Can the Economy of Giving model serve as a model for the unchoking algorithms in BitTorrent networks?

University of Amsterdam

Faculty of Science

Information Studies – Business Information Systems

Master‟s Thesis

Can the Economy of Giving model serve as a

model for the unchoking algorithms in

BitTorrent networks?

Preface

Title

Can the Economy of Giving model serve as a model for the unchoking algorithms in BitTorrent networks?

Abstract

In this study, an economic model called Economy of Giving and the peer-to-peer file sharing field is discussed. The purpose of the thesis is to explore the notion of that model and find out whether it can be related to the P2P sharing field and more specifically around BitTorrent technology. Literature and internet research was conducted in order to come to a result and eventually come up with a suggested mechanism.

Throughout this thesis it was possible to come up with a suggested mechanism that works according to the basic principles of Economy of Giving model and can complement the existing BitTorrent‟s mechanisms. However, there might be the restriction that it would probably work better in private closed torrent communities for several reasons according to research. Therefore an alternative implementation of the Economy of Giving model is suggested that applies to public networks well.

So, that was done after additional analysis and modification of the initially proposed mechanism. Finally, the main finding is that the model can be related to the P2P file sharing field and that the choking mechanism is actually an implementation of the model, even though it already exists in BitTorrent with a slight difference on how it would work. Thus, it is shown that BitTorrent‟s unchoking algorithm is an instance of the model and necessary function to have the suggested mechanism working properly.

Keywords

Economy of Giving; BitTorrent; Peer-to-Peer; File-Sharing; Qualitative; Unchoking Algorithm; Tit-for-Tat; Optimistic Unchoking.

ii

Acknowledge ments

This project would not have been possible without the support of many people. I would like to give many thanks to my supervisor, Peter Weijland, who offered guidance and support and helped make sense of confusion. Also I would like to thank my family who endured this long process with me, always offering support and love.

iii

TABLE OF CONTENTS

Preface... i

1. Introduction ... 1

1.1 Aim of the Study and Research Question ... 2

1.2 Significance of thesis ... 3

1.3 Definitions... 3

1.4 Outline... 3

2. Method ... 5

2.1 Literature Research ... 5

2.2 Internet Research setting ... 5

2.3 Semi-Structured Interviews... 6

3. Literature and Internet Research ... 7

3.1 Economy of Giving ... 7

3.2 BitTorrent... 12

4. Related Work ... 18

5. Design of Prototype mechanism ... 20

5.1 Economy of Giving related principles ... 20

5.2 Description of proposed mechanism... 21

6. Results and Analysis ... 25

6.1 Relation and Feasibility of proposed mechanism ... 25

6.2 Additional Suggestion... 25

7. Discussion and Conclusions ... 29

7.1 Research Questions Revisited ... 29

7.2 Limitations ... 29

7.3 Future Research... 30

7.4 Concluding Remarks... 30

8. References ... 32

9. Appendices... 35

9.1 Appendix A: Interview Protocol ... 35

9.2 Appendix B: Interviews ... 36

iv

1

1. Introduction

Peer-to-peer (P2P) systems have become a really popular medium for file sharing especially for huge amounts of data because of the technology they provide for cooperative distribution among users. In P2P systems costs for sharing data are distributed, meaning that each peer not only downloads but also uploads part of files, thus allowing faster download speeds compared to a server which could potentially crash when it gets too overloaded. File sharing exists in many different forms, protocols and networks such as Gnutella, eDonkey and BitTorrent, with the latter being the most popular since it is responsible for a considerable amount of traffic on the Internet according to a report by Envisional (2011) and this is where this thesis is going to focus due to the popularity of the protocol.

Figure 1 Distribution of Internet Protocols, Env isional 2011

There are still however issues in P2P systems that must be overcome, one of them being the fact that when there are no counter measures to prevent peers from not uploading any files, they free-ride (Adar & Huberman, 2000) and the reasons behind this have been extensively studied (Mol et al, 2008; Ripeanu et al, 2006). Peers who download files without distributing any in return, are called free-riders and one of the ways to prevent this is by providing several incentives to make them contribute by uploading.

Since P2P systems involve resource exchanges and utilization of services and there is a desire for equilibrium on the amount of contrib ution of files among peers, it allows viewing these systems as markets where traders exchange goods and always expect

2

something in return when they offer something, unless it is a gift. Furthermore, according to Krishnan et al. (2004) it is important to understand the economics dynamics of P2P networks in order to develop mechanisms and protocols that will efficiently operate in the long run. Early discussion by Buyaa et al. (2001), has already focused on the possible economic models in P2P and classified the economic models into categories. Also, research has already focused in the economic aspect of P2P systems by trying to relate it to economic notions such as auctions (Levin et al., 2008) or gifting (Ripeanu et al., 2006) and analyze individuals‟ sharing behavior and even more studies have focused on how peers can be incentivized to share more by uploading content after downloading a desired file.

Stepping onto this view, in this thesis I am going to try and relate P2P file sharing to a notion called Economy of Giving, which will be presented and explained to the reader in a later section of the document.

1.1 Aim of the Study and Research Question

Primarily the aim of this study is to try and relate the notion of Economy of Giving to other domains as suggested by Weijland (2014) and in this case computer science field and more specifically P2P file sharing and the unchoking algorithm that BitTorrent uses as mechanism in order to prevent freeriding. Avoiding useless bandwidth usage and computational cost is one of the main reasons why peers free-ride (Feldman, 2003). Hence, a mechanism that would take into account this parameter would be of great importance and this is the target of this thesis; to suggest a mechanism that can be derived from the Economy of Giving principles and thus be related to P2P field.

It should be noted however, that the purpose of this thesis is neither to delve into nor to come up with new mathematical foundations for Economy of Giving or delve into computing issues such as performance and scalability of a suggested mechanism. This thesis serves as an exploratory research and the target is to introduce the notion of Economy of Giving in an easy to understand way and try to relate it with P2P systems through the suggested mechanism that will emerge from analyzing the basic principles of Economy of Giving model and to discuss how these could be implemented.

Therefore, the research question that comes out is,

“Can Economy of Giving be related to P2P file-sharing field and more specifically to the BitTorrent’s unchoking algorithm?”

As sub-question the following is formed,

“Is it possible to come up with a mechanism based on the principles of the Economy of Giving model?”

3 1.2 Significance of thesis

The significance of this thesis is actually the introduction of an economic notion to the reader and the exploration and relation of this notion with another field, as suggested by Weijland (2014) and more specifically an effort on establishing the unchoking algorithm of BitTorrent as an instance of the Economy of Giving model. In order to come up with useful results this thesis will suggest a mechanism for BitTorrent networks. Even though there are mechanisms that try to prevent free-riding, research has focused mostly on how users can be incentivized to contribute and how to prevent free-riding by having either counter measures, such as banning the user and not allowing further download, or motivating the gifting behavior of people and how this can be used to develop new technologies (McGee & Skågeby, 2004). Also there is extensive research on how incentives can help increase contribution of peers, but this thesis is trying to present a novel approach by introducing a new economic model and deriving a mechanism out of the basic principles of it. The focus is not on humans, but on how a mechanism could help in the effort of not allowing free-riders to download parts of files; leaving out the human factor. BitTorrent‟s mechanisms and policies fail to prevent freeriding and unfairness across nodes according to Bharambe et al. (2005), so in an effort to relate the economic model to the P2P field, a mechanism is going to be suggested that complements the current ones and offers insight or helps to lead future researchers to novel ideas or approaches.

1.3 Definitions

Throughout this thesis, I am going to use several terms which might need explanation so as to facilitate understanding. In brief, BitTorrent peers are referred to as leechers if they are downloading (and uploading) a file and seeders if they have the complete file and are uploading it to other leechers. A Tracker stores information for an available set of peers who own a file. This set is called a swarm and is responsible for helping leechers in discovering other peers and allowing them to download the files in pieces which consist of blocks and make a complete file once downloaded. Typically, trackers, also called tracking servers, are active BitTorrent entities that keep information about which peers are currently sharing a given file.

A glossary of the specific terminology used throughout this thesis is included in section 10.

1.4 Outline

The remainder of the document is structured as follows. In section 2 I will present what type of research I am conducting in this thesis and will also present the methods used to gather material. In section 3, I will introduce the Economy of Giving by presenting literature on the foundations of giving and gifting and s how basic differences in order to better explain the notion. Furthermore, an online community which follows a model similar to Economy of Giving is going to be presented in order to show an actual application of another economical model and how it can be done. Also, background on the BitTorrent P2P file-sharing protocol is provided to show how it works and the basic mechanisms are analyzed. Adding to that, some online

4

file-sharing communities are presented so as to analyze the different policies followed in different cases. Section 4 will include some related work that has been done with regard to BitTorrent protocol and economic notions and models. Next, in section 5 the suggested mechanism will be described after analyzing the principles that are going to be used from the Economy of Giving model and section 6 will include the results and feasibility of the suggested mechanism and therefore the relation to P2P systems.

Lastly, the thesis will conclude with section 7 by including the concluding remarks, revisiting the research questions and suggestions for future research regarding this work.

5

2. Method

This thesis serves as an exploratory qualitative research for a notion and an effort to relate it to another field (economics to computer science) and it is based on gathering data by literature research complemented by Internet research. I searched (a) Google Scholar; (b) the references in the articles I found; (c) online full text collections of publishers (Springer Link, Sage Journals online) in order to complete this research. However, the literature on the P2P field was mostly old. In order to get recent views on the field, two interviews were conducted with researchers in P2P who were asked questions that could cross validate informatio n already gathered from sources such as published scientific papers or were found online.

2.1 Literature Research

As for the P2P field, literature research was focused on finding information on how BitTorrent protocols work, what has already been done in the field and where researchers focus in order to solve problems such as free-riding.

As far as Economy of Giving is concerned, research focused on finding notions related to the foundations of the concept of Economy of Giving such as reciprocity and how primitive societies exchanged goods and gifting, in order to introduce the notion to the reader more easily and to highlight the difference with gifting behavior.

2.2 Internet Research setting

Web study was conducted in order to get more recent views on the P2P field from blogs or relevant technical sites and forums related to the field. Internet research was focused on finding torrent sites that were handling distribution of files in a different way among them. Some torrent sites that were found include; The Pirate Bay1, IsoHunt2, EZTV3, TV torrents4, TorrentLeech5 and PolishTracker6. These are just a sample of the total amount of similar sites. The Pirate Bay, IsoHunt and EZTV are public communities. The remaining three, however, are private torrent sites which means closed communities about which information would not be easy to obtain. Nevertheless, web study showed that there is research done already in both types of sites and alongside information from technical blogs it provided a sufficient amount of data. Apart from that aspect, web study was also done in an effort to find potent ial online communities that embrace some kind of economic model that can be related (similarities and differences) to Economy of Giving and proved to be successful; the online community found is named JOATU and will be presented in the next section in detail.

1_{Available under the link: http://thepiratebay.se} 2_{Available under the link: http://isohunt.to/} 3

Available under the link: http://eztv.it/ 4

Available under the link: http://www.tvtorrents.com 5

Available under the link: http://torrentleech.org/ 6

6 2.3 Semi-Structured Interviews

Semi-structured interviews were conducted with researchers from the P2P field. Unfortunately, the effort on interviewing more researchers was not fruitful so I had to continue with two of them after receiving some negative answers or no answers at all. However, since the purpose of the interviews was to support existing somewhat older literature, the amount of interviews was sufficient, especially since the outcome was the same and the information was cross validated. Semi-structured interviews were selected as a type by giving more freedom to the interviewees so as to gain more insight from them (Robson, 2011, pp. 285). Eventually, this proved to be useful since interviewees almost answered some of the coming questions without even being asked and there was no problem since I did not have to stick to a strict protocol as I would have done in a fully structured interview.

The interview protocol can be found in Appendix A and the answers of the interviewees in Appendix B.

2.3.1 Participants

The interviewees were found after doing a sufficient literature study and gathering some names from scientific papers. Interviewees had experience in P2P, their field of research was specifically P2P networks and they have both been involved in development programs.

I contacted the interviewees via email and arranged a time, date and place to conduct the interviews. It should be noted that the interviews were conducted in different contexts; one was a face-to- face interview while the other was via Skype (tele-conference). Also, the interviews were both recorded in order to assist in taking notes after asking for permission from the interviewees.

More specifically the interviewees were:

 Assistant professor Spyros Voulgaris – Computer Science Department, Faculty of Exact Sciences, Vrije Universiteit (VU)

 dr. A. Arno Bakker – Scientific programmer on large-scale distributed systems in the VU‟s Computer Systems group

2.3.2 Interview Context

As mentioned above, the context of the interviews differed. One was face to face while the other was via Internet and more specifically a popular VoIP program called Skype. However, as cited by Robson (2011, pp. 292), both face to face and interviews via the internet produce viable data and since social cues of the inter viewee were not important there is nothing to argue against an interview by Skype for the purpose of this thesis.

7

3. Literature and Internet Research

As noted by McGee and Skågeby (2004), in economics even an “altruistic act” of giving is conceived in terms of immediate or expected return. However, authors state that giving is not easy to be explained in terms of traditional economic concepts such as transaction, profit or utility. Therefore, in an effort to define and present the notion of Economy of Giving, literature research has been conducted in the foundations of giving and several views of researchers in forms of giving have been presented. More specifically, the following forms of giving are going to be highlighted as background knowledge:

 Altruistic and Charitable Giving

 Gifting

 Reciprocal Exchange

Definitions, behaviors and what drives giving in each of the above categories is going to be presented in order to gain the ability to spot similarities and differences with the model of Economy of Giving and thus make it easier for the reader to fully understand it in section 3.1.4 below. In brief, literature found on giving was mostly relative to economic notions of exchange and reciprocity. Also, an online community marketplace based on an economic model is presented in order to show how theory is applied to practice.

Furthermore, internet research is presented in addition to literature research so as to define the basic policies of the BitTorrent protocol and explain the mechanisms used in an effort to prevent users from free-riding. Lastly, several communities and the way they work are also presented in order to see how these try to stop free-riding.

3.1 Economy of Giving

3.1.1 Altruistic and Charitable Giving

One form of giving as found in the literature is giving out of pure altruism, meaning that some people give something for the welfare of others and as a notion it can be considered to be the opposite of selfishness. Though, research on this topic has focused on charitable giving, the forces that drive people to give without asking anything in return are analyzed to determine whether true altruism is possible. According to Andreoni (1989), there are two reasons why people give and they are; (a) people contribute to the public good because they demand more from it; (b) they get a private good given the name “warm glow” the latter being a selfish motive. Furthermore, a more recent study on charitable giving by Bekkers and Wiepking (2010), showed that the forces that drive this type of giving are; (a) awareness of need; (b) solicitation; (c) costs and benefits; (d) altruism; (e) reputation; (f) psychological benefits; (g) values; (h) efficacy. Therefore, even though it is not clear whether this type of giving can be considered pure due to the motives that are behind every individual, in any case it is giving without demanding and this serves the purpose of reference here.

8 3.1.2 Gifting

Sometimes people give without expecting something in return, something that in layman‟s terms is called “gifting” (McGee & Skågeby, 2004).There is a strong desire of the humans to give, whether that is advice, digital goods or other resources. In commercial economies where exchange-based transactions take place, individuals act in order to increase their own return while in gift economies7they may act for the good of the community or others in it. In their article the authors also, mention that usually reciprocity is what motivates sharing among people, but they indicate that people sometimes just enjoy giving while not expecting something in return and categorize such behaviors as follows in order to explain what motivates them:

 Pseudo-Gifting

Giving something away, but there is a benefit for a business such as business cards or free product samples or personal benefit such as meeting new people.

 Social Gifting

An individual makes a gift for a social function either because of social pressure to maintain social agreements or the expression of social relationships.

 Ideological Gifting

The explanation given here is that someone is giving due to ideals, in the sense that "this gifting will make the world a better place."

 Altruistic Gifting

This is when individuals just gift and generously offer something without expecting anything in return, similar to altruistic giving already analyzed above.

A more technological approach by the authors also covers what people can gift online since because of the web there are a lot of communities where someone can contribute and they categorize as follows:

 Expertise

That refers to the fact that many people share their expertise on internet forums for free and get nothing in return. Similarly, people who moderate discussions not only get nothing but they also face criticism from other users.

 Artifacts

This actually refers to the file-sharing phenomenon with music, movies, books and software where individuals share via the internet or the help of the internet – give other users information about what he offers.

7

Definition fro m Business Dictionary:

An economy based on giving in the context o f re lationship rather than making transactions simply for profit or personal materia l gain.

9

 Storage and Bandwidth

Includes projects based on users‟ donation of unused resources such as storage or bandwidth. Examples include universities that donate space on their computers in order to make resources publicly available.

As per the categories of expertise and social gifting, some researchers believe that involvement in the open source development happens due to the size of such communities since they are small and there is social pressure to contribute (Zeitlyn, 2003). More research on the topic by Hars and Ou (2001) identifies two types of motivation for participation in open source projects and these include (a) internal motivation, altruism and identification from the community – in this case it is similar to social pressure – and (b) external rewards such as expected future returns or personal needs with the latter category playing a greater role in the motivation of actively participating.

3.1.3 Reciprocal Exchange

As defined by Kranton (1996), reciprocal exchange (also called gift exchange)has to do with agreements where people give goods, services, information or money in exchange for future compensation in kind and as it is cited, this phenomenon was mostly seen in primitive societies. However, researchers have shown that this phenomenon is still prevalent in modern societies as well. Reciprocal exchange relationships take place between two people who know each other and the reason for this bond is because the more people engage in such an exchange the harder it becomes to exchange commodities on a market, which is one of the reasons it is still pervasive in modern markets. Also according to the author, in a two-people reciprocal exchange, if one “cheats” and does not return any goods back to the supplier when it is expected, then their relationship is terminated since one of them (the receiver) broke the unwritten, however existing rules, of their honest and reliable relationship. The receiver though, can easily utilize another connection with another supplier especially in markets where anonymity prevails and the only benefits that the receiver is losing are the ones that he could have from this one broken relationship, but use others without any problem. In this case as stated by the author, the only big problem that could come up is the case where many people in a market had engaged in reciprocal exchanges between them and thus the search for a new connection would be of great cost to the “bad” receiver. This phenomenon is mostly observed in primitive societies or current traditional societies where the markets are small and people are willing to provide goods to others in anticipation of future settlement of the exchange and if there are reputation mechanisms where a “bad” receiver would find it more difficult to settle a new transaction the cooperation can be stronger.

3.1.4 Online community marketplace

Jack Of All Trades Universe8 (JoatU) is a web application designed to foster the

8

10

generation and growth of multiple local and global online economies. This application allows individuals to offer services as well as products and it also uses a currency called “Jack Of All Trade Unit” (JoatUnit) supporting local community projects. JoatU exists as an online “hyper- local” marketplace for people living within a small distance and are interested in offering or receiving goods and services through a variety of types of exchanges. It is an “anything- for-anything” barter exchange system. JoatU is designed for the underemployed, unemployed, for those with goods to rent or sell or for those interested in acquiring new skills or volunteer.

It is a transitional economic platform that does not limit itself to one style of market economics. It allows capitalists to sell their products and services for legal tender, communists to enact direct one- for-one exchanges, and socialists to benefit through the creation of the JoatUnit. The JoatUnit is a means of exchange that is generated to represent the value of someone‟s work done for the benefit of the community. The algorithm of generating this complementary currency is different from other currencies, and is based on the evaluation by the community. For example, one could teach a class, or offer to lend a tool to the decentralized tool library and be paid in the community currency for their offering.

The vision of JoatU is to build a community that seeks to promote human welfare, thus the system is designed to help grow a community- generated welfare state that supplies for the basic needs of its citizenry. The value of the unit is only derived from its acceptance; therefore it is imperative that a portion of the user base accept the JoatUnit for a portion of their trades. There is no JoatUnit generation for one-on-one trading; these trades are organized directly between the parties involved and can be done with exchange of goods, cash payment, Bitcoin9 payment or JoatUnit. Another thing is that as JoatUnit currency exists it can encourage the community to share a variety of different services and products with each other. This means that one person can exercise her/his different skills and does not have to specialize in just one. It could be seen therefore as reciprocal exchange of goods or services within an online community. However, JoatU uses reputation mechanisms by having reviews for every supplier so it is not easy to act in a way that is not accepted in a market.

JoatU serves as an example found from internet research where economies and models can be applied not only in standard markets, but also in online communities and many other fields such as technology and computer science; especially in P2P, networks economics are closely related to the way these work since they include transactions among peers and exchange of resources and files.

3.1.5 Economy of Giving Foundations

The Economy of Giving model incorporates characteristics from every type of giving and also has principles that differentiate it from them. According to Weijland (2014), Economy of Giving would have application in a world where goods are exchanged or given away without any currency to express the value for each transaction. Instead, a means of a system of social credit would be used in place of an alternative money of account. Therefore, the basic principle of this model is that every individual keeps track of a mental account of social credit in relation to another person individually.

9

11

Both participants in the transaction keep track of these credits and always take a look at their balance. So, it becomes obvious that the model of Economy of Giving makes use of the notion of “Account Balance” in the sense that a supplier keeps record of all the transactions with another entity and expects him to counterbalance that at some later point.

Furthermore, according to the Economy of Giving model, transactions between two users are being done in relation to values of some attributes. One of these attributes is account balance that is already introduced above. The other one is called yield which is referred as the present value of a future settlement of transaction as perceived by an entity who supplied a good and its relation to the account balance; the higher the account balance the lower the yield value. For instance, if entity A has given several goods to entity B and has not received much, or anything, in return, then the account balance of social credit is increasing and thus A values every future settlement of transaction less in relation to B, meaning that the yield value of the transactio n is low and it is less likely that another transaction is going to happen and added as another debt. This principle is called “The law of diminishing returns”.

The author explains another important principle on how the model works and more specifically what happens when a supplier can give away some goods and he has to select between two receivers. In this case the “Highest Yield Rule” axiom (HYR) is used which in simple words means that the supplier will select to have a transaction with the entity for which the yield value is higher than the other. For example A values a transaction with B higher than a transaction with C so it starts the first one until a certain point. Then it checks again the account balance and the yield value for both entities and in case the yield value for the transaction with C is higher now it will pick C, otherwise it will continue with B.

In order to explain more principles of the model, one would have to dive into complex mathematics and this is not intended here, so the more important attributes of the model were explained in order to give insight to the reader as to what this model is about. As far as the other types of giving are concerned, it can be obvious that giving with altruism, charitable giving and gifting are much d ifferentiated from giving as explained in the Economy of Giving concept, even though if gifting is done with an ultimate goal of settling a future debt resulting from some gift then one can see a small similarity here. However, reciprocal exchange has much in common with this type of giving with the only difference being that reciprocal exchange is based on a two people relationship and the premise of knowing each other well, while Economy of Giving models argues that an entity selects a transaction based o n an account balance of social credit and how this is related to the value of the yield for this specific transaction and it can easily change the choice between transactions with different entities as explained above via the HYR axiom.

3.1.6 JoatU & Economy of Giving

JoatU and Economy of Giving have more differences than similarities. JoatU successfully embraces an economic model but not similar to Economy of Giving as it will be explained here. In JoatU people can exchange goods, something that is also introduced in Economy of Giving, thus it can be considered one common point.

12

Another similarity is that there is an alternative expression of currency in both cases. In JoatU, users apart from exchanging goods they can pay by money or even with the JoatUnit. This could be associated in a way with the account balance and credit system notion that exists in Economy of Giving which is to be applied in a world where there is no currency to express the value of goods. Hence, the creation of a way to represent currency exists in both cases and constitutes a similarity.

However, there is one big difference that clears out how Economy of Giving works. In JoatU if a person agrees to give away a product or a service to someone else and the receiver will not pay d irectly then this is considered bad behavior while in Economy of Giving this would not be seen as a loss, instead a social credit in terms of debt would be added to the account balance of the supplier for the receiver as mentioned above.

So this comparison was done in order to make more clear what Economy of Giving is and which properties and principles of JoatU exist in it and which do not.

3.2 BitTorrent

Literature on BitTorrent was mostly older than 2-3 years, so interviews with researchers were conducted in order to get recent views on the protocol and its mechanisms and thus support the literature. In this respect internet research was conducted as well, mostly in technical blogs to get recent valuable information.

In brief, P2P file sharing10 is totally different from traditional file downloading. Instead of using a web browser, a software program is used to find computers – called peers – that have the requested file. For instance, a computer user is running a file-sharing client and requests a file to download. The client then queries other peers connected to the internet who also have the software. When a peer that has the file is found, then the file sharing software can download the files from that peer‟s hard drive. The most used P2P file sharing protocol is BitTorrent (Envisional, 2011; Zhang et al., 2011; Chen & Jarvis, 2009), hence this is going to be analyzed below.

The BitTorrent ecosystem consists of three major components: peers, peer discovery mechanisms, and torrent-discovery sites (Zhang et al., 2011). Torrent is the collection of peers participating in the distribution of a file and each one is identified by an identifier called info-hash. A peer in a torrent at given time is called either a leecher or a seeder; where the peer possesses a part of the file and the whole file respectively. The process is the following:

When a peer wants to start a new torrent, it needs to seed the content, so it is becoming the initial seeder, register the torrent with a tracker and upload the torrent to a torrent discovery site. A torrent file contains Meta data about the file to be shared including; filename, size, number of pieces, hashing information for the pieces (so that leechers can verify the integrity of the pieces they download ) and the URL for the tracker. A tracker is a mechanism for peer discovery and acts as a central server keeping a list of all peers participating and when a peer joins a torrent it actually

10

Information on P2P file sharing retrieved fro m: http://computer.howstuffworks.co m/bittorrent1.ht m

13

registers with one or more trackers. A tracker can be contacted by peers to obtain information for other peers in the network, called a swarm, as shown in the figure below.

Figure 2 A peer asks tracke r for informat ion about other peers 11 (Norberg, 2006)

Peers learn about the existence of a specific torrent from torrent discovery sites and these can be either public or private online sharing communities. After this point, a peer gets a torrent, the IP address and port number of the tracker and joins the BitTorrent overlay for that torrent. Then, peers become leechers or seeders if they have the complete set of pieces of the file. Peers use a BitTorrent client and can communicate with other peers via the BitTorrent protocol in order to connect and exchange pieces of files. When only the initial seeder has the file, the newly joined peers are connected directly to him and request the pieces. When the pieces start spreading and other peers now own some too, the peers start to trade and exchange the pieces between each other instead of downloading directly from the seeder. Eventually, the process continues by having seeders uploading data to leechers (Figure 3), while the rest of the peers interact with each other to decide who to download data from or who to upload to. Mechanisms responsible for this are going to be explained in the next sub-section. Finally, once a leecher finishes downloading a file, it can join as a seeder or leave immediately thereby becoming a freerider or also called exploiter (Chen & Jarvis, 2009).

11

Norberg, A. (2006). Introduction to BitTorrent. Presentation at Umeå University. http://www.rasterbar.com/products/libtorrent/bittorrent.pdf

14

Figure 3 Peers included in a torrent e xchange part of files with other peers (No rberg, 2006)

There is an interesting visualization of how BitTorrent works by representing seeders and peers exchanging files. It is written in JavaScript and is available online from the website TorrentFreak.com in the following address: http://mg8.org/processing/bt.html

15 3.2.1 Policies and mechanisms

The basic philosophy of BitTorrent protocol is that it focuses on the bandwidth of peers to efficiently distribute contents on a large set of peers and also that the transactions are handled among peers without the interference of any other entity. For instance, leechers contact the tracker at set intervals to ask for a new peer list in order to boost their download speed, although the tracker is not involved in the actual distribution of the file even though it connects and communicates with peers. This list of peers follows some policies and rules such as a predefined threshold of how many peers can be included. Typically that limit is 20 peers while the high limit is 80 and if the set size falls below that number then the peer will contact the tracker to obtain a new list of IP addresses of other peers (Legout et al., 2006). Furthermore, another policy which is followed by BitTorrent clients is that each peer reports its state, amount of bytes uploaded and downloaded, to the tracker every 30 minutes or when disconnecting. Adding to that, each peer also sends messages to every other peer in its list when it has downloaded and verified a new piece of a file.

Since, in BitTorrent each peer is responsible for trying to increase its own download rate; the effort is made by downloading from whoever they can from the swarm and uploading with a variant of tit-for-tat12. When a peer is said to cooperate it means it uploads and when it decides not to cooperate it „chokes‟ other peers; meaning temporary refusal to upload to a specific peer who is not uploading (freeride r). This process is done by using two mechanisms called “Choking Algorithm” and “Optimistic Unchoking” (Cohen, 2003). More specifically, this is how the choking algorithm works: A leecher prefers peers with better upload rates and after it has downloaded a file, it is selective about who to upload to (unchoke). Decisions as to which peers to unchoke are based strictly on current download rate from them (Cohen, 2003). This explains how BitTorrent uses a tit- for-tat strategy. A peer then usually unchokes whoever has uploaded to it and every 10 seconds (amount of time sufficient to give TCP the time to have the transfers in full capacity) it will check other peers in the neighborhood and choose new neighbors to unchoke based on their actions the last 20 seconds and split it upload bandwidth for the peers it is uploading to. An example to illustrate the process is the following: If peer A is not uploading to peer B then peer B will not unchoke A and therefore A will not be able to download from B. On the contrary, if A uploads to B with a good speed, then B will unchoke A and A will increase its download speed. So the philosophy is that in an ongoing transaction among two peers, if one stops uploading to the other will eventually be choked and will not be able to download and it is one of BitTorrent‟s ways to prevent users from freeriding while providing incentives such as faster downloading speed to users who seed more – the more peers to download from, the more speed you get. As far as the optimistic unchoking is concerned, it means that every peer randomly chooses another peer in the swarm to upload to every 30 seconds. This is really useful because it allows new peers in a swarm that cannot upload anything to other peers since they do not posses any part of the file, meaning that they would stay choked otherwise, to download parts of it and be able to upload too from now on.

12

The basic idea of tit-for-tat is that one agent will a lways cooperate as long as its opponent cooperates.

16 3.2.2 Torrent-Discovery sites

The way to learn about ongoing torrents and be able to get them is via discovery sites such as Pirate Bay, IsoHunt and many more. According to Zhang et al. (2011), some of them provide also tracker services; Pirate Bay is one while IsoHunt is not. Here it should be noted that most of them provide a .torrent file with all the Meta data it contains, but there are also sites that only serve as search engines and provide links to .torrent files such as Torrentz13.

Pirate Bay and IsoHunt and EZTV are examples of public communities where users do not need to register in order to download a torrent. These public communities have no download restrictions or any basic rules by any means and rely on BitTorrent‟s mechanisms to prevent freeriding. However, most of them also have discussion forums, content rating, comments and user profiles which can actually serve as an incentive for some users to contribute and get good comments on the content they offer or if someone mentions their contribution on a forum.

On the contrary, private communities also exist and have rules that must be followed and incentives that vary per site and require registration before being able to download torrents so as to curtail freeriding. More specifically, Meulpolder et al. (2010) focus on the following popular communities and mention specific characteristics about them:

 TorrentLeech: This community has closed membership and one can only be a member if invited by an already registered member. There is a ratio policy (specific amount of upload for a specific amount of download). Each member has to seed with a ratio of 0.4, meaning 4 bytes uploaded for every 10 downloaded, or seed the complete file for more than 24 hours. Another requirement is to have a minimum overall ratio of the same value. If a member is not following the rules he is warned and if in 5 days time he is not able to regain the minimum ratio, his account is deleted.

 PolishTracker: This community also accepts new members only by invitation. It has a stronger ratio policy (1.0) or seed the complete file for at least 48 hours and a 0.55 ratio overall, which is considerably more than TorrentLeech. Furthermore, members who do not follow the rules are warned and then banned from the site. In this community they aim to keep exposure to a minimum, thus ensure more efficient participation and easier control.

 TVTorrents: Invitation here is also the only way for someone to be able to register. This tracker makes use of a credit system in the following way; in order to download, the cost is one credit and in order to get credits you have to upload. Credits are given per uploaded byte. When a member has no credits, he cannot download at all.

Some rules regarding such private sites might include the content of uploads also. Sites may require that the files uploaded should be of a certain quality format. Also invitations have a reflection on the user who invited a person. In some cases, the

13

17

selfish behavior of someone who is going to get banned unless he conforms to the rules, could also cause the banning of the member who invited him, depending how strict the policy of the site is. Finally, if a user cannot find someone to invite him in such a private community, there is the process of an intense interview to get an invite and selling invites is strictly prohibited in most cases (Lifehacker14, 2010).

As a final important note on BitTorrent communities, research (Meulpolder et al., 2010) has shown that the ratio of seeder/leecher is more than ten times larger than in public sites and that peers seed more in private communities in addition to having seeders for all data provided in a private site, thus making BitTorrent‟s mechanisms for preventing freeriding unnecessary.

14

18

4. Related Work

Different studies about the BitTorrent Protocol have been done over the course of the last years, ranging from mathematical modeling to empirical studies. Many of these studies focused on its relation with economics and different models in an effort to better analyze BitTorrent‟s mechanisms and suggest solutions to problems that exist in the protocol. Levin et al. (2008), focused on proving that BitTorrent does not provide users with incentives to contribute and follow the principles of the protocol. Instead they suggested that an auction model can help study BitTorrent and improve the incentives. They contributed by showing that BitTorrent is not using tit-for-tat strategy so accurately, since it shares some properties of tit-for-tat, and that the auction model would fit better in describing it and give insight in designing a better incentive mechanism. The approach they took led them to suggest a replacement of the original BitTorrent‟s unchoking algorithm and that is a proportional share auction clearing where peers are rewarded with an amount of good part to how much they bid – the more you give the more you get philosophy.

Moreover, Ripeanu et al. (2006), study gifting behavior and several P2P file sharing communities and suggest changes for the BitTorrent protocol and the design of the communities. More specifically, they study motivations for contributing and analyze freeriding in several online file sharing communities, public and private alongside with the ways they use to prevent it, and the reasons behind it. Finally, they come up with two suggestions for the protocol and the torrent sites in order to increase the seeding time. First, is to change the default client behavior to stay online for a time respective to the size of the file downloaded and second, that administrators of torrent sites could make users conform to a maximum file size to be published since they found that smaller files had higher seeding ratios.

Finally, another research focusing on incentives and freeriding was conducted by Jun & Ahamad (2005) and showed that freeriders finish downloads as fast as peers who seed a lot, meaning that they are not punished by any means by the existing protocol. Another observation they did on BitTorrent was that tit-for-tat strategy is not as effective, because it starts the transaction first and then by checking the current contribution of each user it takes action. This, however, means that the “bad” user has already downloaded a part of the file. So they suggest a new simple mechanism (tit-for-tat based) in order to prevent freeriding, and the basic principle is that peers maintain the upload (u) and download (d) amount for each link and calculate the deficit of that link for these values. A fragment expressed as constant c is also defined and a variable f≥1 called “nice factor”. Each peer then, ensures that the deficit is restricted up to certain point – u-d ≤ f*c at any time –a maximum upload rate is allowed, leading to an even upload rate for all links and the amount a peer is going to risk to establish one connection is defined. So within this condition peers benefit more by cooperating rather than trying to freeride by exploiting the user, since a bigger upload rate will lead to a bigger upload rate from other peers as well.

Other aspects of BitTorrent have also been studied and these include mostly performance and scalability issues, but are not closely related to this study so there are no references included in this section.

19

What differentiates this study from the above referenced papers is that here the intention is not to replace the mechanism of BitTorrent or provide more incentives for peers to make them contribute, but instead, by analyzing the basic principles of an economic model, try to suggest an extension that could boost the effort of BitTorrent unchoking mechanism to prevent freeriders from exploiting the P2P way exchanging of files.

20

5. Design of Prototype mechanism

After having done literature and web research in order to get insight in the BitTorrent‟s mechanisms and principles, this section will include which principles of the Economy of Giving model could apply on BitTorrent. This effort aims not only in relating that economic model to P2P file sharing field, but also to provide a novel approach while trying to prevent freeriding. After analyzing the principles a case scenario will be presented alongside with pictures so as to facilitate the reader‟s understanding of the concept.

To begin with, the first principle to be used will be the Law of Diminishing Returns. More specifically this includes the yield of a transaction between two users; meaning a value affecting the future settlement of a transaction with a specific peer. For this purpose the current contribution level of a peer to another can be checked and have a credit assigned for that peer. For instance, if a peer is not contributing anything, then eventually that user will be choked as already being done by the existing BitTorrent‟s mechanism. Then, if the same transaction is about to be settled in the future between these two peers, the peer who contributed will value that transaction low and will have to decide whether to settle this transaction or not. This can be done either by taking into consideration another principle from the Economy of Giving model called Highest Yield Rule (HYR), which is optional in the design and described below, or not settling the transaction at all after the first value point that has been credited. The first option is preferred in order to increase chances of finding more peers and giving the chance to peers with bad behavior (freeriding) to change their BitTorrent clients and start contributing.

So, the second principle would be HYR and would apply in the peer selection part where a peer has to decide whether to start a transaction with another peer or not, thus punishing it for freeriding. More specifically, in case a peer is about to start a transaction with another peer the first thing would be to check how it values that specific transaction compared to other available ones from other peers. That way if a peer has not uploaded much or not at all to that peer during a past transaction and there is another available transaction for the same part of file from another peer which is valued higher, then the latter one would be preferred over the former. By having this principle users can be forced to seed in order to have more chances in being selected for having a transaction with other peers over other ones in a swarm.

Finally, in order to carry out these functionalities, the principle of account balance should also be included. Without the account balance serving as storage for credits it cannot be possible to store values for a transaction with another peer and retrieve them later in order to check and calculate which transaction is valued higher according to the yield value. When a peer is not contributing to another peer then the latter one increases the account balance for the former peer by one point. However, the opposite can also be applied. That means that when a peer A has an increased account balance for another peer B, but in a future transaction the peer B eventually starts contributing then the account balance for him can be reduced as would happen with a debt between two traders.

21 5.2 Description of proposed mechanism

Therefore, a change according to these principles can be done on BitTorrent clients in order to alter the way they settle transactions with other peers. A scenario will be illustrated here in order to make clear how the suggested mechanism could work in the P2P environment. First of all, when a peer enters a swarm it gets the list of peers from the tracker and it can see which peers have missing pieces of a specific file it wants. The peer then chooses other peers either based on the algorithm of rarest- first

15

or it just selects randomly by taking the initial risk to upload to a peer that might not give anything in return. The example presented here will include three peers in total for demonstrating purposes. All peers own parts of a file in order to have an exchange of resources. So the “initial seeder” case is not presented here, where all new peers connect to one peer who has the complete file and is just distributing it to others until they can also distribute after getting parts of the file.

To begin with, peer A starts transactions with peers B and C. However, only peer C is uploading files back to peer A while peer B is not (computer firewall or configured client not to do so are two of the reasons for that). The same goes for B and C. C is contributing while B still not sending anything back. In this case, according to Economy of Giving model that is not seen as loss, but credits are going to be added in the account balance of A and C for B. However, since the unchoking algorithm is checking the current contribution of the other peer, B is going to be choked in order to stop freeriding and a new transaction with another peer will start both for A and C in case there are more peers available.

Figure 5 Peers e xchanging resources in a swarm (B is not contributing)

15

The piece of a file that is owned the least by peers and therefore is the rarest. This approach ensures high availability, wh ile classic downloads are sequential.

22

Figure 6 Peer B is choked by both A and C

and its account balance is increased (and decreased for others) respectively

When the account balance of peer A for a peer B is increased then the account balance of peer B for peer A is simultaneously decreased; showing this way that this peer did contribute.

The scenario continues as follows, in a future moment two peers B and C have the same piece of file that peer A is looking for. Peer A then will check the account balance for each peer and according to the HYR axiom the yield value for every transaction will be checked and compared between the two transactions. This will be done according to the following rule; higher account balance  lower yield value. When the yield is low that means that the transaction with that peer is valued lower and is not preferred over another transaction which is valued higher. So, A will pick the peer with the lower account balance. In this case I set the account balance for peer C at 1 credit and the account balance for B at 3 credits; thus C is selected for a transaction with A.

However, in this case there are two things to be considered; (a) prevent freeriding, (b) achieve higher download. As per (a), this can be done by using the existing BitTorrent‟s mechanism called unchoking algorithm or by not having another transaction with a user after a credit has been added for him in the account balance. However, this would cause a problem for (b), since the fewer peers you transact with, the less download speed you get. But the combination of both mechanisms (existing and suggested) can result in better outcome. So, when a transaction is initiated the current contribution level (download rate) is checked by the unchoking algorithm. This is done in an effort to connect to more peers by taking the risk of uploading to freeriders, but still have the chance to stop the tra nsaction with them. Then the

23

account balance principle can exclude peers with freeriding behavior in the past. But, the latter measure would not exclude a peer from a future transaction from the first credit added to the account balance. That system would work by giving more chances to peers to change their behavior but in the meantime decrease the upload rate towards them in order to treat them like freeriders and optimize the likelihood that they will conform and start contributing too. Then, this measure would do the best selection available in terms of increased account balance among many peers and give chances to other peers who might also have an account balance but still lower than others‟ as shown in Figures 7 & 8 (for demonstrating purposes the peer selects only one between the two by comparing the account balance for each). This can also be complemented by an upper limit of how many times a peer will be given the chance to stop freeriding and if it went past the limit it could be choked permanently from the other peer, because exploiters could also take advantage of this soft handling.

Figure 7 Peer A checks the account balance for each of the peers that have the piece of file it wants

In case a peer B already has credits added in the account balance of another peer for it (as described in the example), if it changes behavior and start contributing then credits are going to be decreased according to the model. This could be done in the following way. If the account balance is higher than a specific point, say 4, then a credit can be reduced if 100 megabytes are uploaded to the other peer (this only serves as an example). If the account balance is lower than 2 then 50 megabytes could do the trick and so on.

24

Figure 8 Peer A chose peer C to transact with due to lowe r account balance and B is left as free rider (case where peer A cannot be

connected to more than one more peer)

Finally, optimistic unchoking mechanism still allows new peers to contribute, but an increased account balance may exclude them from future transaction. The disadvantage of the mechanism is that still a peer might have to defect by the initial risk of uploading to a user who is not contributing is there. However, this flaw is balanced out with this suggested mechanism by the fact that this will not happen many more times and also might lead to better download speeds since the peer who has uploaded more each time can be chosen among others.

Therefore, the mechanism suggested here is an extension, so optimistic unchoking and unchoking algorithm still apply and their use would actually be required even if these two mechanisms did not already exist. Additionally, these two mechanisms are complemented by the credit system, the account balance and HYR p rinciples. This helps in punishing freeriders and forcing them this way to contribute in order to be able to download in a certain future moment. It should be noted here that this mechanism works only between two peers for both same and different files. However, it does not work as a reputation mechanism where information about a specific peer is shared with other peers in the swarm. The reason for this is because Economy of Giving does not have such a feature directly, but this will be elaborated more in t he next section.

25

6. Results and Analysis

6.1 Relation and Feasibility of proposed mechanism

The mechanism suggested here is based on the principles of the Economy of Giving model and this allows talking about relation of the model with the P2P file sharing field in that way. However, the question that comes up is the following; is the suggested mechanism feasible? In order to help answer this question both researchers from the P2P field were asked whether a mechanism that would use an account balance which would store credits for specific peers could be feasible and useful. Their answers were analyzed and found to be similar since they both agreed that it can be feasible and could complement the existing mechanism of BitTorrent, but they were not sure whether that much useful in the “wild” internet. By that they explained that the chances of meeting with the same peer for a second time are so little so it could be an overload of service and data usage space without really an important reason to do so. So, the application of such a mechanism in public communities such as Pirate Bay, IsoHunt etc might not function properly, by causing overload issues or even be useless as the interviewees noted.

However, it could be applied in a small-scale network such as a private community. In that case the number of peers participating is way fewer than in a public community due to strict registration constraints (invitation policies etc), especially in communities such as PolishTracker where the philosophy is to keep the exposure of the tracker to a minimum. Thus, the possibilities of meeting the same peer again are increased by far and a mechanism as mentioned above could be very useful in the effort to prevent freeriding and have all the users contributing by increasing the sharing ratio by far. Therefore, it comes out that the relation, in terms of a protocol design, that is trying to be achieved in this study is dependent on the community, private or public, as a first point until an actual development of the protocol can be done and rely on testing results to come up with more conclusions and notes. So, as a first step after this study it is possible to talk about feasibility and possible behavior but only in the design phase and no further.

The purpose of this research as an exploratory research was to relate the model to P2P and establish the unchoking algorithm of BitTorrent as an instance of the Economy of Giving model. Literature alongside with internet research, interviews and the analysis of the principles of the economic model resulted in the design and suggestion of an extension mechanism for the existing mechanisms of the BitTorrent protocol. Hence, I would argue that this research resulted in successfully fulfilling the initial research goal.

6.2 Additional Suggestion

So, by analyzing the principles of the model it was possible to suggest an extension for the BitTorrent protocol. However, it turns out that this might not be possible for public communities because the chances of two peers meeting again are very low. There can be a modification of how the account balance and yield curve principles are handled in order to suggest a solution for this issue and eventually having the

26

extension working in public communities as well and completing the relation of the model with the P2P field and showing that unchoking algorithm would be used here as well as a necessary function.

Such a modification could work as a reputation mechanism by making data publicly visible to every peer in the network. Such a concept is not introduced by Weijland (2014) regarding the foundations of the Economy of Giving model, but a similar idea sits in how the yield curve is affected from reputation and how that reflects on the loss of yield. An example to illustrate this is the following; if A has a better reputation with B then the yield curve will decrease more gradually reflecting less loss of yield even if that peer is in the red. So it is a design issue on whether having the yield curves publicly visible or not and thus the reputation also, but the concept is there. Therefore, in this modification we assume that the trackers monitor the transactions among peers in each swarm. Every peer still has an account balance where it stores credits for peers who are not contributing. This account balance though, is also monitored by the tracker which also has the task of calculating the yield curve for every peer defining that way the strength of diminishing returns and how each transaction will be valued. A peer is not able to see another peer‟s account balance, but is able to check the yield value directly from the tracker. That way the overload for every client is also decreased, since performance and storage issues and now up to the tracker. Thus, clients can still work efficiently and transactions will not be affected at all. Now, other peers will be able to know when a peer did not contribute and when he was behaving normally by uploading and downloading parts of files in the network. So, when a peer behaves badly it will find it difficult to be connected to other peers since they will already know that it is a potential freerider by checking the yield curve for that specific peer.

An example to illustrate this will include four peers. It is assumed that the swarm first has three peers which are exchanging parts of files. However, only peer A and C are contributing while peer B is free riding as shown in Figure 9. The result will be that peer B will be choked by A and C, so the two latter will add a credit to their account balance for that peer. Since the tracker monitors the account balances and transactions, it is now able to calculate the yield value for every peer. Therefore, when a new peer enters the swarm in order to download a file what happens is the following. Assuming a new peer D is entering the swarm and already has a part of file to upload, but it needs a part that both B and C have and he has to choose the one who has more chances of contributing back; meaning the one with a higher yield curve. So peer D communicates with the tracker to get the list of peers that are currently in the swarm alongside with the yield for every one of them. Then peer D will check the values calculated by the tracker for all peers, hence having the ability to leave out or choke from the beginning the one with low yield curve (freeriders).

The case here is also when peer D has only one more space available for connection with a peer so it will pick the peer with the highest yield order to leave out the potential freerider. For demonstration reasons the yield values in Figure 9 are represented by three variables; x, y, z and they are equal. In Figure 10 the yield curves for C and A have increased by one for demonstrating reasons to show that they are higher than B‟s yield which is a freerider. Furthermore, same rules apply as described in chapter 5 (more chances would be given to a user in order to stop freeriding). So

27

this representation shows the selection case between two peers for a piece of file and the one with the highest yield is picked.

Figure 9 Swa rm of peers e xchanging parts of a file

Figure 10 Pee r D checks the yield curve and pic ks peer C instead of peer B who is a free rider

28

Therefore, by further analyzing and modifying the suggested mechanism always according to the principles of the model we come to the conclusion that the choking mechanism is an implementation of the Economy of Giving model and thus directly related to P2P field. This mechanism is used by BitTorrent and the model as well with the difference that BitTorrent‟s choking is based on current contribution during a transaction while the model‟s choking is based on the peers‟ yield curves monitored by the tracker (or calculated by every peer in the initial design). So Economy of Giving‟s mechanism by combining the account balance, the yield curve calculation and the choking mechanism it can prevent freeriding more effectively. That mechanism works like reputation mechanism and it could not only solve the big population problem of the initial suggestion according to the model, but also it could leave out the initial risk every time a peer takes when it connects to a new peer. This would be done easily since when a peer enters a new swarm it gets the list of peers and their yield curves so it can immediate ly choke the freeriders in the first place.