4 ms·
Decentralized systems suck. Look at any protocol or network service designed to be fully decentralized. Inefficient routing, unstable communication, high laten
by peterwwillis 13y ago
Decentralized systems suck.
Look at any protocol or network service designed to be fully decentralized. Inefficient routing, unstable communication, high latency and low bandwidth. You have to do a lot more work to achieve things that would otherwise be simple and straightforward.
Take human communications, for example. They're largely decentralized by default, but we use centralized services to improve efficiency. If some piece of information needs to be disseminated to a large population over a period of time, we first record the message, then we post it in a centralized location viewable by lots of clients[1]. Those clients can then go back and distribute the information to its connected peers[2], but technology also allows us to distribute subscriptions of this information to a wider population in a variety of media[3]. ([1]stone tablets, [2]town crier, [3]newspapers/television)
Modern people are afraid of centralization in the Internet because they are suddenly realizing they have no real control over the content or medium. But this is the case with all other forms of centralized mass communication, too. People are also afraid of their lack of control over their data, which is tantamount in these days to a form of property. Anyone would freak out if the contents of their house were suddenly only available at the discretion of some conglomerate!
But you don't need "decentralization" to keep a backup of your data. Most of it originates from you anyway, and you can store it before you publish it. Some service providers even provide methods to download all your data, though obviously you can't rely on everyone providing you that option (unless someone passes a law...)
On top of all this, the internet is already a system of decentralized networks and services. You can just hook in your own services into the network at any point therein and maintain them yourself. The only restriction is really by the ISP, and there's lots of those to choose from in most developed countries. Personally I think we should consider the internet a combination of utility and public highway, since at this point we all need access, but we also need competition due to the sensitivity of limits on network access.
- rthomas6 13y agoWhat about things like Bitcoin and Bittorrent? Those are decentralized protocols that both work great. I think the real problem is that decentralized protocols have no way of determining what information each user should see or not see. I also think this is a solvable problem, but that it is hard.
- peterwwillis 13y agoBittorrent does not work great, to my mind, compared to a centralized service. It's absolutely abysmal on network traffic, is inherently slow, and depending on the communication, practically unusable outside affinity groups. Toward that last end it's useful, but i'd hardly consider it great in general. I don't know enough about Bitcoin to say if it's efficient but i'm aware that transactions can move around the network without much effort, which is nice. But there's big downsides to the transparency of transactions there vs cash or credit, not to mention having your "wallet" stolen vs more traditional credit systems that are more durable for general global commerce. Both of those things are useful. They're just not as efficient or reliable as a centralized service can be. Centralized services & communication have their own problems, but generally speaking the bigger they are the more robust they are, which is often the opposite in truly distributed systems.
- oib 13y agoWhy is bittorrent abysmal on network traffic and inherently slow ? Both facebook and twitter use it for server deployment because they utilize network resources far better than centralized systems. I would make the opposite claim: distributed systems can be far more efficient or reliable than centralized service.
- peterwwillis 13y agoThe short answer is that the effective use of the protocol is almost entirely based on the quality of the network of peers involved in distributing the content. For a relatively small internal network that can be quality-controlled by a single team, this may work more effectively than a single centralized service. For the Internet, this is retarded. See PDFs below for some example reasons. The long answer would have me explaining all of the different use cases of both decentralized and centralized services and a comparison for each. For example, compare bulk file transfer to a domain name service in truly decentralized networks. Keeping domain names requires massive coordination of peers over a gigantic index which can't have conflicts. Bulk file transfer is almost the opposite, as few people have the records (torrent) and the task is mainly in distributing the index (the file). Both are simpler, more efficient, and more reliable if you take out the middle-man of trying to coordinate with more than one peer. http://www.imd.uni-rostock.de/veroeff/ImprovedBitTorrent_Danielis.pdf http://www.imd.uni-rostock.de/veroeff/ImprovedBitTorrent_Dan... https://engineering.purdue.edu/~isl/TON12.pdf https://engineering.purdue.edu/~isl/TON12.pdf https://www.ietf.org/proceedings/74/slides/P2PRG-0.pdf https://www.ietf.org/proceedings/74/slides/P2PRG-0.pdf
- amirmc 13y ago> "... the internet is already a system of decentralized networks and services. You can just hook in your own services into the network at any point therein and maintain them yourself." Sure, but I'd argue we should be making it easier for developers to build more decentralised systems. Very few people seem to be working on tools and OSS products at that level and that's because the problems are challenging (of the kind that a startup wouldn't/shouldn't attempt - hence an active area of work in university [1]). There are plenty of things that require centralisation but I'd argue that just as many don't need to be centralised but just end up that way by default. [1] http://nymote.org http://nymote.org (I'm working on these tools with others)
- peterwwillis 13y ago> I'd argue we should be making it easier for developers to build more decentralised systems Why?
- amirmc 13y agoOk, let's start with your points. "Inefficient routing, unstable communication, high latency and low bandwidth": If I have two devices (phone and laptop) on the same wifi network (e.g. home or work) and I want to send something between them, why should it first go via a cloud service? In my case, photo backups from my phone to my laptop, which happens via PhotoStream at the moment, or files from my laptop to my phone. If a route can easily be formed across the nearest AP then it's shorter, more stable, probably higher bandwidth and certainly lower latency than via the cloud. As the number of devices I own grows, this use case will become more prevalent (think "Internet of Things"). "You have to do a lot more work to achieve things that would otherwise be simple and straightforward": Sure, but you haven't asked why this is the case (and who 'you' is), which is just as important to examine. At the moment, you as a developer have to do more work in order to build decentralised systems because not enough free OSS tools exist yet. The main reason that using a central server is easier is because the tools, techniques and knowledgebase have become 'standardised' but that doesn't preclude there being better ways of doing things. I'm advocating tools that make it easier to build decentralised services and products. Decentralised systems are more resilient than their centralised counterparts. Any large-scale service/product ends up being decentralised behind the scenes and if you don't believe me, ponder for a moment how the likes of Google, Facebook et. al., manage to deal with so many users across the globe. They have to make their own tools to solve the same kinds of problems I'm alluding to. I'm merely advocating that we try to 'productize' such tools so that any developer can pick them up and incorporate them from the very beginning. Real world examples of such things are how Skype worked in the early days, Dropbox's LAN sync [1] and even Spotify [2], where P2P can improve user experience and reduce their bandwidth costs. I don't know how their systems work but this clearly demonstrates the business value in P2P. As you say yourself, "the internet is already a system of decentralized networks and services" but don't dismiss the work taken to get there and the value that the lack of centralisation brings. Those same benefits can be gained elsewhere provided the appropriate tools/infrastructure are made available and useable. Hypothesis: If there were FOSS tools that make it easier to create secure, decentralised systems, more developers would create secure, decentralised products. [1] https://www.dropbox.com/help/137/en https://www.dropbox.com/help/137/en [2] http://community.spotify.com/t5/Help-Desktop-Linux-Mac-and/Unadvertised-P2P-feature/m-p/400160#M46512 http://community.spotify.com/t5/Help-Desktop-Linux-Mac-and/U...
- w0rd-driven 13y agoI think you hit on why Diaspora hasn't really taken off. There may be technical challenges I gloss over but centralization denotes authority to me. Most design by committee OSS seems to suck really hard but having that one strong sense of leadership, like in the case of Linus seems to make a huge difference in comparison. I know I'm mostly talking about humans but I think there's part of our condition that doesn't really promote decentralization in a pure state. I think we could and should try to promote it to make future endeavors more successful but it may be too strong of a current to swim against if I can use any more analogies...
- amirmc 13y agoI think Diaspora was an idea ahead of it's time. Market-timing is critical for a new project/startup and I think there weren't enough users in the world who understood both the costs and benefits of Facebook (or social media) to properly evaluate and consider alternatives. I think the world is different now and there is a growing group of people who care and would consider alternatives and others who 'have nothing to hide' and will continue to hand over personal data. Edit: An example of my view above is the press attention that things like ArkOS, CozyCloud and others are now getting. It remains to be seen how many of these build large, sustainable user bases but it does indicate some level of 'market pull', which (imho) we should try and match with 'technology push'.
- wmf 13y agoIt sounds like you're talking about two different things. The Diaspora software could have strong central leadership (with an open source license providing a right to exit that may never have to be exercised) while the deployment could be distributed.
- DenisM 13y agoWell how about central planning vs free market? Logically, the former must win as more efficient per your own argument, and yet in the long run free market wins.
- wmf 13y agoI can't find the link, but I saw an estimate that even today's supercomputers aren't powerful enough to centrally plan a modern economy — there are literally millions of variables.