Openwrt SQM set up for Xfinity docsis 3.1 connection.
Hey guys, so I’m pretty new to this openwrt stuff. Only about a year in or so. Before I thought gaming routers were the “key” to fight bufferbloat but realized it was all just gimmicky. My friend pointed me towards openwrt and cake has been somewhat good for me. So I have a question for the link layer adaptation tab, it asks for per packet overhead. I have a cable modem connected by coaxial cable and then it’s Ethernet cable wired to my openwrt router. I’ve read that if my isp does standard dhcp I should use Ethernet with overhead with none as the per packet value but it’s hard to really determine what’s the correct answer. I was just wondering if anyone has any insight on what could be optimal. I’ve tried values of 44,42,18. And by the way my connection from Xfinity is 2000mb download and 350mb upload. Some insight and tips on how to better my connectivity with openwrt would be very much appreciated. Thanks guys.
So on docsis 18 is the correct overhead to specify (unless the ISP uses ds-lite), but are you sure your 2000/350 plan is implemented via docsis and not via some sort of fiber link?
P.S.: Getting a router with a beefy enough CPU to shape 2Gbps is a bit of a challenge
P.P.S.: on docis consider ACK-filtering and setting an suitable MPU value
Please connect to your OpenWrt device using ssh and copy the output of the following commands and post it here using the "Preformatted text </> " button:
Remember to redact passwords, MAC addresses and any public IP addresses you may have:
Maybe not... the often not even enabled PIE AQM in upstream direction is not terrible, but it also is not flow queueing nor IP fair... Personally I would consider that (if actually enabled by a docsis 3.1 ISP) a nice back-stop, but I would not rely on this. The bigger issue is that this only mandated for the typically much narrower upstream direction.
CableLabs later developed low latency docsis (covering both directions) however I would not touch with a 10 ft pole, this is not a well engineered solution, but IMHO engineering by hopes and prayers... Caveat, that is my personal opinion, and others consider it decent to the best since sliced bread.
Which claim? That it is not generally enabled or that PIE is inferior to cake, or the one about low latency docsis being badly engineered?
Regarding the first I got told by people on advertised docsis 3.1 plants that pie is not active, so whether you believe me or not, I am convinced this is not universally enabled...
PIE being inferior to cake, simply compare:
PIE:
cake:
For low latency docsis being not a marvel of engineering, consider that at its core it espouses L4S, but L4S itself is too little too late:
Yes it is a 2000 and 350 connection. I am in a mid split area where Comcast upgraded our speeds and currently I’m using tbe nano pi r6s it can handle 2 gig. I also got an x86 intel n100 machine that I might give it a try
Honestly the AQM by Xfinity sucks so bad on downstream… it’s pretty okay on upload. Very good numbers that are kinda consistent thru tests. But I was under the assumption that if they already do aqm and I’m using sqm on my router it’s just extra processing… maybe I’m wrong. Idk.
ACK filtering on egress only, the MPU should be set to
From: https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm
For DOCSIS Cable - Choose Ethernet, and set overhead 22 (mpu 64). For rates > 760 Mbps, set overhead 42 (mpu 84), because 1 Gbps ethernet between modem and router affects the worst-case per-packet-overhead.
I would however believe that overhead 18 would be enough, the 22 is based on the theoretical chance an ISP might use a hidden VLAN tag, but as far as I know that was never actually observed.
Not sure how to deal with the switchover from diocsis being the bottleneck to ethernet being the bottleneck for small packets, what ethernet speed are you using between the docsis modem and the OpenWrt router?
They likely use PIE, if that delivers latency that works for you great (do not let my obvious dislike for the docsis AQM/traffic sheduler color your own experience. I follow Rich Brown's rule of other people's networks, if you are happy with it, I am happy too )
What do you mean what Ethernet speed? My wan port for the router is 2.5G and my docsis cable is the new cm3000 from NETGEAR which is a 2.5G port as well.
I really don’t like the aqm from Xfinity it’s the worst thing ever for internet. Especially for docsis. I just know when I set overhead to 44,42 things seem sluggish especially when I’m playing an online game. 18 felt very fluid for me. Idk how this correlates with gaming. Also on 18 overhead I seen lower jitter numbers but I didn’t test enough to see if there’s proof of consistency there. I just want to know what you guys think of what the best config of sqm would be.
Here is the thing, 1 Gbps ethernet operates at a gross rate of 1 Gbps with an MPU of 84 bytes (minimal frame size 64 plus 20 bytes of overhead outside of the frame), docsis standard has an MPU of 64 and no additional overhead, so with minimally small packets 1 Gbps ethernet becomes the bottleneck at around 760 Mbps and for rates higher than that one should account the ethernet numbers...
Actually, personally I would just do exactly that and set mpu to 84 and overhead to 38 and not bother much further...
Yeah that should not be the case though, configuring the overhead larger is essentially equivalent with configuring a smaller shaper rate (assuming maximally sized packets) so this should not affect perceived fluidity...
As I said at those rates, I personally would configure the ethernet values and ignore that under most traffic mixes that will be to conservative and waste some potential throughput, as at 2000/350 Mbps I simply would not care enough to even calculate the throughput sacrifice any more precisely.
Ahhh I see. I would try your values as suggested. But if anything 18 overhead would be the best correct? Also any other advice I should take in for SQM set up or for network connectivity? Perhaps ecn suggestions.
)