardware. Consumer/SOHO all-in-ones don't scale as well for line-rate performance across the board, and/or in beefier networks. For the next level, you want to start thinking discrete components: dedicated gear for routing, switching and wireless. It will will yield better results in each area, plus easier upgradability in the future, and a great way to improve your skill set.
 
Last edited:
Separates are probably a better way to go if you are pushing the edge of things. If nothing else it is much easier to trouble shoot separates.
 
Yup, I agree, QoS doesn't really do anything unless you're nearing capacity on your line, which is why it doesn't really do anything for me or my usage scenario. Also understand how ICMP pinging works, some games don't use ICMP and use the packet delivery for estimating your latency. Having used the Asus implementation, it really doesn't seem to do anything and I get a lot of ping thrashing while in game (once again not measured with ICMP). Playing something like HOTS or Fortnite, the variance is all over the place.

Specifically looking for good packet prioritization. Also listing off features from a box doesn't really help what I'm looking for, I can do that already.

Are all new routers firmware locked like Asus now?

Timestamping.
It works out the true latency on the input route but doubles it presuming the latency on the return route is the same.
I dont get lag spikes.But i do get delayed gameplay on EA games. Other games are ok mostly.
 
Also, from the looks of it, the Netduma R1 is nothing more than a MikroTik RB951G-2HnD flashed with OpenWRT and whatever packages (CAKE, etc.), plus a custom web GUI, as opposed to stock LuCi. The 600Mhz MIPS chip in there isn't going to be able to drive more than ~100Mb/s of CAKE-governed throughput, so if you want to utilize anywhere near your full 460/22 line, you'd either have to enable CAKE on upload only (which helps, albeit not completely)

Linksys WRT1900acV2/ACS running OpenWRT is pretty stout on the Routing side with the SQM scripts installed - CAKE runs fine there, and on a 460/22 line, it can all be done in SW.

WRT's as Routers with OpenWRT are outstanding, and somewhat under appreciated... I would posit that they're better at this than Broadcom or Qualcomm IPQ80xx there.

I've always liked the hardware on those boxes, it was always the Linksys SW, and WiFi drivers for the FOSS builds that were problematic, and folks have put effort into the Marvell WiFi drivers with some levels of success.
 
The last Linksys anything I used was the WRT54G (in the first years of DD-WRT and Tomato); great initial platform for my first foray into third-party firmware. I've since moved to discrete routing, switching and wifi using SMB and enterprise gear, so I can't really comment on their current stuff from personal experience, although I'm sure with Marvell SoC's, they could be solid with the right software stack.
 
Sorry for the almost 2-month old bump, but curious as to which way you ended up going for a router?

Hah I didn't. Untangle doesn't support network MAC cloning (they're going on 11 years on that one). DDWRT supports no new router models (people are still recommending R7800), despite development for current models ongoing. Newer routers are coming out, but still have the same sh*t level firmware from decades ago. Pfsense is like managing a toddler and doesn't seem to give as good of performance as dedicated routers.

People like to talk crap about DUMAOS, but honestly that's what we're paying for? Good firmware and decent hardware. Sadly that's a crapshoot as well. Seems like they put the majority of their effort into their first foray and unless you want to pay $500 for their top of the line router you're probably going to have a miserable experience from the reviews I've read.

I still own a RT-86u, I had hoped DDWRT support would eventually make it's way to it, but owning it for over two years it hasn't happened. I actually just purchased a R7800 off of eBay so I can put DDWRT on it and get CAKE, which is supposed to be darn good from what I've read. So I bought inferior hardware to get superior firmware. I understand Merlin is trying and I do still have Merlin on my 86u, however it's not DDWRT and still occasionally needs reboots after a few weeks of uptime as it slows down. The manual/automatic QoS isn't as good as DDWRT as well, which is what I'm really after.

I'll probably resell it if my experience isn't beneficial, however DDWRT has been extremely positive before I went to my 86u and no longer can use it.

Couple things, when I talk about DUMAOS, I'm talking about the XR700 not their original PoS on their website. I sent them a suggestion asking them to consider going the route of DDWRT and have people pay for licenses for their own router, but that just got a 'that's a good idea, maybe in the future' and that was it. Also good thing about forums is they last years, so I'll probably be back here in the future still looking for a solution to a problem that still exists.
 
Last edited by a moderator:
I am still leaning toward using a switch for prioritization of packets especially after I setup my daughters' small business for 19 IP phones on her layer 3 switch I setup. The switch has 8 queues to use. I used the built-in Macros for the Voice VLAN. It works real well. There are no drop outs or funny words. The phones are doing their job as well as the switch.
 
i just want to point out that irregardless of internet speeds, you need QoS, as it can significantly improve things, such as your score on dslreports and experience when you have a lot of people or applications using it at once.
 
I am still leaning toward using a switch for prioritization of packets [...] The switch has 8 queues to use.
Priority queues do enhance FIFO, but they don't help quell buffering itself, which can and will happen at the gateway when egressing/ingressing packets out to a link with a highly differing speed (ie. DSL or low-speed cable/fiber). Now you may say the solution is to simply just "buy more bandwidth", and while that may be doable in many instances, it often isn't an option. That's where a good queuing algorithm and shaping ("SQM") on the gateway can come into play, and does tend to make a world of difference at avoiding the consequences of excessive bufferbloat on the WAN interface. I'm not sure what speed your daughter's business internet is, but chances are it may be high enough and/or her total traffic load may be low and/or sparse enough where the lack of SQM at the gateway (I presume the ISP unit or a Cisco RV?) isn't really exposed all that much enough to confidently state that it's needed. That said, for any points where enough simultaneous traffic is loaded onto a slow-enough WAN link, packet priority alone won't address the inevitable bloat; SQM will, or at least to a way larger extent.

This is what @System Error Message is referencing, what @dtaht would likely agree with (albeit explain in probably more accurate terms than I can) and what I maintain is required, in many instances, to truly deliver effective QoS overall.

TL;DR - Here's a talk by Dave on where we're at with this as of about a month ago.
I found the comments on Ubiquiti, Mikrotik, Cisco, Juniper and Arista to be interesting, too.
 
Last edited:
I just have not seen that kind of slowness now days. Yes, 15 or 20 years ago when we were running 5 meg DSL. The DSL was a huge bottle neck in those days. And there were not any good priority systems back then. I tried the one in Smoothwall back then and it cut my internet speed down to 3 meg.

Yes RV340 router.
 
Last edited:

Similar threads

Latest threads

Sign Up For SNBForums Daily Digest

Get an update of what's new every day delivered to your mailbox. Sign up here!
Top