Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
On Mon, Feb 2, 2009 at 8:10 AM, Bernie Innocenti ber...@laptop.org wrote: IEEE chose to make wi-fi networks look like 802.11 LANs, similar to ethernet. It might have been a bad idea in retrospect, but now we have to live with it. AFAIK, the bulk of the problem with multicasts over 802.11s (and not all wi-fi networks) is that those must be propagated at the slowest possible link speed in order to reach all nodes. This is irrelevant, really. Protocols are designed with certain assumptions. Those assumptions (mostly having to do with the behavior and cost of broadcasts) were true when the protocols were designed, and are no longer true today. This is the way of all software, it's not unique to 802.11s in some way. Like Martin, you are confusing mDNS with DNS-SD. Ok, but how would the laptops advertise their SRV records without multicast DNS? Wait, are you perhaps suggesting to use DDNS to publish those services on a nameserver running on the XS? That is how DNS-SD works, yes. --scott -- ( http://cscott.net/ ) ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
On Mon, Feb 2, 2009 at 8:58 AM, Bernie Innocenti ber...@laptop.org wrote: Morgan Collett wrote: Also don't blame avahi for the fact that we send out updates every time you alt-tab between shared activities, so that your icon can jump to the appropriate snowflake on everyone else's Neighborhood Views... as well as sending who joined and left... Mature GUIs have a common pattern to avoid too much graphical flickering on possibly rapid state transitions, such as setting a busy pointer or disabling buttons while some operation is in progress. I don't know if it has a name, but the algorithm is exactly the same for de-bouncing mechanical keys: you propagate the event only after the state has settled for a certain amount of time. This would take away a certain percentage of spurious updates, but the number basically remains proportional to the number of users so it doesn't scale much better. http://wiki.laptop.org/go/Network_architecture#Direct_presence_interrogation describes an algorithm to ensure that the amount of traffic in the net is kept proportional to the number of users. (The algorithm you describe is actually proportional to the number of users squared or cubed, because the size of the messages as well as the number of such messages increases with the # of users. If a mesh network is involved, the number of rebroadcasts necessary is another factor roughly proportional to the size of the network.) --scott -- ( http://cscott.net/ ) ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
C. Scott Ananian wrote: This is irrelevant, really. Protocols are designed with certain assumptions. Those assumptions (mostly having to do with the behavior and cost of broadcasts) were true when the protocols were designed, and are no longer true today. This is the way of all software, it's not unique to 802.11s in some way. You make it look like there was an alternative to broadcasts in a peer to peer network, but I don't see any way out unless you want to have master browsers with elections in the best Windows workgroup tradition. Anyway, stuff that doesn't exist yet. Wait, are you perhaps suggesting to use DDNS to publish those services on a nameserver running on the XS? That is how DNS-SD works, yes. I do not understand the security side of it, and how old records get garbage collected unless you do a periodic refresh. -- // Bernie Innocenti - http://www.codewiz.org/ \X/ Sugar Labs - http://www.sugarlabs.org/ ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
On Mon, Feb 2, 2009 at 3:13 PM, Bernie Innocenti ber...@laptop.org wrote: C. Scott Ananian wrote: This is irrelevant, really. Protocols are designed with certain assumptions. Those assumptions (mostly having to do with the behavior and cost of broadcasts) were true when the protocols were designed, and are no longer true today. This is the way of all software, it's not unique to 802.11s in some way. You make it look like there was an alternative to broadcasts in a peer to peer network, but I don't see any way out unless you want to have master browsers with elections in the best Windows workgroup tradition. I think you've left the topic. You also seem to be trying to solve five different problems at once, without acknowledging that the solutions might be different (even if the abstraction is the same). Anyway, stuff that doesn't exist yet. What? WfW was just a nightmare? Whew. Wait, are you perhaps suggesting to use DDNS to publish those services on a nameserver running on the XS? That is how DNS-SD works, yes. I do not understand the security side of it, and how old records get garbage collected unless you do a periodic refresh. Research, young butterfly. I'm not finding this thread very useful, sorry. --scott -- ( http://cscott.net/ ) ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
On Mon, Feb 2, 2009 at 14:18, C. Scott Ananian csc...@laptop.org wrote: On Mon, Feb 2, 2009 at 5:13 AM, Bernie Innocenti ber...@laptop.org wrote: Martin Langhoff wrote: On Mon, Feb 2, 2009 at 6:39 PM, C. Scott Ananian csc...@laptop.org wrote: My suggestions: DNS-SD and libepc (http://live.gnome.org/libepc/). There's no need for Sugar-specific solutions here; we just need to use existing standard solutions. Yep - I want existing standard stuff, but the devil we know seems to swamp the spectrum with 802.11s. When I read the Zeroconf book, I got the impression that the _standard_ was carefully designed to minimize needless broadcasts and scale well in real scenarios. I can't comment on the current Avahi _implementation_ though. This is true for wired networks; not necessarily true for mobile and/or wireless networks. Like Martin, you are confusing mDNS with DNS-SD. --scott Also don't blame avahi for the fact that we send out updates every time you alt-tab between shared activities, so that your icon can jump to the appropriate snowflake on everyone else's Neighborhood Views... as well as sending who joined and left... Regards Morgan ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
Martin Langhoff wrote: On Mon, Feb 2, 2009 at 6:39 PM, C. Scott Ananian csc...@laptop.org wrote: My suggestions: DNS-SD and libepc (http://live.gnome.org/libepc/). There's no need for Sugar-specific solutions here; we just need to use existing standard solutions. Yep - I want existing standard stuff, but the devil we know seems to swamp the spectrum with 802.11s. When I read the Zeroconf book, I got the impression that the _standard_ was carefully designed to minimize needless broadcasts and scale well in real scenarios. I can't comment on the current Avahi _implementation_ though. Even if the standard itself is flawed, designing a custom protocol to do the same thing is going to be a lot of work and probably end up facing the very same design issues that made the IEFT's standard inadequate for us in the first place. When it comes to non-trivial networking protocols, I don't trust any given individual to be able to do a good job without going through an *extensive* iterative design process with public reviews of interim drafts. What's hardest about networking is that it looks deceptively easy at first :-) -- // Bernie Innocenti - http://www.codewiz.org/ \X/ Sugar Labs - http://www.sugarlabs.org/ ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
Morgan Collett wrote: Also don't blame avahi for the fact that we send out updates every time you alt-tab between shared activities, so that your icon can jump to the appropriate snowflake on everyone else's Neighborhood Views... as well as sending who joined and left... Mature GUIs have a common pattern to avoid too much graphical flickering on possibly rapid state transitions, such as setting a busy pointer or disabling buttons while some operation is in progress. I don't know if it has a name, but the algorithm is exactly the same for de-bouncing mechanical keys: you propagate the event only after the state has settled for a certain amount of time. This would take away a certain percentage of spurious updates, but the number basically remains proportional to the number of users so it doesn't scale much better. -- // Bernie Innocenti - http://www.codewiz.org/ \X/ Sugar Labs - http://www.sugarlabs.org/ ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
C. Scott Ananian wrote: When I read the Zeroconf book, I got the impression that the _standard_ was carefully designed to minimize needless broadcasts and scale well in real scenarios. I can't comment on the current Avahi _implementation_ though. This is true for wired networks; not necessarily true for mobile and/or wireless networks. IEEE chose to make wi-fi networks look like 802.11 LANs, similar to ethernet. It might have been a bad idea in retrospect, but now we have to live with it. AFAIK, the bulk of the problem with multicasts over 802.11s (and not all wi-fi networks) is that those must be propagated at the slowest possible link speed in order to reach all nodes. Like Martin, you are confusing mDNS with DNS-SD. Ok, but how would the laptops advertise their SRV records without multicast DNS? Wait, are you perhaps suggesting to use DDNS to publish those services on a nameserver running on the XS? -- // Bernie Innocenti - http://www.codewiz.org/ \X/ Sugar Labs - http://www.sugarlabs.org/ ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
Re: [Sugar-devel] Service announcement scheme - (Re: A small request.)
On Mon, Feb 2, 2009 at 5:13 AM, Bernie Innocenti ber...@laptop.org wrote: Martin Langhoff wrote: On Mon, Feb 2, 2009 at 6:39 PM, C. Scott Ananian csc...@laptop.org wrote: My suggestions: DNS-SD and libepc (http://live.gnome.org/libepc/). There's no need for Sugar-specific solutions here; we just need to use existing standard solutions. Yep - I want existing standard stuff, but the devil we know seems to swamp the spectrum with 802.11s. When I read the Zeroconf book, I got the impression that the _standard_ was carefully designed to minimize needless broadcasts and scale well in real scenarios. I can't comment on the current Avahi _implementation_ though. This is true for wired networks; not necessarily true for mobile and/or wireless networks. Like Martin, you are confusing mDNS with DNS-SD. --scott -- ( http://cscott.net/ ) ___ Devel mailing list Devel@lists.laptop.org http://lists.laptop.org/listinfo/devel
