Developers of gtk-gnutella, please allow us to search again after 180s. I get hardly any results with 600s between search prods.

I doubt that any host I have ever connected to can even reach me after 600s they go by so fast. I don't know if that is broken behaviour, but regardless it helps if I search again after 180s.

This is on a modem - it seems like you have set up a lot of the stuff for broadband by default. Perhaps you could have two 'default' configurations that can be chosen at the click of a button - modem and broadband.

It's really hard to get any connections to stay up, at least on modem, they drop like flies. I find that to keep around 4 up I need to set a target minimum of around 8 to 12, and maximum higher than the minimum.

180s between searches <-- no way that is flooding the network. I used to use gnut, which could send out searches once per second, I think I remember. Now that may lead to network degredation, but really 180s is quite conservative unless you are running like <UL>hundreds</UL> of searches in parallel.

Nos

Researching every 180 seconds is way too much. The developers of limewire have done some tests, and concluded that researching less than every 45 minutes is unhealthy for the gnutella network.

Hein

And keeping 4 connections up with a modem is too much. Every connection takes 1-2 KByte/sec in, and approx. 1 KByte/sec out. A modem can only take 56 Kbit/sec=7Kbyte/sec in the most optimal case.
I think you'd better keep 3 connections max. In gtk-gnutella 0.91 you can switch 'prefer compressed nodes' on in the advanced preferences. For you as a modem user this is good.

I think the next version of gtk-gnutella will support ultrapeers, which will drop bandwith requirements for modem users significantly.

Hein

Hein,

Be f*cked!~

I haven't seen a search result in days, not since upgrading to 0.91.

I had recompiled 0.90 with the limit hacked down, but this version is more complex to hack, and so far I haven't been bothered.

You know what to do with your theoretical figures. Get a modem and try it, d!psh!t.

Bad for the network .. do I care what is bad for a network that is too broken to provide search results to modem users? No. If the network can't handle 10 queries being sent out every 3 minutes, then limewire should be figuring out why it's so broken.

Maybe on your broadband connection does every connection take 1-2k/s - personally I think something's broken if that is the case, anyway.

For me, gtk-gnutella generally reports less than 1 1/2 k per second, mostly 0-1. What I tried to say is that the connections are dropping off at such a rate that I have to be constantly trying to find new ones at a ratio of about 2-3:1 .. ie for every connection I am trying to keep up, I have to be looking for 2 or 3 to replace it, since many of them do not connect .. either timeout or some other refusal.

And before you tell me what is wrong with my setup, no, this is not BECAUSE I am looking for too many connections. If I set it to 1 I get about ... zero. Nix .. nothing ... SFA. If I set it to 3 or 4 as you suggest, still around 0 to 1 connections at any given moment.

Ultrapeers .. hopefully may help this for gtk-gnutella users .. in general I think the network was a lot healthier for gtk-gnutella users before there were any ultrapeers at all. Now many of the hosts I try to connect to tell me to f*ck off.

Please do not try to tell me how my PC is working .. next you'll be telling me when I'm running out of hard drive space. Thanks, but .. do not assume I am stupid and you know better than me how my connection is faring.

Nos

Well, well! You're easily offended....?? Why all the swearing?

Maybe you didn't noticed, but I tried to help you.

Every time a new connection is made with a new gnutella-node, all the searches you have are sent to those new nodes with 10 seconds between every search. So you don't have to adjust the re-search time to search a new node.

And yes, gnutella in its current implementation is broken because it isn't healthy to research every 3 minutes.

Hein

well .. why not ..

Quote:

Maybe you didn't noticed, but I tried to help you.

Hmm mm .. often in these forums 'help' is disguised as authoritarianism .. or is it the other way around, anyway ...

Quote:

Every time a new connection is made with a new gnutella-node, all the searches you have are sent to those new nodes with 10 seconds between every search. So you don't have to adjust the re-search time to search a new node.

Thanks for this info. If it is true, then there really is something wrong with gtk-gnutealla 0.91 afaics.

I genuinely have seen my a) ability to maintain stable connections and b) search results diminish drastically with each upgrade since about 0.89. Downloads usually never finish either, but this is probably unrelated and more to do with the files I have been trying to get and the people sharing them.

Quote:

And yes, gnutella in its current implementation is broken because it isn't healthy to research every 3 minutes. Hein

Personally I am soo skeptical about your information above, for one reason: if your description is accurate, then continually breaking connections and making new ones should INCREASE my search results. But as I have said, it most certainly has not.

Do you have any references for either the theoretical models, computer models, or practical demonstrations of the theory that searching every 3 minutes is 'harmful' to the gnutella network? How is the 'harm' measured, and what outcomes were given priority? For which types of gnutella connections was it considered 'harmful', and were there any types of connections where it is 'not harmful' or 'less harmful'?

Nos

It is harmful if all the Gnutella nodes do the same. Recall that you're normally seeing about 16384 hosts in your horizon.

If everyone requeries once every 180 seconds for just one query, it will make the query traffic be 91 queries/seconds.

This is about the average traffic one gets today.

Gtk-gnutella is already nice to allow retries every 600 seconds. Other Gnutella clients authorize only 1 requery every 45 minutes!