I fell down a rabbit hole when I asked an LLM "why is Ink slow and Ratatui fast, and don't say 'because Rust'".
This led to a "10,000,000x faster" drop in replacement to Ink using a Rust backend. The goal here isn't to displace Ink but more of a bit of naive research into how you could make it faster.
I tried to distill the findings in this doc. Please don't use Ratatat beyond research, it's just an experiment.
As a pihole user for years I recently bought a firewalla blue. Installed pihole on the firewalla, turned off firewalla ad blocking, and done.
I can VPN to my home ad blocking network from anywhere, have more insights into my home network shenanigans, and still use my personal block list built over years. Super easy and most importantly, done.
I'm not sure I get it? Why not run OpenVPN on your pihole's RPi, forward the port on your router, bingo-bango-bongo?? What extra are you getting with the firewalla? Is it 'just' ease of administration (which is probably worth the price!)?
Did you research if the Pi could be setup with additional software (apart from pi-hole) to handle all that (or most of what) the Firewalla provides? Seems like that’d be a lot cheaper if one doesn’t need very high network performance.
I would agree with this. Get the latest model T series you can with your budget.
I would avoid the E, X, L series as they are either too cheap(in quality) or either too expensive (X series) due to thinness. The T series strikes the perfect middle ground and it well known for available replacement parts and upgradability.
I was part of a software project for BHS several years ago. We were taking the cad models for a entire plant and rendering them in the web with info/marketing media for each module.
The speed that they could sort material at different stages was incredible. I don't remember seeing a single "robotic arm" picking anything. It was all air and mechanical sorting. Very cool stuff.
They should have videos of most their modules on their site for the curious.
The first recycling robot was installed only three years ago.[1] There's been considerable progress since then. BHS is now installing and reselling the MAX-AI robotic picker, which seems to be a recent addition to their product line. This is out of the experimental stage and in use at high-volume sites. Expensive up front, about $200K each and $160K for installation and integration. Each one replaces about 2 people, plus the robots can run 2 or 3 shifts, so payback time is maybe 3 years.
There are some VC-funded companies in this area, but the vision/AI part is pretty much solved. It's getting the mechanical systems to be robust enough in a very dirty environment that's hard.
The big insight is that most of the separation has to be done by cheaper processes operating on the bulk material stream. The robotic system is just to pull out stuff that didn't get sorted correctly by screens, drums, shakers, magnets, blowers, and vision-based air jet sorters. An all-robot system is too slow and too expensive, but quality control robots for getting from 5% contamination to 0.5% contamination are cost-effective. Here's a good video of the whole process.[2] 70 tons an hour.
That makes it possible to get contamination levels down to China's new standards and the ones of US plants which can use the stuff. "Because of China's voracious appetite for all types of U.S. scrap exports over the past 20 years, MRFs "had the luxury of being able to sell the mixed rigid plastics [and] mixed paper without a lot of sorting and quality control." ... "Some materials are up in price … PET is actually increasing. It's a supply and demand thing," Butler said. "For the most part, if you can separate that material, there are domestic markets for that right now."[3]
So, in the recycling industry, some are whining about China's policies and others are fixing the problem, shipping products, and making money. It's small and medium sized cities that can't keep up. The big cities have big plants with multiple 70 ton per hour processing lines with the machinery running 2 shifts a day, and direct deals with companies that can use the plastics, metals, etc. A small town doesn't generate enough recycling to do that, and trucking unsorted material long distances is too expensive.
Very insightful and cool to hear about new tech making it into the MRFs. I was impressed back in~2014 so I bet there are way neater techniques now days.
Mechanical sorting (like shoe sorters) is very effective for discrete objects. Really cool to see stuff flying down a belt at 20mph+, get scanned/imaged from all sides, then thrown down one of 50 chutes without stopping.
As the first LR evangelist (there are many now) at my workplace I am always curious how you guys do what you do. Looks like pure magic.
I've poked around in the console a fair amount but would be really interested in two things you listed above:
- "We use the MutationObserver API for capturing diffs of the DOM"
- "We use Protobufs as a wire format for reducing bandwidth"
I know you all write pretty awesome JS articles on a regular basis but I have so many questions about how LR works! Spill the beans but not the secret sauce!
This led to a "10,000,000x faster" drop in replacement to Ink using a Rust backend. The goal here isn't to displace Ink but more of a bit of naive research into how you could make it faster.
I tried to distill the findings in this doc. Please don't use Ratatat beyond research, it's just an experiment.