4 ms·

Immutablity doesn't scale. Users like having experiences. It's much easier to build something like Google/Uber/AirBnB on GUN/CRDTs than only-hash based or app

Immutablity doesn't scale. Users like having experiences. It's much easier to build something like Google/Uber/AirBnB on GUN/CRDTs than only-hash based or append-only protocols. End users don't use something for the protocol, they use it because the app developer created something that gives them value. We need to stop pitching decentralization to the fantasies of developers, and start giving value to makers, shippers, and users.

lijogdfljk 7y ago
> Immutablity doesn't scale. I imagine you're thinking of specific implementation(s) of immutability. Regardless, my use case is not that of Google; my use case is of the person. It's of being able to share, own and read information with trust. > Users like having experiences. This feels like a straw man. I very much supported UXs in my post; I even said I love bloated feature rich user experiences. I don't think anyone intends to entirely give up feature rich web. We often just want options. Right now we only have the mutable, feature bloated web. While I love it for applications, it seems terrible for information. > It's much easier to build something like Google/Uber/AirBnB on GUN/CRDTs than only-hash based or append-only protocols. Quite possibly; I do not claim to have all the answers. What I have is a desire. A desire to be able to refer to content, and not have someone change the meaning of that content. In the age of misinformation I don't think we have a choice. In the same way that a signed binary can be important, I think signed / immutable information is important. Do you disagree? > End users don't use something for the protocol, they use it because the app developer created something that gives them value. Agreed; which is why I was talking about value I see in the features of immutability I described. As well as the anti-value (is that a thing?) I see in the state of the mutable web as it stands now. > We need to stop pitching decentralization to the fantasies of developers, and start giving value to makers, shippers, and users. Saying decentralization is a fantasy seems in spite of some of the most successful technologies in human history. This is odd to me. edit: as well as new federated software embracing decentralization. You being against this is perplexing to me.
- marknadal 7y ago
  Granted, there are certainly worse implementations of immutability than others, but immutability as the foundational architecture (even in its best design) creates bottlenecks. Yes, a person owning their data should mean they can change it! Fair point on UI/UX, you got me here - you are right. It is trivial to build immutability on mutability, this is what MD5/SHA checksums do all the time on things like downloads. But if your base system is immutability, it is hard to do anything else well. To your question: I agree signing is important. Doesn't mean it needs to be immutable! Very well said! I'm impressed with how both precise and concise that was. Good point. Oh sorry, decentralization is great, just its target use case should be "developer fantasy" but "user value add". Sorry for the confusion here. I'm pro decentralization (I build p2p protocols).
  - lijogdfljk 7y ago
    > Yes, a person owning their data should mean they can change it! Well being able to change data is of course needed; so that goes without saying (I should hope). However what I mean by immutability is rather that I think the web should allow me to link to a piece of content, and ensure that what I link cannot be changed from underneath me if I choose not to. I feel like I'm describing features and you're describing implementation. To me, immutability is a feature. I'm not at all referring to blockchain or any of that junk - merely that I want to be able to link (in the web sense, url/etc) to content - to discuss it with other people and etc, and have that content not change. Optionally of course I should be able to link to mutable content. However we have mutability down. The feature add I'm talking about, is immutability. > It is trivial to build immutability on mutability, this is what MD5/SHA checksums do all the time on things like downloads. But if your base system is immutability, it is hard to do anything else well. Sure, but that's an implementation detail. I was not (or did not intent to, at least) say that the "base implementation" must be immutable. All I care about is that immutability is a first class citizen, and linking immutable content is possible. To build user features around immutability, where as nothing like that exists in the current "web".
    - marknadal 7y ago
pfraze 7y ago
Discussion around the technical side isn't developer solipsism. It's part of the process for product design. Append-only logs do scale. Dat 2.0 uses a hash-trie structure inside its file metadata log to act as pointers for fast lookups. Partial replication means you don't have to pull down the full history. Mark, you work on a competing product to ours. I'd appreciate it if you didn't FUD our work. We don't do that to you.
- marknadal 7y ago
  DAT is one of the better protocols because it has evolved its approach and is starting to adopt CRDTs. I'm FUDing append-only logs and immutability, not products - don't conflate the two, or else you discourage legitimate discussion around the scalability of different architectures. Question: Do you have to still store the full history? How would you explain scaling something like Uber with this model, then? You're gonna have 100s of GPS coordinate updates per second, for each car, potentially millions of them.

lijogdfljk 7y ago

> Immutablity doesn't scale. I imagine you're thinking of specific implementation(s) of immutability. Regardless, my use case is not that of Google; my use case is of the person. It's of being able to share, own and read information with trust. > Users like having experiences. This feels like a straw man. I very much supported UXs in my post; I even said I love bloated feature rich user experiences. I don't think anyone intends to entirely give up feature rich web. We often just want options. Right now we only have the mutable, feature bloated web. While I love it for applications, it seems terrible for information. > It's much easier to build something like Google/Uber/AirBnB on GUN/CRDTs than only-hash based or append-only protocols. Quite possibly; I do not claim to have all the answers. What I have is a desire. A desire to be able to refer to content, and not have someone change the meaning of that content. In the age of misinformation I don't think we have a choice. In the same way that a signed binary can be important, I think signed / immutable information is important. Do you disagree? > End users don't use something for the protocol, they use it because the app developer created something that gives them value. Agreed; which is why I was talking about value I see in the features of immutability I described. As well as the anti-value (is that a thing?) I see in the state of the mutable web as it stands now. > We need to stop pitching decentralization to the fantasies of developers, and start giving value to makers, shippers, and users. Saying decentralization is a fantasy seems in spite of some of the most successful technologies in human history. This is odd to me. edit: as well as new federated software embracing decentralization. You being against this is perplexing to me.

marknadal 7y ago
Granted, there are certainly worse implementations of immutability than others, but immutability as the foundational architecture (even in its best design) creates bottlenecks. Yes, a person owning their data should mean they can change it! Fair point on UI/UX, you got me here - you are right. It is trivial to build immutability on mutability, this is what MD5/SHA checksums do all the time on things like downloads. But if your base system is immutability, it is hard to do anything else well. To your question: I agree signing is important. Doesn't mean it needs to be immutable! Very well said! I'm impressed with how both precise and concise that was. Good point. Oh sorry, decentralization is great, just its target use case should be "developer fantasy" but "user value add". Sorry for the confusion here. I'm pro decentralization (I build p2p protocols).
- lijogdfljk 7y ago
  > Yes, a person owning their data should mean they can change it! Well being able to change data is of course needed; so that goes without saying (I should hope). However what I mean by immutability is rather that I think the web should allow me to link to a piece of content, and ensure that what I link cannot be changed from underneath me if I choose not to. I feel like I'm describing features and you're describing implementation. To me, immutability is a feature. I'm not at all referring to blockchain or any of that junk - merely that I want to be able to link (in the web sense, url/etc) to content - to discuss it with other people and etc, and have that content not change. Optionally of course I should be able to link to mutable content. However we have mutability down. The feature add I'm talking about, is immutability. > It is trivial to build immutability on mutability, this is what MD5/SHA checksums do all the time on things like downloads. But if your base system is immutability, it is hard to do anything else well. Sure, but that's an implementation detail. I was not (or did not intent to, at least) say that the "base implementation" must be immutable. All I care about is that immutability is a first class citizen, and linking immutable content is possible. To build user features around immutability, where as nothing like that exists in the current "web".
  - marknadal 7y ago

pfraze 7y ago

Discussion around the technical side isn't developer solipsism. It's part of the process for product design. Append-only logs do scale. Dat 2.0 uses a hash-trie structure inside its file metadata log to act as pointers for fast lookups. Partial replication means you don't have to pull down the full history. Mark, you work on a competing product to ours. I'd appreciate it if you didn't FUD our work. We don't do that to you.

marknadal 7y ago
DAT is one of the better protocols because it has evolved its approach and is starting to adopt CRDTs. I'm FUDing append-only logs and immutability, not products - don't conflate the two, or else you discourage legitimate discussion around the scalability of different architectures. Question: Do you have to still store the full history? How would you explain scaling something like Uber with this model, then? You're gonna have 100s of GPS coordinate updates per second, for each car, potentially millions of them.