Information on the SourceHut outage

Let’s break it down:

* “URL for getting discovery” … I am not sure if I understand what you mean. Does anybody search for a project by going to GitHub.com and search in that search box? I always come from somewhere else (PyPI, package in my distribution, somebody mentioning a project somewhere). For read-only browsing repo, one could use GitWeb/cgit or something of that kind.

* bug reports … this is a real problem. I was playing with hosting my own bugzilla, but it doesn’t work, because nobody in their sane mind creates a new account on yet another tiny website they don’t care about. I guess, filing reports to email list is the only available option, because it gets sufficiently low threshold for access. The ticketing system probably would be just a read-only website (with bugzilla or anything else, see https://todo.sr.ht), or perhaps the future lies in git-bug and its web interface to display bugs.

* getting access as a new user … if the preferred communication channel is email, then this whole question goes away.

* PRs … git send-email(1) and an email list?

* releases … some static website?

I don't think a fully distributed system is easy to use. Look at Mastadon for how that approach fails to scale outside of the tech audience since there is no simple answer to the question which website should I go to which is how people think about it. However you can make it more distributed than say GitHub for sure. https://docs.pagure.org/pagure/ for instance stores a lot more of the associated metadata data (tickets, prs, doc etc) directly in that repo so when you clone it, you get all of it.

The problem with that is that only a very small percentage of the population has the technical skills or financial resources to be able to set up that kind of hosting capacity, and even then the scale of hosting that they can support is extremely varied. Mastodon is a great example of that - sure it's not (that) hard to set up your own self-hosted server, but very few people can set up a server which can handle enough traffic to be useful for much, so most people use a few big centralised servers.

Git is inherently well suited to fully distributed hosting, because every single git repo (as a fairly good first approximation) has *all* the data you might want to host - by its very nature, the data is already in lots of physically independent buckets everywhere there are people interested in that bit of software. The problem is, those buckets are almost all in locations where they can't be involved in any kind of public access - they're on people's personal or development machines, or spread across servers in support of deployment tooling, and so on. So it's distributed *storage* capacity, but not distributed *hosting* capacity.

The closest thing I can think of to that kind of distributed storage capacity is hosting at the individual ISP level - effectively having each individual pushing the data up to their own ISP, who then provides a hosting service for it. Which is kind of already a thing, since most ISPs do provide some level of hosting service for their clients, but it's not the kind of thing that can scale up to the level of a sizeable forge replacement, definitely not a replacement for something on the scale of github, and probably not even the scale of a smaller forge. And there's a simple reason for that: hosting costs money and resources, ISPs aren't making money from it, and therefore they don't offer it as more than an afterthought or as a small-scale paid service.

I can't think of any alternatives to ISPs as a source of truly broad-scale distributed hosting - if anyone has suggestions I'd be interested in hearing them.

The key problem is really that there are *massive* economies of scale in the hosting business - the incremental costs of adding capacity to an existing data centre are insanely small compared to the initial cost of the data centre itself (up to a point, obviously), so lots of smaller data centres is a lot more expensive than a few big ones, and "truly" distributed "data centres" (one for each individual) is scaling massively in the wrong direction. So any time you want to scale up a service, there's a very strong cost push towards big centralised hosting - it's just vastly cheaper that way, as well as being easier to implement and manage. We're not going to move away from that any time soon, and probably never - this is pretty much inherent to the nature of the technology.

But when you actually look at the world, there are *lots* of data centres out there, of various sizes and providing various levels of service - we probably don't *need* to distribute the actual physical hosting capacity much further than it already is. The bigger problem for the kind of distributed service that you're talking about is the fact that very few people have the *practical capability* to use that hosting capacity - the technical skills, the financial resources, the willingness to jump over the various structural hurdles that are put in their way by the modern hosting service model.

ISPs generally need their own data centre presence just to be able to do normal business, which is why they're probably the first choice for getting around that - anyone with an Internet connection already has a paid contract with an organisation that has the technical capability to provide hosting. If the goal is ultimately to allow *every* individual to be able to push their own little bucket of shared/shareable data up to an external hosting provider then this is surely the best starting point. But for ISPs, hosting services are going to be a money sink, as well as exposing them to extra risks (apparently random DDoSes being the perfect example) - they need a revenue stream to offset that. If you want it to be universal then that revenue stream needs to come out of everyone's pockets, which then makes everyone's ISP bills higher; for most people who aren't particularly interested in hosting services (except for the kind of thing they post on Facebook or similar) that's going to be unacceptable. All of which adds up to ISPs not having any reason to provide hosting services for their users, even if they technically could do so reasonably easily.

The gist of it all is that the way the world is structured (right now, at least) simply doesn't make truly distributed systems viable, certainly not compared with a centrally hosted version of the same system.

The only way I can think of to change that would be to change the way that things are paid for - if it was possible for the hosting provider to get money from people's *actual use* of a service (rather than having the people running the service get that money, and then use some of it to pay the hosting provider) it'd be far more viable for organisations like ISPs to get into the hosting game. In fact, if that was possible then it would provide financial incentives to move towards distributed hosting - with a centralised model all that money would go to the centralised provider (Facebook, etc) rather than the people who actually created the content that the centralised provider was hosting. With the kind of usage payments I'm thinking about, distributed hosting would see those usage payments going to the right place.

But it's easier to *say* "get money from people's use of a service" than it is to make the world actually work that way. A *lot* easier.