From Wikipedia, the free encyclopedia

This is the talk page for discussing improvements to the Gnutella2 article.
Please sign and date your posts by typing four tildes (~~~~). Put new text under old text. Click here to start a new topic. New to Wikipedia? Welcome! Ask questions, get answers. This is not a forum for general discussion about the article's subject.	Be polite Assume good faith No personal attacks Be welcoming	Article policies No original research Neutral point of view Verifiability

Contents 1 Flame War 2 Clean up 3 History (also NPOV) 3.1 Why compare Gnutella to Gnutella2 3.2 Gnutella2.com

[edit] Flame War

is there somewhere a link to that G1 vs G2 flame war ? IT would be quite interesting to read, and perhaps good addition to th article.

I'll probably add some of the Flame War, most likely the majority being Vinnie's amusing contributions to it, as cites later on. Currently, I'm still trying to dig up the last cite about the debate of Gnutella2's packet format vs. Gnutella's GGEP, but as I continue this search fruitlessly, I'm beginning to think that it took place either on the BearShare forums, or more likely the old Shareaza forums, both of which have been lost... when I give up (And remove the cite, because it's probably not going to come, and you'll just have to take my word for it as somebody who live through, participated in, and was an amused spectator of the whole spectacle) I'll probably compile a list of the various flames. -FrYGuY

I gave up on the one cite where Gnutella developers claimed that GGEP provided plenty of flexibility for Gnutella 0.6, and believe it was, in fact, on either the old Shareaza forums or the now defunct BearShare forums. I do offer my word that the claim is factual, but if that goes against the "No original research" rule it can be removed, as it's a pretty minor part of the article. Also, I have added some of Vinnie's more... colorful... contributions to the flame war as citations. -FrYGuY 03:09, 7 September 2006 (UTC)

I think the BearShare messages (Vinnie?) should be moved here, to the talk page. They really aren't Encyclopedic and lend little to the article. People maybe curious as to the content, so links here might be acceptable. The posts by Rapheal and Adam seem somewhat reasonable as references. Bpringlemeir 05:28, 11 September 2007 (UTC)

[edit] Clean up

I added a cleanup tag to this article since the majority of it seems to be written POV. Plus it just sounds bad. Needs an overhaul. Drano 09:56, 24 December 2006 (UTC)

I removed the cleanup tag. It's been there for two months. If you provide examples of what seems to be written in POV I'll happily change it, but until then the cleanup tag seems to just be clutter. -FrYGuY 06:19, 20 February 2007 (UTC)

I have re-added the cleanup tag. This article doesn't cite any reliable sources. Single mail messages from a mailing list are not reliable sources. It maybe fine to have one of these in an article. However, citing publish literature and academic studies might lend some credibility to this article. Bpringlemeir 23:40, 10 September 2007 (UTC)

[edit] History (also NPOV)

I suggest removing the historic bits about the so-called flame war involving the lead developer of BearShare. First, it's a fart in the atmosphere adding no useful information. Second, it always was very obvious that Vinnie was making fun of the Shareaza fanboys posting there who could not abide any kind of criticism against the maker of their beloved Raza. By ripping this out of context and just pointing to certain posts, this might very well be lost on the reader. I find it obvious that this collection of links is only provided as slander. Not everyone has the same sense of humour and these posts certainly weren't the highlight of discussions at the GDF. Anyway, keep in mind that Wikipedia articles must be NPOV. By selectively pointing to certain posts, you are clearly in violation of this rule. As G2 is heavily based on discussions that took place at the GDF to which Vinnie Falco contributed like other Gnutella developers, you owe this guy just as much credit as any other discussion participant. At the very least, you could put this childish bickering to a rest. --82.141.60.211 18:59, 15 March 2007 (UTC)

As a participant in that "so-called flame war", I can tell you that Vinnie was not "making fun" of people. He was pissed. He insulted the developer of Gnucleus, John Marshal (Swabby), the lead developer of Limewire, Adam Fisk, and one of Vinnie's own employees (Though that one was somewhat jokingly), along with people defending the decision (myself included). There was no real context... he honestly just started calling people various explitives, and his flame of Adam Fisk seemed to have happened because he was so angry he insulted even those on his own side. Slander? Well, ignoring the fact that slander in transitory use like speech, libel being what you seem to be implying, it's only slander or libel if it's false, defamatory, and believable by a reasonable person. Seeing as it's true, it does not meet those qualifications. As far as being NPOV, since the part of the article is the developer disputes which erupted, and the most notable and memorable part of those disputes was the flame war, and Vinnie's post being what made it a flame war, I believe it is notable. The thread is in those links, however, and you're more than welcome to add other developers flames, but the main problem with that is that it was a pretty one sided "war". Vinnie basically went ballistic, but everybody else, including Adam Fisk and Gregorio, who remained against Gnutella2, and John 'Swabby' Marshall, who remained supportive, all discussed their points with calmheadedness. The closest to a Flame from somebody other than Vinnie that I can think of would be this post... so feel free to add it.

So, in summation: I believe it to be NPOV because it's posting to basically the sole 'flamer', however the correct method of correcting this if you believe it to be otherwise would be adding links to non-Vinnie flamers. I believe it to be useful because I was specifically asked to add a section on the flame-war to the article, so clearly some people found it useful. If you find it to be distasteful, welcome to the club. So did I. So did many people. But that distasteful nature is part of the reason the schism between the two camps of developers was so severe, and thus in an entry on Gnutella2, it is quite notable. If you believe it to be "picking on Vinnie", then I can't do much against that other than to say that he contributed plenty to Gnutella over the years, mostly good, but THAT isn't notable in this article. Sometimes, a single negative contribution overshadows a lengthy beneficial career.

In that vein, I am now readding the second until you have a more persuasive reasoning than "it's mean" or "it's distasteful". Please keep in mind the old truth that the answer to bad speech is not no speech, but more speech. If you feel that the article sheds Vinnie in an unfair light, either add counter points if it fits in with this article, or add to Vinnie's article the many, many contributions he had to BearShare. -FrYGuY 07:24, 26 March 2007 (UTC)

These cites are not really "Encyclopedic". They are more like some sort of grade school tit for tat. Just state facts. I think that neither Gnutella nor G2 developers think that having two networks is good; it should be obvious to most readers that this causes some tension. Due to the lack of real references, I have added the NPOV tag.

For instance, the limited research I did found that the random walk protocol of G2 is not really superior to most search algorithms. It has slightly less bandwidth, but returns far less results. The paper I cited didn't even begin to discuss DHT based methods (which are currently active on the Gnutella network via Mojito). There are many papers and studies that show DHT methods are superior to other search methods. This is at least one factual error in the article that has produced biased statements. Please cite reliable sources, to make this neutral. Bpringlemeir 23:51, 10 September 2007 (UTC)

Uh, what? "It has slightly less bandwidth, but returns far less results."? I think you meant "uses slightly less bandwidth", but even then the metric is questionable, and I would like to see such a study [edit: Just checked out the study you posted. Read it again... it EXPLICITLY excludes Gnutella2 from the "random walk" category.]. For one thing, search results are dependant upon other nodes returning query hits. Perhaps you are refering to the fact that the end nodes (leaves) are responsible for most of the work of their search, rather than the network topology? Second, DHT based methods have both advantages and disadvantages. The primary advantage is that a search for a hash returns almost (I'll get into that "almost" in a second) global results with minimal bandwidth. The disadvantages are that searching for keywords requires kludgy or alternate systems for searching, produce 'stale' results (meaning nodes which are no longer active or which no longer possess that file), and can and do (here's that pesky 'almost') occasionally suffer from network splits, meaning that nodes separate into sub-networks effectively partitioning search results.

On top of that, nowhere in the article does it CLAIM superiority! Point out what's NPOV if you're going to mark it as such. Not having cites (Which, as Gnutella2.com and the specs are apparently dead now, I can't readily provide) does not makes it NPOV, it makes it needing citations. Mark it appropriately. Until then, I'm removing the NPOV. -FrYGuY 00:39, 11 September 2007 (UTC)

[edit #2: The Gnutella network still uses a query flooding approach. QRP, while a valuable addition to the protocol, simply allows the last hop to be avoided if it is not useful. The basic mechanism is still query flooding.] -FrYGuY 00:50, 11 September 2007 (UTC)

Alright, you seem rather perturbed, so I will leave the page. The paper actually excludes Gnutella2 because it is difficult for them to analyze the Hub/leaf topology. However, a paper like this would add a lot of merit to this article. I have replace the cleanup and npov tag with a references tag.

I think you also misunderstand QRP. It may pragmatically be "last hop only" with peers on the current Gnutella network. However, each ultra should mark the query route table with a hop count. It gathers these from patches to the query route table forwarded by other ultra nodes. When a query is received, the ttl and the hop count in the query route table determine whether the query should be forwarded. In theory that is not last hop and at least some clients on the Gnutella network do implement this. Also, the dynamic querying is much like the random walk in Gnutella2. So, I really think that currently the two networks are nearly identical. The article is rather one sided and doesn't take into account any of these facts. I hope you take this in the good spirit that I intend it. Bpringlemeir 02:13, 11 September 2007 (UTC)

The specification for Dynamic Query could be read to verify that the current search algorithms are rather similar between the networks. Dynamic query has been in use since 2004/2005 on the Gnutella network. It is a factual in-accuracy to say that the current Gnutella network uses query flooding, as I have wrote in that wiki. Bpringlemeir 02:25, 11 September 2007 (UTC)

Hop count is marked on the SEARCHES, not the QRT. In Gnutella (0.6, current), assuming all ultrapeers have 'standard' 30 ultrapeer connections, and searches are Max_TTL 4 like the spec says they should be now, searches are sent out to first 1 ultrapeer, then 30 more if not enough results are returned, then 900 (assuming no overlap), then finally 2700 (again, assuming no overlap). The last hop at any given stage are checked against the inter-ultrapeer QRT. While there are some really clever methods on top of it, it is still, at heart, a query flooding search. A Gnutella2 search starts with one of the leaves' hubs. from there, the hub's neighbors QRTs are checked, and forwarded to neighbors. These little islands of about 7 hubs at a time are crawled until sufficient results are returned. So no, the two networks really aren't exactly 'identical'.

My main reasons for being "perturbed" were the NPOV label (That's fixed now. The article probably does need some clean up.), and the fact that you were claiming that the article promoted Gnutella2 as 'superior', something I went out of my way to avoid when I wrote it. I specifically wrote it to include both advantages and disadvantages of the scheme compared to Gnutella. And yes, there are both. If you can find any studies on the various networks themselves, I would be keenly interested. Unfortunately, most of what's out there is simulation, or assumptions, or entirely out of date (There are plenty of studies on Gnutella 0.4, which is entirely useless now that even base Gnutella 0.6 is far more efficient, and modern Gnutella 0.6 is similarly far ahead of base 0.6). Similarly, comparing them to DHTs is at best misleading. While some parts are far more efficient, other parts are significantly lacking for the purpose of file-sharing networks. -FrYGuY 02:41, 11 September 2007 (UTC)

You still mis-understand query routing. See this picture. Assume three ultras in a chain A-B-C. If C sends a table to B, B increments the table values and then sends it to A. A will insert another incremented value (if it already doesn't have a lower one). The value can be used with the TTL to determine whether to forward queries to QRP capable ultras. The ultra also keeps tables sent from its neighbors, to determine which route should be taken.

Also dynamic querying should start off with small TTL values. For rare files it is similar to what you have stated. It will end much quicker for more popular content (possibly reducing TTL, etc). Are you sure all of this is still simple query flooding? Do none of those features give the current Gnutella searches similar properties as stated for G2 searches.

It is more efficient, as continuing a search does not increase the network traffic exponentially, queries are not routed through as many nodes, and it increases the granularity of a search, allowing a client to stop once a pre-defined threshold of results has been obtained more effectively than in Gnutella. It also increases the complexity of the network and the network maintenance required, as well as requiring safeguards to prevent a malicious attacker from using the network for Denial-of-service attacks.

I didn't mean the two networks are literally the same. I said that in this context "the current search algorithms are rather similar between the networks". Sorry, I split the text and wasn't so clear. Bpringlemeir 04:38, 11 September 2007 (UTC)

Huh, man I've been out of the loop. I was unaware that Gnutella had QRTs over more than one hop now. Anyway, yes, it's still query flooding. Reread what I wrote. First 1. Then, if not enough results, 30. Then 900. As in, an iterative extension if results are not enough. It's still query flooding, albeit a very clever, very flow-controlled version. The basic premise of search propigation remains the same. -FrYGuY 05:40, 11 September 2007 (UTC)

Well, actually that is still not correct. First 1, then up to 30 and then up to 900. However, this will be highly limited by QRT. A typical QRT fill is 10%, so it would be 1, 3, and 9 if the TTL is the maximum. Bpringlemeir 05:52, 11 September 2007 (UTC)

Sadly, it doesn't work like that in practice. Generally, the same bits in the routing table are filled, as much of the content on the network is similar, meaning that most of the searches will get flow-controlled into oblivion, or will mostly get sent (and hopefully retired quickly). However it's entirely possible for searches to constantly hit, and yet garner no search results. -FrYGuY 17:43, 11 September 2007 (UTC)

Do you have references? Bpringlemeir 19:53, 11 September 2007 (UTC)

Ugh, it was gone over in the GDF when somebody was talking about increasing the size of the QRTs to help sparse it out, lemme see if I can find it. -FrYGuY 20:54, 11 September 2007 (UTC)

After a very brief search, the closest I came is this[1] post by Oso. Basically it explains why arbitrarily increasing the sparseness of the table will offer basically no help in filtering searches to forward... -FrYGuY 21:04, 11 September 2007 (UTC)

That particular message basically says that there is no point in reducing the maximum QRT fill to less than 10%. Someone was advocating a 1% or 1/2% fill and Serguei was making points against this. Bpringlemeir 02:21, 12 September 2007 (UTC)

Also, you are correct about the QRT being relevant for the last hop only. The original research paper suggests a multi-hop table. Bpringlemeir 02:33, 12 September 2007 (UTC)

[edit] Why compare Gnutella to Gnutella2

Part of my NPOV claim, Why bother having a section named "differences between Gnutella and Gnutella2"? Why aren't there sections on differences between FastTrack, eDonkey2000, DirectConnect, BitTorrent, Napster, etc. ? Bpringlemeir 05:32, 11 September 2007 (UTC)

Because the network is not called FastTrack2, ED2k2, DC2, BT2, Napster2, or so forth. People assume that the networks are related (And they are, Gnutella2 being forked from Gnutella), and thus a section on the differences between the two. -FrYGuY 05:40, 11 September 2007 (UTC)

You can state they are different without comparing them. Bpringlemeir 05:45, 11 September 2007 (UTC)

Yes, but comparing the fork, to the network which it forked from, by explaining the differences between them seems to make sense. -FrYGuY 17:43, 11 September 2007 (UTC)

[edit] Gnutella2.com

For those wondering, Gnutella2.com was hijacked, hence why it has been down. If you edit your host file to redirect "gnutella2.com" to 69.4.33.4 you can still access the page. However, it's still useless for citation as it has been for the few weeks because, unless you edit your hosts file, you're simply going to get redirected to the hijacked domain squatter. -FrYGuY 09:11, 30 September 2007 (UTC)

I tried this. I also tried the url [2]. The best I got was a page saying "It works!". Can you give an example of how to reach these pages? Also, does the licensing for the text prohibit it from being mirrored? Or is it expected that the squatter situation will be resolved soon? I have always seen the domain registered to a Michael Stokes in Perth, WESTERN AUSTRALIA. I am not beating a dead horse. I am genuinely interested in the protocol. Thanks, Bpringlemeir 14:28, 30 September 2007 (UTC)

This is the best description of the process. I don't know how to direct link as the server is using virtual hosts, and I don't know how to find the "real" URL. The license of the G2DN wiki is GFDL 1.2, so it's entirely possible to mirror. Unfortunately, I am not an administrator of the domain, and haven't been able to get any answers in terms of getting it fixed or a time frame of when it's going to be fixed. -FrYGuY 06:02, 1 October 2007 (UTC)

This link at the web archive works for me. Thanks, Bpringlemeir 17:16, 1 October 2007 (UTC)

The site is mirrored at http://g2.trillinux.org/. I've corrected the links in the article. 217.26.163.26 07:41, 7 November 2007 (UTC)