RTMP: web video innovation or Web 1.0 hackā€¦ how did we get to now? (Demuxed 2019)

It was fun to go back in time and recall why Flash was great in 2000, when IE 5.5 had just been released and you couldn’t rely on CSS actually working. In prepping for this talk, I worked really hard to try to express what Web development was like then and why people loved Flash: “200K of cross-platform goodness.” Flash made the Web work for high fidelity interactive graphics 20 years ago, which I think helped drive Web standards to support more than text, images and links.

“We wanted to support all the people on the internet.” It still boggles my mind how we could support low-latency way-back-then and now when computers and networks are faster it seems impossible… sometimes I try to visualize what is happening to the bits as I wait for something to happen. (I really do know why things are slower now, and it’s not just about the tech, but that’s a different story.)

HTTP tunneling worked much better than you would expect… sometimes it’s good to make something work for everyone, even if not optimal.

I fondly remembering Doug Engelbart telling me that his Augment system (built in the 1960s) had more features than Flash Player 6 + Flash Communication Server in 2002. (It’s great when your heroes tell you that your great accomplishments are not all that interesting.) Later, he did acknowledge that making this stuff available to everyone on the Web was “pretty good.” Inspiring widespread adoption, creating an ecosystem is a different kind of innovation.

The thing that makes it an ecosystem is that each essential component can be bought from multiple companies and is available as open source. At first Flash was essential, now much later, Flash doesn’t really matter anymore to the relevance of the RTMP protocol.

Today there are 500M IP camera on the Internet, about the number of people on the Web when Flash Player 6 was released. SmartHome video sensors have insane growth. The future of video is not about how to catch up with latency and resolution of live broadcast TV (though that will happen), it’s about how we can be integrate video streams from new devices, how we can help the machines help the people by creating new applications. Maybe RTMP will be a part of that, what do you think?

As a newcomer to Rust, my suggestion for 2020 theme is to fulfill the promise of “empowering everyone to build reliable and efficient software” by finishing what’s started (rather than adding new features), continuing the focus on good docs and good tools, and expanding to develop a coherent ecosystem.

Rust empowers you to reach farther, to program with confidence in a wider variety of domains than you did before. — Rust Language Book forward

Overview themes

2020 roadmap: finish what’s started, fulfill the promise

2021 edition: scalability – Can newcomers to Rust create a real-world, complex system without recreating basic components or contributing to the language itself?

  • more scalable systems written in Rust
  • experienced C/C++ engineers can easily transition to Rust
  • more scalable ecosystem
    • commonly needed libraries are available
    • new engineers can easily become contributors

Keep doing

  • Tooling is great! rustup toolchain, feature flags, online/offline docs make it easy to experiment with new Rust/crate features, even as a relatively new Rust programmer.
  • Transparency (like this call for blog posts, RFC process including roadmap)
  • Focus on good docs & good error reporting is incredibly helpful. Keep iterating on this!

Feature requests

  • safety beyond memory safety and concurrency. For example: URL parsing should be in std have a shared implementation that supports common use cases — it’s risky for Internet apps to not have a stable, well vetted URL parser, why are there three? (That’s a rhetorical question. I know why, but don’t think there need to be. See twitter discussion [Update 12/16/2019: I’d like to see convergence in URL parsers (or perhaps a shared common library) the way the community seems to be converging around serde.]
  • async all the things! I think this is already the plan. I look forward to async I/O (network and files) to be supported in the std library. I appreciate the thoughtfulness about safety, factoring out useful core concepts (like Pin/Task), and ensuring compatibility with Futures and Tokio crates. Consider other async use cases: GPU, OpenGL
  • lifetimes visualization would accelerate learning curve on resolving compiler errors, good ideas in this thread

Slow down to speed up

In my experience writing documentation often uncovers design issues and bugs. RFC template has a guide-level explanation section, which is great, and taking that one step further to writing baseline docs before declaring a feature “stable” would create positive pressure for community focus. Some ideas for process improvements…

  • A crisp “definition of done” could help focus the community. Consider adding requirements to releasing ‘stable’
    • RFC updated to reflect what was completed and is still open
    • stable reference docs are complete or include link to RFC
  • Consider WIP limits: how limiting work-in-progress increases speed

It seems in keep with Rust values to create a strong incentive to support contributing writers who are working to take the feature over-the-line and encourage new engineers to contribute. It is easier for new contributors to work with APIs that are documented or clearly dive into a work-in-progress, aware that they are contributing to finishing something.

Other Improvements

  • Documentation shouldn’t require deep knowledge of Rust (example: https://stackoverflow.com/questions/56402818/how-do-i-interpret-the-signature-of-read-until-and-what-is-asyncread-bufread-i/56403568#56403568)

Background

The reason I’m learning Rust is that I am experienced engineer with a need to write performant, low-profile client/server code. I’m excited about the idea of writing one body of code that can (potentially) work across native desktop, mobile, servers… and with cross-compilation to WebAssembly (Wasm), also browsers and edge servers.

Arguably, C works for all my needs, it even cross-compiles to Wasm. I want to like Rust better. I do in theory, but in practice, it’s got a lot of sharp edges (which is saying a lot when comparing it to C).


Updated Dec 2019 to modify the introduction, so excerpt is useful (if rust2020 summary post is ever updated. Original text below:

answering the Rust programming language call for blog posts as input to the 2020 roadmap

Caveat: I am new to Rust. There’s probably stuff I don’t even know about that is more important than anything here. This is just me doing my part to give back to the awesome Rust community.

Slightly modified vegan nut loaf to avoid tomatoes. Nut loaf recipe can be also be used to make burger patties, pan fried or baked, or can be stuffed into puff pastry for a puff pastry loaf or small hand pies.

Requires food processor to grind up various ingredients.

Ingredients

Nut loaf

1 1/4 cups (180 g) nuts (raw or roasted cashews, raw walnuts, raw pecans)
1/4 cup (33.5 g) sunflower seeds (or more cashews)
2 tsp oil (or 1/4 cup broth)
1 cup (160 g) chopped onion
4 cloves of garlic chopped
1 cup (96 g) chopped mushrooms (shitake or hen of the wood)
1 cup (140 g) cubed butternut squash
1 tsp each thyme, sage, rosemary, oregano
1 tsp smoked paprika
1/2 tsp black pepper
1/4 tsp each cinnamon and nutmeg
1/2 to 3/4 tsp salt
2 tbsp soy sauce or tamari
2 eggs (or 2 flax eggs for vegan)
2/3 cup (81 g) breadcrumbs

Sauce

1/4 cup roasted red peppers
1.5 tbsp soy sauce/tamari (or use 1/4 tsp salt + 1 tbsp broth for soyfree)
2 tbsp maple syrup
2 tsp apple cider vinegar
1/2 tsp (0.5 tsp) garlic powder

Instructions

Preheat the oven to 350 degrees F.

  1. Toast the raw nuts and sunflower seeds in the oven at 325 deg F (160 C) for 5 mins.
  2. Food processor: pulse nuts and seeds to a somewhat coarse meal
  3. In skillet over medium heat for 2 mins
    • oil or broth.
    • onion, garlic and a pinch of salt
  4. Add mushrooms and a pinch of salt and cook until some golden edges. (3-4 mins)
  5. Add butternut squash and mix in. Add a splash of water, cover and cook until the squash is tender.
  6. Mash and transfer to a bowl.
  7. Add in the spices, salt and mix in.
  8. Add the chopped nuts and seeds, soy sauce, eggs, breadcrumbs and mix well.
  9. Taste and adjust salt, herbs and flavor. ** The flavor will get more pronounced on baking. If the mixture is too wet, add a tbsp or so flour. You want it to be just slightly sticky. If too dry or crumbly, add a splash of broth. If you like sweeter, profile, add a tbsp of tomato paste or some chopped dried fruit such as dried cranberries or apricots.
  10. Transfer to a parchment lined pan. (I used lightly oiled cookie sheet.) Lightly press to shape. Do not pack too much.
  11. Bake for 25 to 30 mins.

Sauce:
1. Blend roasted red peppers into a smooth paste
2. Mix the rest of ingredients in a sauce pan over low heat for 15-20 minutes.
3. Taste and adjust as needed.

Take the loaf out of the oven. Spread the glaze over the loaf and then bake again for 20 to 30 mins. Let cool for 15 mins before removing from the pan. Then cool completely before slicing. Serve with gravy or cranberry sauce or both!