Lately, it seems that a lot of problems have been blamed on network congestion. The increase in the number of clients is not a problem in and of itself. The problem is that so many of the clients are not sharing files (or sharing very few).

When many of the clients don't share files, it becomes much harder for a search to hit. For instance, if I send a search out in which few clients are sharing files, my TTL may die before the search finds a match. Each client sharing zero files needlessly reduces my TTL.


My suggestion is this....

1) By default clients should NOT share files. (for obvious security and configuration issues)

2) If a client does not share files, no other client will connect to it. It should not be stored in any host catchers. In other words, if the client does not share files, it should not recieve (or accept) incomming connections. (outgoing are still fine, of course)

3) If a client does not share files, no searches will be routed to it. (there's nothing to search there anyway. The drop in TTL can better be used somewhere else)


These rules basically do one thing: it ensures that clients not sharing files are dead ends on the gnutella net. The only traffic they recive are their own search results.

This concentrates search queries to clients that are actually sharing something, drastically increasing the chance for search hits for everyone. Plus, it would reduce the gnutella net traffic some without harming the gnutella net.

I'm not certain if this would require a change in the protocol. I don't think so. Clients not sharing files could simply never respond to a PING with a PONG. Thus its IP never gets published on the network. Also, should a client's IP become published for some reason, it could simply refuse all incomming connectons.

Anyway, what do you think? I don't mind critisism as long as it is reasonable and not simply flame.

This won't work, i think it could possibly damage the network's infastructure. This could also make logging onto gNet more difficult. And I bet it won't make a big difference in fixing network congestion.

im not really an expert but i don't think its such a good idea, but hell i could be wrong...

I agree with Mr - I think such action would break the network. Let's say you have 3000 hosts online, with only 10% sharing any files (and perhaps even that's being liberal). If search requests weren't passed on to clients without any shared files, that means that the entire network's search routing would be forced upon only 300 clients.

The added search requests (essentially 10x as many) would kill the bandwidth of those people who were sharing... People whose bandwidth was already being used to serve files. A week later, nobody would be sharing files because the "price" to do so would be too high.

I wish I had some suggestions on how to fix the congestion, but I don't. Meanwhile I'll keep playing with redir and ipchains, hoping to find a local solution to the firewall problem...

Shaun

Well, yes, it _could_ increase the number of seach requests to the remaining clients that are sharing files. BUT... I don't think it would do so significantly, or even at all.

Think about it this way... In theory, every node will see every search once on a good, healthy network with todays system. If this is true, my suggestion would not actually increase the traffic through the sharing nodes at all. Granted this is a best case scenario and likely is not actually true in the real world. But I definately don't think traffic through those nodes would increase 10x.

Lets say that today, a search may propogate through only 1/3 of the network. With my suggestion, traffic MAY increase 3x in this case through the "sharing" nodes while reducing traffic everywhere else.

The benefits I see from this sort of strategy are that small TTLs will more easily propogate the "sharing" network. Searching will be much faster and much more thourough while decereasing overall network activity.

It would not break the network, but it would make it more efficient (granted, at some expense of the sharing nodes).

Oh, yeah, and while I am thinking about it, another change that would have to be made for this to work is that "sharing" clients would NEED to accept LOTS of incomming connections from "non-sharing" clients, no number limits or at least a very large limit (30-50). Else, if only the sharing nodes could accpet incomming connections, the network would quickly run out of open spaces a non-sharing client could connect to. This should not be a problem as the "non-sharing" clients should only be sending out an occasional search querry and getting some responses back. Unless he is a spammer, in which case it would be easier to block him.

Is anyone working on some sort of gnutella network simulator program? A network simulator would go a long way to answering some of these and other questions. It would give us a better understanding of how the viral population network model of gnutella really operates and how to improve its efficiency.

I admit that I may not know what I am taling about. (I am a mechanical engineer by proffession. We always made fun of Comp Sci and Comp Eng people).

I figured this was sort of relevant to the thread - if you haven't read it already, check out the study "Freeriding on Gnutella." It's an interesting read. You can find it at http://www.parc.xerox.com/istl/group...eeRidingA.html [image]http://noc7.levitate.org/shaun/test.gif[/image]

Shaun

D'oh, sorry about the malformed image tag in the previous post. Just testing something and I got the coding wrong

Here is another interesting read. It basically attempts to do the same sort of thing I sugested above, but gives preference to high speed connections instead of big share'ers.

http://coralia.ctw.cc/gnutella/proposal.htm

I really like this proposal, and my ideal of giving some preference to sharing clients could also be incorporated.

Well, I share about 5 Gig of files (only complete and useful stuff - this is also very important!), but it's not a big deal to do so if you're on a LAN connection and uploads running with a total speed of ~250K don't affect your own download speed. I can understand people with slower conections who don't want to share - I would do it neither if it affected the speed of my own downloads! I don't have too much of a problem with the fact that I'm playing the role of a fast server for others as long as I can also get some good stuff (btw. right now after 35h uptime my netstatistics says "bytes out:9.752.491.172", "bytes in:1.950.241.450" ...).

if people are not willing to share their things, then you should not share your things with them!! did we not learn this in kindergarten? just my own personal opinion, i do not care much for greedy people(like lars ulrich or dr. dre) they want to blame the fact that they are not selling records on us instead of accepting the reality that they are has-beens. after all, wasn't there a record number of cd sales last year?

Limewire offers an option on connections with "Freeloaders". I think Bearshare also can block freeloaders.


Another thought. If a connection is "high speed" even if not sharing wouldn't it be better used "deeper' than My Modem connection with lots of shared files. I can only give you connection to two servants where the high speed could give you many. Limewire sets me up for two outbounds and 0 inbound's, this puts modems on the fringe of the network.

[This message has been edited by lightstone (edited 03-03-2001).]