You know what they say. The first 50 percent of the restructure accounts for the first 50 percent of the development tokens. The remaining 50 percent of the restructure is approached asymptotically as tokens→∞.
Once you get outside Cloudflare's free tier, the Bunny setup in the original post beats it pretty quickly. Cloudflare costs $0.30 (Workers invocations) + $0.36 (R2 class B operations) per million requests. Bunny charges for bandwidth but not per request (!!), so assuming an average of 1kB/request (map vector tiles should be a bit less than this in my experience) that's $0.005-$0.06 per million depending on which region and tier you're serving from.
The real "too good to be true" might be GitHub Actions building a 100GB+ PMTiles file for free
I think PM tiles are quite a bit larger. Just loading at a low zoom level a view of the world triggers requests of ca. 500kB on my site, though that might also overscan a bit or include the file header. From my calculations, for Bunny to be cheaper the avg request should be at most 660kB at $0.01/GB which should probably hold true (I think just the worker invocations alone are already equivalent to 300kB of traffic on Bunny)
Huh yeah you're right, I was off by about 100 orders of magnitude. pmtiles.io shows tile sizes with "Show tile bounds" checked and they're mostly in the 30-100 kB range.
Bunny's volume tier costs $5/TB served, i.e. $1 gets you 2~6.7 million tiles served, probably closer to 3.
But Cloudflare also allows public R2 buckets so we could potentially ignore the Workers cost, i.e. $1 gets you 2.8 million tiles served. I think caching just works in front of R2? So you get significant gains if your visitors are mostly looking at the same areas and not zooming in all the way to their house every time.
So they come out pretty evenly or with Cloudflare a bit on top depending on the scenario. Bunny does has the advantage of easy data replication across regions (I don't know of a good way to achieve this with R2, although D1 has some kind of replication feature in beta).
Both are still far, far cheaper than almost any vector tile subscription I've come across. And, of course, more expensive than self-hosting (at least at the hobbyist level).
When you have a CDN in front of object storage it's not so easy to compare as the origin requests go out in chunks. For caching in front of R2, you need the worker, as Cloudflare only permits file sizes of up to 512MB regardless of the range requested. And then there's the latency difference of working directly with R2 (Protomaps calls out latency up to 500ms and on a previous blog post I measured 300-600ms on HEAD requests against R2[0])
The DuckDB devs measured it ~30x faster than PostgreSQL (and ~3x faster than Arrow Flight) for bulk transfer, which pg is known to be very slow at, and a bit faster (<2x) than pg for small writes up to 8 threads, after which they say that pg continues to scale well with number of threads but DuckDB plateaus.
That they've put it behind Cloudflare doesn't mean it's "served by Cloudflare"
Maybe Cloudflare gives them a good rate, or is donating service, but it's also possible or even likely that IANA is just using Cloudflare as a vendor, and therefore is paying for all the traffic, which is why they don't want people relying on it or hammering it all the time
In the nomenclature of RFC 2119, one MUST NOT design computer systems that rely on the correct operation of an HTTP server at [example.com.] Plus, it's *_really_* rude to pound on a small-scale service being provided as a courtesy.
> Oh, certain browser will not work with this blog if it cannot negotiate ONLY for Cha-Cha/Poly. It's by design as a showcase of why that particular web browser refuses to do that.
I assume the "particular web browser" is Chromium, which won't load it on any OS I've tried. On Windows, Firefox and the built-in curl.exe also refuse to connect.
Article makes it look like nothing happened in the embed/low-memory/single-threaded malloc space in decades since Doug Lea's malloc.
I just implemented TLSF for my minimal Wasm libc: fragmentation is just as good, performance is a lot more consistent (and on average better), for a significant reduction in code size.
Interesting! I have also implemented a version of TLSF for my (currently single-threaded) programming language. I also found it to be good for my use case (embedded / minimal code size). I'll compare it against yours. One improvement I did is some kind of preallocation for small blocks.
https://github.com/thomasmueller/bau-lang/blob/main/src/main...
I don't get what this is supposed to prove. That my browser doesn't allow websites to declare what encryption they use? I am pretty sure there's a point to all of this, but please let me (and my browser) choose which encryptions I want to trust.
Isn't the idea that the server and client should agree on a common subset? Here, the server's subset is very small, but not esoteric – so then surely it's the client that's lacking in features?
They were my first VPS. I've kept it running continuously the whole time; I have its IPv4 address memorized. Didn't think I'd ever want to give it up but here we are. Lesson learned, no more emotional attachment to infrastructure
Others have answered your question. A lot of the sites listed here https://anubis.techaro.lol/docs/user/known-instances/ have dynamic content; Git web interfaces in particular (Codeberg, the Linux kernel, FFMPEG, and more are on the list) are vulnerable to poorly or maliciously configured scrapers.
~51 in favor of the proposal. 20 of those specifically stated a desire for a blanket or as-strong-as-possible ban. 17 mentioned paying for the service. Many (sorry, didn't count these!) stated that the reason they joined SourceHut in the first place is because of the strongly opinionated culture and that an LLM ban aligns well with that.
~19 opposing the proposal to some degree. (I am counting Greg K-H's comments as a very weak "oppose".) 5 specifically mentioned paying for the service, and 1 stated they will stop paying if the policy takes effect. Some stated that they don't personally use LLMs, or don't like them or the effects they have, or would support a weaker ban (e.g. only ban fully vibe-coded projects or excessive resource usage).
You're missing the same point that the blog post is missing. What they're doing is much less like replacing rand() with 4 and much more like setting seed(4) before generating any numbers. There is no "loss of quality" unless you're already using a temperature of 0.
I know people disagree over definitions of "hard fork". Gram does not send patches to Zed but does occasionally merge fixes from Zed. There is not any specific effort to keep (non-AI) feature parity.
reply