3 ms·

How so? There only issue I can think of is upload speed, which is improving for many, and would improve more prevalently if more average users had a clear need

How so? There only issue I can think of is upload speed, which is improving for many, and would improve more prevalently if more average users had a clear need.

schoen 9y ago
The original BitTorrent wasn't aware of network topology at all, so it could randomly use a relatively expensive or congested link just as easily as a LAN connection. You could get better use of network resources overall if you usually preferred to get pieces from peers that are closer to you on the network. I think the BitTorrent company and/or community were studying this issue at some point and so they may have come up with alternative algorithms that prefer at least AS-locality or something. As an extreme case, imagine that there are two seeds, one being another computer on your LAN and another being a host in Timor-Leste (which at one point in the 2000s reportedly had a single T1 connection for the entire country). It will be nicer for you and for the Timorese if you can somehow arrange to simply download the file from your LAN seed. Here's one example of someone trying to improve this property: https://en.wikipedia.org/wiki/TopBT https://en.wikipedia.org/wiki/TopBT
- thomastjeffery 9y ago
  I would argue that in the long run, it is better to demand more bandwidth than is available, so that infrastructure builders can react accordingly.
  - schoen 9y ago
    I like the overprovisioning concept fine, but it's going to be pretty expensive to overprovision for the possibility of high-speed connections between all random pairs of hosts everywhere in the world regardless of geography. We'll probably have to persuade a lot more people to value this degree of connectivity ideologically, because demand for it isn't really getting expressed much right now. I don't want to sound like a Bellhead here, but the packet switching infrastructure that's been built out has embodied some assumptions about its likely patterns of use. To change those patterns of use a lot (which I'd like to see too) probably needs more infrastructure. Even things like ADSL are a problem here. I agree that path asymmetry doesn't feel at all healthy for the Internet -- it feels like it's baking in producer and consumer roles -- but there are a few counterarguments that may give us some pause: * It's never been cheaper for people who specifically want to provide services to others to rent servers in colocation facilities. It's getting cheaper every year. So, maybe people who care to do so could just rent some servers. (Of course, this might give them less autonomy because there are more people who can decide to kick them off or shut them down, and they might still not have the infrastructure to distribute media that they produce at home to the servers they rent. On the other hand, running things at home may make anonymous publishing harder because the associated broadband service is usually tied to a physical address by the ISP.) * There's never been a huge amount of expressed demand for symmetric network connections, even when they've been available as an option. (Maybe this is different in different countries?) The economic division of network operators into "eyeball networks" (major net traffic sinks) and "provider networks" (major net traffic sources) seems to be getting sharper over time. The network operators themselves might have an economic incentive to try to get their users to equalize the traffic in both directions -- especially in terms of arrangements around settlement-free peering -- but they haven't managed to change most users' habits very much this way. * I think of myself as a very pro-decentralization Internet user, yet whenever I run "ifconfig wlan0 | grep bytes", my download volumes consistently dramatically outpace my upload volumes, even though I have a symmetric broadband connection both at home and at work! So apparently I'm not being the change I want to see in the world.
    - thomastjeffery 9y ago
      Your first point is actually a boon for P2P. If it is cheap enough for users to rent bandwidth closer to each other, that will solve the same problem CDNs solve. As for points two and three, decentralization does not mean 100% symmetry, just relatively more. I have a symmetric connection, and can forward ports without any fuss, and rather enjoy those advantages. Even so, I download quite a lot more than upload. I don't expect that to change much, even with significantly more P2P traffic. There are a few things where I am really frustrated with other's inability to forward ports, especially video games and VOIP.
always_good 9y ago
Well, doesn't your answer of "it's improving for many" admit to the reality that it sucks right now, answering your own question? Besides, even if we have more bandwidth, it doesn't change the fact that p2p would require more of it. Also, if you asked the average person if they want Spotify to be peer-to-peer + use more bandwidth vs centralized + use less bandwidth, do you think they are in a position to care much beyond the bandwidth difference? Especially on mobile devices? I certainly would choose the latter. What do I care about making Spotify's operations cheaper at my own expense?
- thomastjeffery 9y ago
  > it doesn't change the fact that p2p would require more of it. But P2P doesn't require more bandwidth. It requires more upload bandwidth, which is arbitrarily constrained. > if you asked the average person if they want Spotify to be peer-to-peer + use more bandwidth vs centralized + use less bandwidth, do you think they are in a position to care much beyond the bandwidth difference? Especially on mobile devices? Of course not. Spotify is already tailored to the current state of internet infrastructure, which heavily favors centralized content delivery. Spotify can afford to foot the bill for their content delivery network. That is why it is no longer "better" for them to use peer to peer networking. > What do I care about making Spotify's operations cheaper at my own expense? For Spotify or Netflix, or any other site that already has a centralized CDN, there is no advantage. For everything else, however, P2P is a great way to scale distribution, and the more we do to incentivize ISPs to increase upload speed, open ports, etc. the better.
cortesoft 9y ago
Well, ISPs did not build their systems with equal upload and download in mind. There is no pure technical reason they couldn't have, but it was what they did. To replace all of that would be difficult, and the large players in the existing system will have economic motivations for fighting any change. There is a pretty good write up here: http://www.hyperorg.com/blogger/2009/10/03/why-up-is-slower-than-down/ http://www.hyperorg.com/blogger/2009/10/03/why-up-is-slower-...
- thomastjeffery 9y ago
  > To replace all of that would be difficult Not particularly. Symmetry is totally arbitrary. It's having the same amount down as before, while having more up that is the problem. > and the large players in the existing system -will- have economic motivations for fighting any change. That's the real problem. The solution is to give them an economic incentive to change. Average users wanting more bandwidth/symmetry is an effective way to do that.
  - toast0 9y ago
    Symmetry isn't totally arbitrary. In many deployed access technologies, scaling head end send speed is easier than scaling user end send speed; sometimes there are settings to increase upload speed at the expense of download speed (ex annex M for adsl2), but sometimes it's an inherently asymmetric system.
    - thomastjeffery 9y ago
      Even so, it would be a good thing for ISPs to change that infrastructure. That won't happen until there is a fiscal incentive for them to do so, so let's not be afraid to give them one.
kodablah 9y ago
It usually requires more copies and more uploaders due to the fact that personal devices' uptimes are much more ephemeral.

How so? There only issue I can think of is upload speed, which is improving for many, and would improve more prevalently if more average users had a clear need.

schoen 9y ago

The original BitTorrent wasn't aware of network topology at all, so it could randomly use a relatively expensive or congested link just as easily as a LAN connection. You could get better use of network resources overall if you usually preferred to get pieces from peers that are closer to you on the network. I think the BitTorrent company and/or community were studying this issue at some point and so they may have come up with alternative algorithms that prefer at least AS-locality or something. As an extreme case, imagine that there are two seeds, one being another computer on your LAN and another being a host in Timor-Leste (which at one point in the 2000s reportedly had a single T1 connection for the entire country). It will be nicer for you and for the Timorese if you can somehow arrange to simply download the file from your LAN seed. Here's one example of someone trying to improve this property: https://en.wikipedia.org/wiki/TopBT https://en.wikipedia.org/wiki/TopBT

thomastjeffery 9y ago
I would argue that in the long run, it is better to demand more bandwidth than is available, so that infrastructure builders can react accordingly.
- schoen 9y ago
  I like the overprovisioning concept fine, but it's going to be pretty expensive to overprovision for the possibility of high-speed connections between all random pairs of hosts everywhere in the world regardless of geography. We'll probably have to persuade a lot more people to value this degree of connectivity ideologically, because demand for it isn't really getting expressed much right now. I don't want to sound like a Bellhead here, but the packet switching infrastructure that's been built out has embodied some assumptions about its likely patterns of use. To change those patterns of use a lot (which I'd like to see too) probably needs more infrastructure. Even things like ADSL are a problem here. I agree that path asymmetry doesn't feel at all healthy for the Internet -- it feels like it's baking in producer and consumer roles -- but there are a few counterarguments that may give us some pause: * It's never been cheaper for people who specifically want to provide services to others to rent servers in colocation facilities. It's getting cheaper every year. So, maybe people who care to do so could just rent some servers. (Of course, this might give them less autonomy because there are more people who can decide to kick them off or shut them down, and they might still not have the infrastructure to distribute media that they produce at home to the servers they rent. On the other hand, running things at home may make anonymous publishing harder because the associated broadband service is usually tied to a physical address by the ISP.) * There's never been a huge amount of expressed demand for symmetric network connections, even when they've been available as an option. (Maybe this is different in different countries?) The economic division of network operators into "eyeball networks" (major net traffic sinks) and "provider networks" (major net traffic sources) seems to be getting sharper over time. The network operators themselves might have an economic incentive to try to get their users to equalize the traffic in both directions -- especially in terms of arrangements around settlement-free peering -- but they haven't managed to change most users' habits very much this way. * I think of myself as a very pro-decentralization Internet user, yet whenever I run "ifconfig wlan0 | grep bytes", my download volumes consistently dramatically outpace my upload volumes, even though I have a symmetric broadband connection both at home and at work! So apparently I'm not being the change I want to see in the world.
  - thomastjeffery 9y ago
    Your first point is actually a boon for P2P. If it is cheap enough for users to rent bandwidth closer to each other, that will solve the same problem CDNs solve. As for points two and three, decentralization does not mean 100% symmetry, just relatively more. I have a symmetric connection, and can forward ports without any fuss, and rather enjoy those advantages. Even so, I download quite a lot more than upload. I don't expect that to change much, even with significantly more P2P traffic. There are a few things where I am really frustrated with other's inability to forward ports, especially video games and VOIP.

always_good 9y ago

Well, doesn't your answer of "it's improving for many" admit to the reality that it sucks right now, answering your own question? Besides, even if we have more bandwidth, it doesn't change the fact that p2p would require more of it. Also, if you asked the average person if they want Spotify to be peer-to-peer + use more bandwidth vs centralized + use less bandwidth, do you think they are in a position to care much beyond the bandwidth difference? Especially on mobile devices? I certainly would choose the latter. What do I care about making Spotify's operations cheaper at my own expense?

thomastjeffery 9y ago
> it doesn't change the fact that p2p would require more of it. But P2P doesn't require more bandwidth. It requires more upload bandwidth, which is arbitrarily constrained. > if you asked the average person if they want Spotify to be peer-to-peer + use more bandwidth vs centralized + use less bandwidth, do you think they are in a position to care much beyond the bandwidth difference? Especially on mobile devices? Of course not. Spotify is already tailored to the current state of internet infrastructure, which heavily favors centralized content delivery. Spotify can afford to foot the bill for their content delivery network. That is why it is no longer "better" for them to use peer to peer networking. > What do I care about making Spotify's operations cheaper at my own expense? For Spotify or Netflix, or any other site that already has a centralized CDN, there is no advantage. For everything else, however, P2P is a great way to scale distribution, and the more we do to incentivize ISPs to increase upload speed, open ports, etc. the better.

cortesoft 9y ago

Well, ISPs did not build their systems with equal upload and download in mind. There is no pure technical reason they couldn't have, but it was what they did. To replace all of that would be difficult, and the large players in the existing system will have economic motivations for fighting any change. There is a pretty good write up here: http://www.hyperorg.com/blogger/2009/10/03/why-up-is-slower-than-down/ http://www.hyperorg.com/blogger/2009/10/03/why-up-is-slower-...

thomastjeffery 9y ago
> To replace all of that would be difficult Not particularly. Symmetry is totally arbitrary. It's having the same amount down as before, while having more up that is the problem. > and the large players in the existing system -will- have economic motivations for fighting any change. That's the real problem. The solution is to give them an economic incentive to change. Average users wanting more bandwidth/symmetry is an effective way to do that.
- toast0 9y ago
  Symmetry isn't totally arbitrary. In many deployed access technologies, scaling head end send speed is easier than scaling user end send speed; sometimes there are settings to increase upload speed at the expense of download speed (ex annex M for adsl2), but sometimes it's an inherently asymmetric system.
  - thomastjeffery 9y ago
    Even so, it would be a good thing for ISPs to change that infrastructure. That won't happen until there is a fiscal incentive for them to do so, so let's not be afraid to give them one.

kodablah 9y ago

It usually requires more copies and more uploaders due to the fact that personal devices' uptimes are much more ephemeral.