I have a WRT1900ACS V2 router running LEDE r8391-162e016fed and am attempting to fine tune configurations this morning as well as add a couple of features. One of the first things I'm adding is SQM. I followed the recently updated guide, but I wasn't sure about one of the settings and results are not what I thought they'd be. Below are test results (after configuring as noted here), how I configured things, then my questions. To note, I have a fiber connection and tested from an ethernet-connected Windows 10 workstation without anything major running.
SQM-Enabled Test Results
Enabled for interface br-lan (WAN connection)
Download Speed: 809280 (90% of measured speed)
Upload Speed: 363600 (90% of measured speed)
Queue Discipline: cake / peice_of_cake.qos
Link Layer Adaptation: Ethernet with overhead
Per Packet Overhead: 18
First of all, I've seen some bad results in forums. With a Bufferbloat score of "B" without SQM, will SQM make much of a difference? I use P2P fairly heavily on one workstation if it makes a difference
The guide offered varying configurations for Link Layer Adaptation depending on connection type, however; didn't mention what it should be set to for fiber. Is my configuration of Ethernet / 22 appropriate?
Does anyone have any other suggestions for changes based on my results & config?
After posting, I changed the overhead setting to "22" (keeping it set to Ethernet) then retested. No dramatic change, though.
To get it out of the way, I did check for qos-scripts, but didn't have it installed.
Concerning the numbers, those were from converting Megabyte to Kilobit - not Megabit to Kilobit (always confusing for me) :). I tried using the below values which are 90% of the values you mentioned and reran the test...
Download Speed: 101160
Upload Speed: 45450
Everything looks great - except - that d/l has been cut in half...
Note: I reset Link Layer Adaptation back to Ethernet / 18 prior to testing. I also started tests a couple of times afterwards but got similar results.
Below are my settings. Yes, I am testing on a Windows 10 workstation which is connected via ethernet to the router. Now that I think about it, it's connected by ethernet, but does have a couple of hops despite being on a different layer (traceroute shows only single hop going directly to router):
Workstation - Gigabit Switch - Netgear Powerline 2000 - Netgear Powerline 2000 - Router
Never had issues or fluctuation with PowerLine adapters, but only internet traffic goes through them as everything is plugged into switch on workstation's end. Also verified no downloads being performed during any testing.
Below are my SQM configs - anything not shown is default:
My house is wired for ethernet, but port is on other side of the room of my office and wife is picky :). I can switch from Powerline to using it with long ethernet cable and retest if you think it would make a difference.
The only other question I have is that prior to SQM, I had ~15Mb/s more download overhead and 5Mb/s upload with a score of "B" for BufferBloat. I know there are many factors in determining the results, but assuming downloads will be that much slower from now on; is speed difference worth the tradeoff of an "A" rating or is the answer based on preference? Also, I'd assume that additional fine-tuning may close that gap. If so, I'm still curious about the Link Layer Adaptation setting or if you had any suggestions for where to start (if it would really help that is.)
You can also use etc1 for your interface I believe (I have a WRT3200ACM which have similar interfaces and use eth1).
As for that extra 15Mb down and 5Mb up, you can work to increase your download speed and upload speed until you find the sweet spot, though in general the 90% rule is there for a reason - you want a little overhead.
br-lan faces towards the internal network and wan towards the external network. The downloading/uploading monikers in the GUI actually are not in relation to the internet but only in relation to the interface sqm is instantiated on. It just happens that these two direntionalities are aligned for the typical case of instantiating sqm on the wan interface. We opted for downloading/uploading in the GUI instead of ingress/egress so that for the common case people's intuition would be correct (as compared to the alternative having to explain the directionality to everybody).
Yes sqm-scripts allows to trade-off latency-under-load-increase against some bandwidth, but you can control with what trade-off you are comfortable with. It is worth noting though that ingress/download shaping is a bit more approximate and generally works more reliably with a slightly larger bandwidth sacrifice than egress/upload shaping.
Well fiber is pretty much uncharted territory for me so it is hard to make recommendations...