Over the course of the past year, we searched for a location devoid of wifi signals, so that we could explore, test, and benchmark wifi without interference from other signals. This proved remarkably hard. No matter where we went in the world, there was ALWAYS at least one AP with a competing signal on the channel. We gradually devolved into testing 5ghz only, where we could usually find clear air on some channel or another. 2.4ghz was simply not benchmarkable with any consistency.
Testing with competing wifi signals is definitely on the agenda, but first getting a clean setup was not optional.
Then summer hit, bloatlab 1 became too hot to inhabit, and campground with 110 acres high in the Santa Cruz mountains, asked if we could fix their wifi, so long as it didn’t cost them anything. SCORE!
So this site ended up being the site of the first-ever test fq_codel and quagga-babeld deployment on wifi and ethernet, on real hardware, in June of 2012. It was moderately successful, and feedback from various use patterns and tests resulted in a serious push to make cerowrt more stable than it was, and the fq_codel algorithm less memory intensive.
The June-August campground testbed consisted of 5 directional 5ghz nanostation 5MPs, 3 picostation 2HPs, 4 cerowrt wndr3700s, 1 cerowrt wndr3800. All the devices were upgraded to then-current openwrt + the current cerowrt kernels, excercised heavily, and various workloads analyzed.
The plan was to expand the network to a ring of 8-10 5MPs covering the entire property, 6 picostation 2HPs, and add a fully fq_codel’d set of firewalls and RTT simulators using donated vyatta hardware… as well as a parallel test backbone and produce a multi-exit-fq-codel - but money ran out in August, and we ran into some pesky bugs that are now mostly resolved as of the “sugarland” release of cerowrt.
A revamped design along the above lines is being put together for deployment early in 2013.
NOTE: To preserve the privacy of the site and the users, raw data regarding various tests will not be made available.