July meeting notes
The meeting started promptly at 6:45 as per typical fashion.
Topic: OpenWRT and SQM
Alex started the meeting with his presentation and some history of the OpenWRT distribution
The basic features of OpenWRT is that it is user installable on your router you already own (if/when supported).
Also it has a web UI called LuCI and it has a package manager called opkg
OpenWRT releases seem to happen on an as needed basis and the latest release as of this meeting is 21.02.0
Currently Alex is using the 19.07.7 release which is what the presentation is based on
Next we learned about how SQM is supposed to work in theory.
The purpose of SQM is to help relieve Internet usage congestion by attempting to address the bufferbloat phenomenon caused by high latency in packet-switched networks caused by excess of buffering of packets
Bufferbloat is kindof like a bottleneck that can happen in many places along the network/Internet path your packets travel and often are made worse due poorly implemented code in stock router firmware as well as many other places including your ISP's network and even your cable modem.
There are ways you can diagnose bufferbloat, Alex used speedtest which actually said he was NOT experiencing bufferbloat causing slower upload speeds...
however.... He didn't have another option to explore and did not have enough to do with his free time so he decided to try SQM anyways
He discovered he could actually trigger bufferbloat in OpenWRT simply by enabling the experimental feature called "Software flow offloading" in the Routing/NAT Offloading section in the Firewall Zone Settings which most likely is an ASIC specific setting...
Next he discussed the LuCI interface for SQM "luci-app-sqm" which is limited in functionality but also because it tries to simplify the configuration but results in a mostly unknown configuration for SQM (unless you know what you are doing beyond just yolo selecting cake and clicking save...)
There is a download and upload speed setting in the Basic settings page. The doc says to set them to something like 15% less than your speedtest speeds.
The queuing discipline is the main part that is the most confusing. In the presentation Alex set discipline to Cake...
Also the Link Layer Adaptation page is :head_scratch:
After configuring the SQM Alex re-ran his speed tests and the speed results were slower for both download and upload and the bufferbloat rating was the same as before...
He then decided to buy a subjectively better router device (moar $$ and moar CPU, dual core ARMv7)
With the new router and no SQM enabled the router was showing a C score for bufferbloat
After adding the SQM feature to OpenWRT he got the bufferbloat score to an A+ :great_success: however the speed test was still kinda slow
Overall we conclude that SQM mostly doesn't hurt anything and makes you feel like you've done something to improve your internet connection speeds and serves as an effective Internet Placebo that may subjectively seem better than without it.