PreVue Channel also had an ad-delivery platform which served ads for the upper half of the prevue channel. The system was completely “headless”, using an A4000 that you could dial into, and upload new ads etc.
I developed that in 1993 — I think it was called “AdVue” commercially.
It was able to slideshow/carrousel the uploaded IFF/ILBM files or JPEGs, as I recall. I somehow managed to write a dithering algorithm for rendering as HAM8. I don’t recall how I chose the palette seed colors, as I didn’t know proper clustering algorithms back then.
I also somehow pieced together the “BBS” like Zmodem/Xmodem/etc. functionality for uploads. Long live Public Domain sources. This was pre GitHub ;-)
I’ve heard that the system was used for several years in both USA and perhaps Central and South America.
I lost contact with the company after going back home to Denmark.
The first time I ever saw JPEG images was from an Amiga magazine (Amiga Format maybe?) coverdisk. If I recall correctly my A2000 would actually take time to display the whole image - it wasn't instant.
I actually went and found the code. It wasn’t JPEG after all, it was Targa, but it still had to be turned into HAM8.
At that time, I had already worked with JPEG on the Amiga, for loading and saving images to/from the GVP TBCplus product. As I recall, it took about a second to load a JPEG image at SD video resolutions, but that was likely with a 68030 or 68040.
There is actually a small community still interested in poking around what little software related to Prevue we’ve found. Did you have any other involvement with the Prevue software?
That's correct. With many concurrent connections, you save memory (from thread stacks) and context swiches (since you don't need to switch thread to process each socket).
If all you want is a single request (and you have wait for results to continue work), you don't gain anything by going async.
> If all you want is a single request (and you have wait for results to continue work), you don't gain anything by going async
This is true for HTTP version < 2 because of the head-of-line blocking. With HTTP 2 I think a single connection will see a better throughput if not massive increase.
I think your project is really cool (perhaps because I've been working on a graph database (Neo4J-compatible) in Rust to learn it, so I understand your motivations...)
While doing that, I came across the 'sled' crate as a wait-free KV-store. If does blocking I/O, though, but it might serve as a starting point for working on the storage side of things. As it uses traits to express the interface, I'm not sure it could be made asyncable using the current language features.
I'll definitely been looking at how you did the async parts.
It's never too late to learn as long as you're learning something you're interested in. Period.
For me it was machine learning, which I started to study in 2016 (at age 43), despite having skipped math at Uni. But it was a lot of fun to learn!
Also, in my experience, you won't pick up a language by doing something unspecific. If you want to learn Rust, start a specific project which lies on the border of what you know; let's say - a HTTP proxy, preferable with some feature that you think is missing. You'll very likely fail to produce something more useful than what's out there already, but you'll learn more than what would by following tutorials.
Whether you should be pursuing an actual degree depends a lot on where you are and want to go -- but keep in mind that getting a degree is more than just taking the classes (learning environment, classmates, etc.). So, if you're primarily doing online classes alone, you might have a harder time than if you we're in a more typical university setting. So, find somebody in the same situation to use as a workout buddy!
Great response! RE: machine learning. Where would you say you are now in terms of expertise? Have you churned out some personal projects? Are you working on it professionally?
I'm working on it professionally, but not with huge datasets in a FAANG kind of place, like the MOOCs would have you wish for.
Once the ML promises hit the real world, where the demand for recognising cats is less acute, and datasets are much smaller (since they're so expensive to curate), it does get less sexy and glitzy. Specifically, I currently work on fraud detection at a government agency, using ML and graph databases.
We have some people who do the ML stuff full time, where I'm the back-up and sounding board (as in I'm the senior/mentor)
So, I wouldn't have been working there, doing that, if it weren't for my drive to learn on the side, I guess. ML is not my primary extracurricular interest anymore, but it feels good to know that I can code up a neural net, or discuss the tactics of building a model pipeline with (mostly) anyone.
I developed that in 1993 — I think it was called “AdVue” commercially.
It was able to slideshow/carrousel the uploaded IFF/ILBM files or JPEGs, as I recall. I somehow managed to write a dithering algorithm for rendering as HAM8. I don’t recall how I chose the palette seed colors, as I didn’t know proper clustering algorithms back then.
I also somehow pieced together the “BBS” like Zmodem/Xmodem/etc. functionality for uploads. Long live Public Domain sources. This was pre GitHub ;-)
I’ve heard that the system was used for several years in both USA and perhaps Central and South America.
I lost contact with the company after going back home to Denmark.