Economics of BitTorrent Communities
Ian A. Kash John K. Lai
Harvard CRCS Harvard SEAS
Cambridge, MA 02138, USA Cambridge, MA 02138, USA
kash@seas.harvard.edu jklai@seas.harvard.edu
Haoqi Zhang Aviv Zohar
Harvard SEAS Microsoft Research SVC
Cambridge, MA 02138, USA 1065 La Avenida, Mountain
hq@eecs.harvard.edu View, CA 94043, USA
avivz@microsoft.com
ABSTRACT 1 Introduction
Over the years, private file-sharing communities built on the Interactions among large numbers of agents on the Internet
BitTorrent protocol have developed their own policies and challenge system designers to not only focus on system-level
mechanisms for motivating members to share content and function, but also to account for user incentives. In systems
contribute resources. By requiring members to upkeep a ranging from eBay to BitTorrent, the design of new rep-
minimum share ratio between uploads and downloads, pri- utation systems and sharing protocols has paid particular
vate communities effectively establish credit systems, and attention to the role of economics in computer systems. In
with them a full-fledged economy. In attempting to un- BitTorrent, incentivizing users to contribute by uploading
derstand the functioning of private communities, most pre- while downloading a file has lead to an effective form of file-
vious studies view such communities as computer systems sharing, that now accounts for an estimated 18% of Internet
and focus on their ability to efficiently distribute files. In traffic [1].
this paper, we advocate for adopting an alternative perspec- Despite BitTorrent’s success, the lack of an incentive for
tive where communities are viewed as economic systems in peers to continue uploading a file after it has been down-
which users adapt to site policies. We report on a year-long loaded has spurred the formation of private BitTorrent com-
measurement study of DIME—a community for sharing live munities. Private communities build upon the BitTorrent
concert recordings—through which we find that users ratio- protocol by developing their own policies and mechanisms
nally react to cost differentials when making decisions on for motivating members to share content and contribute re-
what files to consume. We observe significant disparities in sources. Communities tend to be organized around a par-
the cost of new and old files, where by consuming newer files ticular interest—live concert recordings, HD movies, or the
frequent visitors can quickly earn credit. We find that users newest TV shows—and registered members can acquire files
compensate for the high cost of older files by downloading of interest in return for sharing files with like-minded users.
more copies of newer files, and by preferentially consuming There are over 800 active private BitTorrent communities [14],
older files during a ‘free leech’ period. each enforcing its own set of rules that have been refined over
time to fit the community’s goals and needs.
Categories and Subject Descriptors Supported by additions to the original BitTorrent proto-
C.2.4 [Computer-Communication Networks]: Distributed col, private communities are able to track how much each
Systems; J.4 [Social and Behavioral Sciences]: Eco- user downloads and uploads. This allows them to require
nomics members to upload a certain fraction of the amount they
download. This regulation, known as share ratio enforce-
General Terms ment (SRE), effectively introduces costs and credits to the
system: a ratio requirement of 0.25 puts a cost of one byte
Economics, Measurement of upload for every four bytes of download. In managing
credit and accounting for consumption (download) and la-
Keywords bor (upload), private BitTorrent communities have evolved
BitTorrent, private communities, peer-to-peer, share ratio from computer systems into full-fledged economic systems.
enforcement Anecdotal evidence from discussions among members in
private communities points to a rich, multi-faceted set of
user behaviors that emerge in response to economic forces.
Their stories and shared advice suggest that users often
Permission to make digital or hard copies of all or part of this work for make economic decisions and trade-offs, e.g., by joining new
personal or classroom use is granted without fee provided that copies are torrents as a way to quickly earn credit that can then be
not made or distributed for profit or commercial advantage and that copies spent on downloading older torrents. As users react to eco-
bear this notice and the full citation on the first page. To copy otherwise, to nomic conditions within the system, understanding how a
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee. community functions and how to improve its design will nec-
Copyright 200X ACM X-XXXXX-XX-X/XX/XX ...$10.00. essarily require an understanding of its economy.
Previous studies of BitTorrent communities (e.g. [3, 10]) ology for obtaining measurements. Section 3 provides high
have typically emphasized their characteristics as computer level measurements on the state of the system. Section 4
systems, by focusing on aspects such as the arrival rate of reports on how users can acquire wealth in this economy, as
peers to a torrent, the quantity of resources available, and implied by the cost differentials between new and old files.
the performance experienced by users. A few recent works We examine how users react to such cost differentials in Sec-
have studied the economics of private communities by using tion 5, where we find that users preferentially download new
theoretical and simulation approaches to examine how ratio files except when old files are made cheaper. We discuss
enforcement incentivizes contributions, and how issues such the implications of our findings in Section 6, with a focus
as lack of credit flow [7] or potential for collusion [9] can cre- on improving the design of private BitTorrent communities.
ate inefficiencies and manipulation opportunities. While the Section 7 presents related work, and Section 8 concludes.
theoretical analysis and simulation results from these works 1.1 Overview of DIME
provide some insight, gaining a deeper understanding of the
economy in private BitTorrent communities will require rich DIME (www.dimeadozen.org) is a private BitTorrent com-
datasets from actual communities that capture how users munity in which users share live concert recordings (bootlegs)
adapt to economic pressures. in lossless audio format. Sharing concert recordings has a
In this paper, we advance the study of private communi- rich history dating back to before BitTorrent, where mu-
ties as economic systems by reporting on a year-long mea- sic enthusiasts would trade tape and CD recordings of their
surement study of the DIME community for sharing live con- favorite bands. DIME provides a community in which to
cert recordings. Using extensive traces of groups of torrents, continue this tradition of bootleg trading, but with the con-
daily snapshots of the credit and activity of all users, and venience afforded by its website, forum system, and BitTor-
snapshots of activity across all torrents, we find that users rent trackers. Shows uploaded on DIME cover a wide range
rationally react to cost differentials when making decisions of music genres, and include a good mix of new shows from
on what files to consume. Specifically, we find that: currently touring bands as well as older shows of interest.
DIME allows open registration, but restricts the maxi-
• there are significant differences between the cost of new
mum number of accounts to approximately 110,000 so as to
and old files,
reduce server load and work for moderators. While the site
• frequent visitors to the site have more opportunities to is typically full, new accounts open up frequently, as existing
earn credit, and accounts that have been inactive for long periods of time are
• users react to cost by downloading more copies of newer periodically removed from the system. The minimum share
files than older files, and by preferentially consuming ratio required from users is only 0.25, a figure that is low
older files during a ‘free leech’ period. compared to other private BitTorrent communities.
While new files provide ample opportunities for earning
upload credit by seeding to new leechers, the arrival of leech- 2 Methodology
ers quickly slows and competition for seeding grows, leaving
seeders of older files with few opportunities to defray the cost DIME’s servers collect information that is reported peri-
of the download. The lower cost of new files suggests that odically by the BitTorrent clients of its members, which it
frequent visitors to the site have more opportunities to earn tracks and displays in the form of HTML pages available to
credit, and more of them do achieve high ratios. We find that all members. We have obtained the following information
users compensate for this cost differential by downloading by performing periodic crawls of the website:2
newer files; while this could be explained by users’ prefer- • Account profile snapshots: We took periodic snap-
ence for more recent content, we find that users consumed shots of the profile pages of all user accounts in the sys-
older files during a “free leech” period where users received tem. These profile pages include static information such
credit for uploading but did not get charged for download- as the user’s join date and dynamically updated infor-
ing. This suggests that many users desire files they cannot mation such as the user’s ratio, upload amounts, and
normally afford, and that providing alternative earning op- download amounts. If a user happened to be online at
portunities may help meet some of the ‘unmet demand’.1 the time of the scrape, we obtain the user’s current ac-
To better understand the rationales for the system’s cur- tivity and upload and download rates. We performed
rent design, we also conducted an interview with DIME daily scrapes between April 28, 2010 and September 27,
moderators, that provides insights into their goals for the 2010, and multiple scrapes per day between December
site, rationale behind policy decisions, and views on how 23, 2010 and January 21, 2011.3
the system is functioning. Their comments suggest that the • Torrent traces: We recorded traces of torrent detail
site is built for robustness, with particular considerations pages from the time a torrent was posted for a num-
for forming a vibrant community in which users can down- ber of torrents. These pages include information about
load files of interest as long as they put in the effort to the seeders and leechers on the torrent and their current
contribute. In comparison, there is less concern about cost upload and download amounts for the torrent. We down-
differentials, where the moderators appear satisfied with the loaded the torrent details pages every five minutes for the
current mechanisms for users to acquire wealth. We pay torrents being tracked. Our first batch of traces tracked
particular attention to the moderators’ comments when dis- 173 torrents posted after April 29, 2010 until June 26,
cussing possible ways to improve the system.
2
The paper proceeds as follows. Section 1.1 introduces Our study is conducted with permission from DIME mod-
the DIME community, and Section 2 describes our method- erators, and with approval from Harvard’s Institutional Re-
view Board.
1 3
By unmet demand, we mean files that people want but Out of 153 possible days between 4/28/10 and 9/27/10, we
cannot afford. are missing 32 days due to scrape failures.
12000
10000
Number of users
8000
6000
4000
2000
0
0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3+
Ratio
Figure 1: Snapshot showing all users’ upload and Figure 2: A Histogram of user ratios from a snap-
download amounts. Users marked in green have do- shot taken on May 1st, 2010 with bin sizes of 0.025.
nated money to the site; users marked in red (in- the snapshot, of whom 7.4% have donated money to the
cluding those covered by the thick green) did not. site. Note that almost all users who have downloaded more
2010. Our second batch of traces tracked 176 torrents than 10GB that are still in the system have a ratio above
posted after June 27, 2010 until September 7, 2010. 0.25 and that many users have a ratio above 1. This shows
that there are many ‘altruistic’ users in the system who are
• Torrent snapshots: In late 2010, we also started to
willing to perform large amounts of upload while consuming
perform snapshots of all active torrents in the system.
less download.
These snapshots crawled the same pages as the torrent
Figure 2 provides a histogram of the share ratios of users
traces, but we did not track individual torrents and in-
in the system. The figure shows distinct increases around
stead took less frequent snapshots of all torrents.
ratios 0.25 and 1. While the spike at 0.25 is likely due to
While most of the statistics we collect are precise, two the effect of ratio enforcement, the bump around 1 seems to
require some care. One is the maximum upload bandwidth suggest that some users may be ‘target earners.’ In partic-
available to a user. This is not easy to measure, even under ular, earning a ratio of 1 implies giving back just as much
optimal conditions, as it depends on many changing factors as you have taken, which may be what some users are after.
such as the load on a user’s Internet service provider and The median user (among those who have ever downloaded
caps set within a user’s client software. Thus, the actual data) has a ratio of 0.51, and about 28% have a ratio that
maximum bandwidth a user can supply varies over time. exceeds 1.0.
DIME tracks the maximum upload bandwidth it has ever
observed for a user. While at an individual level the re- 3.2 Enforcement Cycles
ported value may be a noisy signal of how much bandwidth DIME does not constantly enforce its ratio, but rather does
a user can typically provide, in aggregate our results suggest so in share ratio enforcement (SRE) cycles, which define par-
it provides a reasonable signal. For example, on average the ticular download amounts after which the system will check
upload rate of a peer is roughly linear in it (see Figure 5(a)). a user’s ratio against the minimum requirement. Users who
The other statistic is the current upload of a peer when do not meet the requirement at that point in time are only
tracking a torrent. We did not perform peer-level measure- allowed to seed torrents they have already completed. The
ments, so we only have access to the data that peers reported limitation is lifted if the user’s ratio recovers to the required
to the tracker. Though we crawl each tracked torrent every minimum. One can view the enforcement cycles as a form of
five minutes, empirically we observe that a peer’s reported loaning, wherein the system allows users (particularly new
upload updates every 20 to 30 minutes. As a result, we can users with no credit) to download for some time before be-
derive upper and lower bounds on the peer’s upload during ing required to adhere to the ratio requirement. SRE cycles
these 20 to 30 minute intervals, but we do not have finer occur at 5GB intervals for non-donating users, and at longer
grained information. When we compute statistics such as intervals for donating users.
upload rates, we assume the upload is distributed equally Figure 3 shows the effect of the SRE cycle as it is triggered
across these intervals, and we make sure to aggregate data at 5GB for low ratio users. Note that most users who become
from many users to mitigate errors due to this assumption. stuck at the enforcement point have uploaded very little, and
In addition to data collected from the site, we conducted the enforcement cycle effectively removes them.
an interview with DIME moderators, that provides insights Looking back at the user snapshot in Figure 1, we see
into their goals for the site, rationale behind policy decisions, that most users who have uploaded at least 1GB are above
and views on how the system is functioning. A full transcript the 0.25 threshold, and virtually all non-donating users who
of the interview is included in the appendix. have uploaded at least 10GB are above the threshold. This
suggests that users who are restricted by the first cycle are
3 The DIME System often free riders that do not intend to become long term
In this section we survey a number of the system level char- members of the community. Such users are initially given
acteristics of DIME. the benefit of the doubt, but the extent to which they can
free ride is limited.
3.1 Contribution and Consumption While donating to DIME does not increase a user’s ratio,
We begin by examining the amounts that users contribute Figure 1 shows that there are a number of users with 100GB
and consume. Figure 1 shows a snapshot of the historical or more of download who are significantly below the ratio
upload and download amounts of all users on February 20th, of 0.25. Presumably, these users make repeated donations
2010. There are 109,891 users in the system at the time of to periodically extend their SRE cycles. From the system’s
2500 1500 1
Fraction of Leeched Torrents
2000 0.8
Number of Peers
Number of Peers
1000
1500 0.6
1000 0.4
500
500 0.2
0 0 0
0 10 20 30 0 10 20 30 0 10 20 30 40 50
Number of Torrents Seeded Number of Torrents Leeched Seeder Leecher Ratio
(a) Seeding (b) Leeching (c) CDF of seeder/leecher ratios in a
single snapshot.
Figure 4: Participation by Seeders and Leechers
significant increase in Figure 4(a) at 10. Since users who
have not donated are only permitted to seed 10 files at a
time, the users seeding 10 torrents at a time are effectively
seeding at capacity.4
A more careful look into the data shows that some seeders
are seeding torrents that are not being downloaded. For
torrents that are actively being downloaded by at least one
user, we see from Figure 4(c) that the median torrent has
only 2 seeders for each leecher, suggesting that download
rates in many torrents may be limited by the amount of
upload bandwidth available.
Figure 3: Closeup of the effect of the share ratio 3.4 User Turnover and Free-riding
enforcement cycle at 5GB.
The ratio enforcement mechanism was put in place to pre-
perspective, this is not particularly bad: there aren’t enough vent free-riding, but selfish users can still take advantage
of these users to cause a problem with the functioning of the of DIME’s free registration policy and open new accounts.5
economy, and donations collected from them can be used to On May 1st, 2010, DIME had a total of 110,190 registered
cover server and other costs. users. By August 1st, a significant number of these accounts,
18.7%, were closed due to inactivity and were made avail-
3.3 Seeder / Leecher Ratio
able to new users. As might be expected, turnover is more
A commonly cited advantage of private communities is that, common for low ratio accounts. 29% of accounts that had
while in public torrents leechers typically outnumber seed- ratio below 0.25 were removed during this period. However,
ers, in private communities all but the newest torrents typ- turnover is still the norm even among significant contribu-
ically have many more seeders than leechers [10]. This gen- tors, as 11% of all accounts with a ratio greater than 10.0
eral observation is true of DIME as well; on May 1st, 2010, were also removed.
the seeder to leecher ratio averaged across the entire system We quantify the amount of free-riding on DIME by ex-
is over 15. amining the impact of turnover on the system. If many
From the perspective of available system resources, the users were whitewashing often and managing to extract large
ratio statistics above may be misleading as a single peer amounts of data, we would expect to see that users who re-
can simultaneously be a seeder and a leecher on multiple main in the system do more upload than download. On May
torrents, in which case his bandwidth will be split between 1st, 2010, the total amount of upload credited to users on
them. To obtain a more informative measurement, we count DIME is 29.6 petabytes, and the total amount of download
the number of distinct seeders and leechers in the system, is 27.1 petabytes.6 The difference is about 10%, which in-
and find that when measured this way the seeder leecher ra- dicates a significant amount of free-riding, but an amount
tio is much less dramatic. In the same snapshot from May that is unlikely to be problematic for the system.
1st, 2010, there were a total of 10,957 users online. Of these
users, 10,488 were seeding at least one file, and 2,444 peers 4 The DIME Economy
were leeching at least one file (some did both), leading to Having provided measurements on the state of the system,
a peer-adjusted seeder leecher ratio of just 4.29. The dra- we turn to investigate DIME’s economy. We first discuss
matic disparity in measurement is due to the disparity in how users make money in this system. In previous work [3],
the average number of files a user seeds or leeches within
4
the snapshot: the average seeder was seeding 6.85 torrents, A small number of very high bandwidth users are allowed
while the average leecher was downloading 1.93 torrents. to seed more than ten files.
5
Figure 4 shows statistics about seeding and leeching be- The user may not be able to do this repeatedly, however,
as the system implements checks for such abuse.
havior. While a tiny fraction of users seed as many as 403 6
These numbers are not the overall amounts of upload and
files simultaneously and leech as many as 61, the median download that has ever occurred on DIME, as deleted ac-
leecher is only in 1 torrent and the median seeder is in 4 counts also remove record of both download and upload from
torrents. While Figure 4(b) decreases smoothly, there is a the system.
120
0-12hrs
25 12-24hrs
25 100 24-36hrs
36-48hrs
>48hrs
Upload rate (MB/hr)
20
20
Upload rate (MB/hr)
Upload rate (MB/hr)
80
15 15
60
10 10
40
5
5
20
0
0 0 5 10 15 20 25 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
Left end point of max bandwidth bucket (kBps) Lifetime of torrent (days) Left end point of max bandwidth bucket (kBps)
(a) Upload rate vs Max bandwidth. (b) Upload rate vs. Torrent age. (c) Upload rate vs. Max bandwidth for
different torrent age buckets.
Figure 5: Effects on Upload Rate.
seeding longer and having higher upload bandwidth have of torrents dying and our measurements recording a rate of
been noted as major contributors to the wealth of individ- 0 for these torrents that are inactive. The large discrepancy
uals. Over the first batch of torrents we tracked on DIME, between the returns from seeding early versus seeding late
we also saw a correlation between bandwidth and upload show that timing may have a more significant impact on the
rates over the first two months of the torrents’ life cycle (see wealth a user can acquire from a file, more than the amount
Figure 5(a)). But while the figure suggests that bandwidth of time spent seeding.
is correlated with upload rates, there are a number of possi- Joining a torrent early allows users to earn more upload.
ble confounding factors. Most notably, we find that upload However, for users who join equally early, it is still the
rates can vary significantly over the age of a torrent, and case that high bandwidth users can earn more upload. Fig-
our results show that joining a torrent early is an extremely ure 5(c) plots the upload rate for users in different band-
effective way to make money, regardless of a user’s upload width buckets during the early lifetime of a torrent. Higher
bandwidth. We detail these findings below, and then report bandwidth leads to higher upload rates, as expected, while
on the implications on wealth. earlier join times magnify this effect through changing the
slope of the plotted relationship. Figure 5(c) also suggests
4.1 Earning money: joining early
that an effective way to compensate for connection speed is
Arriving early in the life of a torrent can potentially be ben- to join torrents earlier. For example, while we refrain from
eficial because the arrival of new leechers slows over time, re- giving precise numbers due to measurement noise, the figure
ducing the opportunities for seeding. However, the effect on suggests that joining in the first 0-12 hours as a low band-
upload is not immediate from this argument, because there width user (50-150 kbps) may yield higher upload rates than
also tend to be more seeders early in the life of a torrent, sug- joining in the first 12-24 hours as a higher bandwidth user
gesting that users arriving early may face more competition (150-250 kbps).
with other users for upload while facing increased demand. A consequence of the effect of joining early is that there
By tracking the activity on individual torrents on DIME, we is a higher cost associated with downloading older torrents.
find that arriving early is a significant way to make money For each unit of time spent seeding, a user can gain back
in the DIME economy, and returns to seeding decrease dra- more upload on a new torrent than on an old torrent, effec-
matically with the age of a torrent. tively making new files cheaper than old files. We have seen
In Figure 5(b), we use the first batch of torrent traces that bandwidth also plays a role as expected; for a specific
(173 torrents) to obtain an aggregate estimate of the upload file and join time, having higher bandwidth also decreases
per period of time seeding over the age of the torrent. For the cost to download the file.
each torrent, we compute an estimate of the upload rate as
4.2 Implications on wealth: visiting often
follows. For each seeder (other than the original uploader of
the file) on the torrent, we construct a sequence of (upload, The cost differential between new and old files have signifi-
(start time, end time)) pairs which gives an estimate of how cant implications on the ability of users to acquire wealth in
much the user uploaded in (start time, end time). We then the system, both in terms of absolute wealth and the rate of
bucket these observations by time (5 hours), so that for each earnings. Absolute wealth refers to a user’s ability to con-
bucket, we have the total upload as well as the total time sume, wherein for two users with the same share ratio, the
spent seeding. From here, we divide total upload by total user with more total upload and download has more credit
time to get an estimate of the upload rate in the time bucket. to spend on downloads before his ratio drops to a particular
We then take the average of these upload rates across all level (e.g., 0.25). Rate of earnings refers to a user’s ability to
torrents in our set of traces. Torrents that had no seeding achieve a particular upload to download ratio given specified
activity in a time bucket are included with a rate of 0. conditions, e.g., we expect that high bandwidth users can
Figure 5(b) shows that the average upload rate on a tor- achieve higher rate of earnings than low bandwidth users,
rent is extremely high in the hours immediately following its all else being equal. In this sense, one can think about ab-
posting, and that there is a severe drop in rate of return over solute wealth as spending potential, and rate of earnings as
the course of the first few days. After five days, the decrease earning potential.
in rate of return slows down, but continues for the lifetime of Since newer torrents provide high earning opportunities,
the torrent. The slow decline in the tail may be an artifact we expect users who can regularly arrive on torrents early
160
Average amounts (in GB)
1
Upload
> 1.0 140 Download
0.8 0.5-1.0 120
Ratio distribution 0.1-0.5
0-0.1 100
0.6
80
60
0.4
40
0.2
20
0
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
0
1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 Frequency of visits (fraction of days)
Frequency of visits (fraction of days)
Figure 8: Average download and upload amounts
Figure 6: Distribution of users by upload to down- (in GB), bucketed by visitation frequency.
load ratio, bucketed by visitation frequency.
ferences in the distribution of bandwidth among users at
different visitation frequencies, we plot in Figure 7 the ag-
2
0-40kBps gregate ratio achieved by users at different visitation fre-
40-80kBps
1.8
80-160kBps quencies, separated out by bandwidth. We see that on av-
160-320kBps erage, regardless of the bandwidth group, users who visit
1.6 >320kBps
the site most frequently earn higher ratios, with the low-
1.4
est bandwidth and highest bandwidth users earning signifi-
1.2 cantly higher ratios when they arrive earlier. This provides
Ratio
1
further evidence that the observed effect may be due to dif-
ferences in earning potential, as caused by being able to join
0.8
files earlier.
0.6 From Figure 8, we see that frequent visitors not only earn
0.4
higher ratios, but also download and upload more on average
than infrequent visitors. In addition to having higher in-
0.2
[0.9,1] [0.7-0.9] [0.5-0.7] [0.3-0.5] [0.1-0.3] period consumption, the proportionally higher upload imply
Frequency of visits (fraction of days) that frequent visitors are also acquiring significantly more
absolute wealth that can be utilized for future spending.
Figure 7: Aggregate ratio vs. Visitation frequency, While some of the difference in absolute wealth earned can
bucketed by upload bandwidth. be explained by higher rate of earning on newer files, it may
also be a representation of actual demand, wherein users
to have the best earning opportunities. Assuming that files with lower demand (e.g., interested only in a few particular
of interest to a user are uploaded more or less uniformly bands) may just choose to visit the site less frequently. It
across time, the likelihood that a file of interest will be new is also worth noting that while newer files provide higher
when a user sees it on DIME is directly correlated with how earning opportunities, visiting the site often is in itself costly.
frequently the user visits the site. To study to effect of visi- We examine how users make economic decisions in response
tation frequency on earnings, we use contiguous daily snap- to such trade-off in the next section.
shots of all DIME users from April 28th, 2010 to July 20th,
2010 to obtain records of each user’s upload and download 5 Responses to economic forces
amounts during this period, as well as the percentage of days To examine how users respond to the forces within the DIME
on which they were seen on the site. After filtering out users economy, we first consider the behavior of low bandwidth
who only visited once or never, and users for whom we did users, who are particularly susceptible to economic pressures
not have at least 60 days of data (e.g., new members) in this within the system. We then share our findings of how users
measurement period, we are left with a dataset containing react to cost differentials by preferentially downloading more
records for 38,583 users. copies of newer files unless older files were made cheaper.
We plot in Figure 6 the distribution of ratios obtained by
users with particular visitation frequencies. We see from the 5.1 Low Bandwidth Users
graph that very few of the most frequent visitors (2%) obtain We have seen that returns to seeding depend on bandwidth
a ratio of less than 0.1, and over 50% earn a ratio above 1. and torrent age. In particular, upload rates are significantly
On the contrary, nearly 40% of the most infrequent visitors higher if a user arrives early on a torrent and tend to in-
earn a ratio of less than 0.1, and only 20% earn a ratio crease linearly with bandwidth over a given period. This
above 1. While this suggests that frequent visitors may be suggests that, all else being equal, low bandwidth users will
presented with more earning opportunities, we also see that earn less credit than high bandwidth users. In this section,
not all frequent visitors earn a high ratio. This is likely due we examine four actions low bandwidth users could take to
to choice: as users who visit often can more easily join newer compensate for this: join torrents earlier, seed longer, down-
torrents to earn credit, they can also respond to their higher load less, and aim for lower ratios.7
earning potential by consuming more older files, or seeding 7
for shorter periods of time (e.g., disconnect soon after the The amount of bandwidth a user has depends at least
partly on how much of his connection he chooses to con-
download completes). tribute. However, previous work has shown that this is quite
To ensure that the observed difference is not due to dif- inelastic [3].
Average Download/Upload (in GB)
120
1
0-40kBps Download
40-80kBps 100 Upload
0.8
Fraction of users
80-160kBps
160-320kBps 80
0.6 >320kBps
60
0.4 40
0.2 20
0
0 0-40kBps 40-80kBps 80-160kBps 160-320kBps >320kBps
10-2 10-1 100 101 102 103 104 105 Frequency of visits (fraction of days)
Age of torrent (hours)
(a) Figure 10: Average download and upload amounts
1 of users bucketed by bandwidth.
0.9
0.8 0−40kBps respond to costs by simply uploading less in proportion to
Fraction of users
40−80kBps
0.7 80−160kBps their download, thus earning lower ratios. Users in the low-
160−320kBps
0.6
>320kBps
est bandwidth bucket earned an aggregate ratio of 0.4 during
0.5
0.4
this measurement period, while users in the highest band-
0.3
width bucket earned an aggregate ratio of 1.7. Since DIME
0.2 only requires users to maintain a ratio of 0.25, aiming for
0.1 lower ratios seem like a reasonable strategy for low band-
0
10
−1
10
0 1
10 10
2 3
10 width users acting in this economy. That said, their lower
Time spent seeding (hours) upload volume and earned ratio imply that they acquire sig-
(b) nificantly less absolute wealth than high bandwidth users,
thus leaving them with less savings for future consumption.
Figure 9: (a): CDF of time of first appearance on This in some sense also limits their ability to strategically
torrents by users’ upload bandwidth. (b): CDF of join early: while the average new torrent provides signifi-
time spent seeding by users’ upload bandwidth. cant earning opportunities, any particular new torrent may
not. Unlike high bandwidth users with higher savings, low
Our results show that low bandwidth users do not join bandwidth users may be less able to account for uncertainty
torrents earlier than other groups. Figure 9(a) shows, on in earnings while attempting to maintain a ratio above 0.25.
a log scale, the CDF of the time after a torrent’s creation In summary, we find that low bandwidth users do not
at which users arrive. Each line represents a class of users compensate for their bandwidth by joining torrents earlier,
within a particular bandwidth bucket. We see that most of but they do seed for longer, download less, and earn lower
the lines are quite similar, with low bandwidth users join- ratios. If we assume that low bandwidth users have the same
ing slightly later than high bandwidth users. This suggests demand as users in other buckets, the combination of seed-
that many such users are unable or unwilling to change their ing for longer yet downloading less suggests that the increase
behavior in order to join torrents earlier. One possible ex- in seeding is unable to fully compensate for the low band-
planation is that, while checking the site more frequently can width of the users. This result agrees with the observation
allow users to join desired torrents earlier, doing so requires by Andrade et al. [3], that while users were generally willing
manual effort and may be costly or infeasible for many users. to seed longer, doing so did not seem to make them much
While low bandwidth users do not appear to join tor- more successful as uploaders. The insights from Section 4.1
rents earlier than other users, we do find that low band- provide an explanation: as the demand for a file drops sig-
width users seed longer than high bandwidth users. Fig- nificantly over time, much of the longer period of time spent
ure 9(b) shows, on a log scale, the CDF of seeding time, seeding will be when there are fewer opportunities for up-
again grouped by bandwidth. Here the ordering of lines is loading.
consistent, with higher bandwidth users spending less time
seeding than lower bandwidth users. For example, the me- 5.2 Unmet Demand and Excess Supply
dian user with a bandwidth between 0-40kBps spends 1.4 While low bandwidth users are particular susceptible to eco-
times as long seeding as the median user with a bandwidth nomic pressures within DIME, all but the very wealthy users
between 80-160kBps, who in turn spend nearly 1.5 times as may need to make economic trade-offs in terms of what files
long seeding as the users in the highest bandwidth buckets. to download based on what they can afford. Given the dif-
A third way to compensate for low bandwidth is for users ference in costs between new and old files, we expect users
to download less. Our findings from observing the change to preferentially download newer files. Furthermore, if files
in users’ download amounts between April 28th, 2010 and were made cheaper, we would expect downloads of older files
July 20th, 2010 show that this is indeed the case. Figure 10 to increase. Based on a natural experiment that occurred
buckets users by bandwidth, and shows that on average low during our study, we are able to test this prediction.
bandwidth users download less than high bandwidth users. From December 23, 2010 to December 26, 2010, DIME
While low bandwidth users download less, there is no obvi- had a “free leech” period, during which downloading did not
ous reason the should want less. The difference in download count against a user but uploading still provided credit. Fig-
amounts thus suggests that the cost of files in the system ure 11 shows the number of active downloads during a three
may be preventing lower bandwidth users from being able week period including the free leech. We observe signifi-
to fulfill their demand. cantly more active downloads during the free leech period
We also see from Figure 10 that low bandwidth users also than during the days before and after free leech, where the
total their bandwidth distribution) is more or less the same dur-
Number of Active Downloads
8000
new ing free leech and at other points in time, this observation
old
6000 very old
suggests that there is an excess supply of available upload
bandwidth among active seeders that is not utilized except
4000 during free leech. While previous work by Andrade et al. [3]
suggest that approximately 75% of torrents are constrained
2000 by upload bandwidth, due to their methodology the result
Free only applies to torrents that currently have multiple leech-
Leech
0 ers (e.g., newer files). For the long tail of older files on
12/19 12/26 01/02 01/09
which there is only one leecher or none at all, our results
suggest that seeders of these files have excess upload band-
Figure 11: Leeching activity before, during, and af- width, but due to the files’ high cost there is typically not
ter a free leech period. enough demand. While this finding points to a potential
inefficiency of the system, having a supply of seeders with
amount of download activity during free leech is 50 to 75% available bandwidth on older torrents does allow these files
higher than during the days following free leech.8 to remain available to users who choose to download them.
In the days before and after free leech, we observe that the
number of active downloads of files uploaded within the last 6 Discussion
week (new files) is nearly identical to the number of active Our ultimate goal is not only to understand the economic
downloads of files older than a week (old files). However, factors driven by DIME’s current system, but also to find
there are approximately 25 times more old files than new ways to improve the site’s function. While one can establish
files at any given time. Thus, users are downloading signif- a number of desirable properties, understanding which im-
icantly more copies of newer files than older files. Some of provements to focus on and what tradeoffs to make depend
this difference may be because old files have already been on the particular goals of the community. While decisions
downloaded by many users. However, we see in Figure 11 can sometimes be made with respect to general user pref-
that while there is no noticeable increase in demand for new erence (e.g., trading off higher download speeds with the
files during the free leech period, demand for old files in- amount of seeding required), most decisions on resource dis-
creased 60 to 70%. Moreover, the demand for files more tribution and availability will force us to weigh the conflict-
than sixty days old (very old) nearly doubled during free ing preferences of different users, and to draw on the political
leech. From an economic perspective, the change in the cost and moral viewpoints of the community while doing so (e.g.,
of new files during free leech is small, because such files al- what kind of users are wanted in the system, who should get
ready provide significant earning opportunities. But for old to consume more, and what kind of files should be allowed).
files that are normally quite expensive, free leech provides
We first discuss our desiderata as motivated by our in-
a significant opportunity to acquire these files for free. The
terview with DIME’s moderators, and then discuss specific
behavior appears to be a rational economic response, where
changes with potential for improving DIME.
files with the largest price discounts received the greatest
additional interest. We also examined whether larger files 6.1 Desiderata
were preferentially downloaded, but the increase appeared
From our interview with DIME moderators, we find that
relatively uniform across file sizes (not shown).
decisions on site policies are often community-driven. For
DIME’s moderators describe the goal of free leech as help-
example, DIME’s ratio requirement of 0.25 is the result of
ing “those who struggle hard but slipped a bit [below the]
a vote among community members in 2004.9 The choice
minimum SR requirement.” By observing the ratios of users
of this relatively low sharing ratio indicates that DIME is
at the start and end of free leech, we found that 937 of 13,345
more open than other communities to the ‘less fortunate’,
users with low ratios between 0.2 and 0.4 increased their ra-
who may have slower Internet connections or cannot visit the
tio by at least 0.01, and 214 of such users by at least 0.05.
site frequently. The fact that every user is granted 5GBs of
This suggests that while not all of the intended users took
download before ratio enforcement begins, strikes a balance
advantage of free leech to earn higher ratios (e.g., some may
between giving new users a leg up, and allowing in some
not have been on the site during the period), a fraction of
hit-and-run leechers to download for free.10 DIME moder-
them did receive a boost.
ators view themselves as a “user-help-desk”, and are willing
While there was no particular bonus for seeding during to help users in share ratio violation to get back on track
free leech, the increase in download activity allows a seeder by providing advice and temporarily extending the enforce-
to earn more credit per hour spent seeding. Interestingly, ment cycle for users with less than 10GB downloaded. One
there was essentially no increase in the number of seeders desideratum may thus be to provide additional mechanisms
during free leech, either overall or among those with low for helping ‘less unfortunate’ users and poor decision-makers
share ratios. Given the increase in the number of active to earn credit, while keeping abusers out of the system.
downloads during free leech, more downloads are supported
by the same number of seeders during this period. Assum- 9
This ratio requirement has not been changed since; there
ing that the characteristics of the population of seeders (e.g., were a number of attempts spurred by particular commu-
nity members, but they “weren’t successful because of the
8 community’s resistance.”
Note that prior to the free leech period our data has only a
10
single observation each day, while during and after free leech The moderators noted that the enforcement cycle may need
we have multiple observations per day. The results during to be extended to account for the increase in file size over the
the latter period captures some of the daily fluctuations in years, but were concerned that “hit-and-run leechers would
usage that are typical of private communities [6]. be able to take even more without giving.”
1
0.9
0.8
0−14 days on site
14−50 days on site
Increasing demand for files
One of the main issues that emerges from our study of
Fraction of users
0.7 >50 days on site
0.6 DIME is that many seeders spend time waiting for leechers
0.5 to appear on their torrents, and do not fully utilize their
0.4
bandwidth. At the same time, there are users who may be
0.3
interested in the files seeded, but choose not to download
0.2
them due to the high cost. One possible solution is to lower
0.1
0
the minimum required ratio, so that users with less ability
−1 0 1 2 3
10 10 10 10 10 to earn credit can effectively spend more, thus increasing
Age of torrent when leecher joins (hours)
demand. However, this may lead to ‘altruistic’ users having
Figure 12: A CDF of times leechers begin to down- to contribute more, and may encouage further free-riding.
load torrents. Users are grouped by the age of their In conventional markets, the price of services that have too
account. much supply and too little demand naturally drops. But on
DIME, all transfers are credited equally, so prices remain
When we approached moderators about the significant dif-
fixed. One can imagine adopting a credit system in which
ference in earning opportunities between new and old files,
uploads and downloads convert to credit based on the prices
they simply responded that “this is the nature of Bittor-
of files. In such a system, one can attempt to adjust the price
rent,” and that it encourages users to arrive which helps to
of currently expensive torrents, by slowly lowering the price
share and distribute files. The moderators also had no is-
over time, by making all files beyond a certain age cheaper,
sues with users visiting the site more often and downloading
or by making the price depend on the seeder to leecher ratio
newer torrents as a means to earn credit, and believe that
in the torrent. This would attract more reluctant download-
this helps DIME be “a vital community.” While forming an
ers, and hint to seeders that they should direct their efforts
active community in which users can download files of inter-
there. A related alternative is to have different prices for
est as long as they put in the effort to contribute (e.g., by
upload and download (similarly to free leech) where down-
joining new torrents regardless of interest) are important,
loaders are deducted credit at a lower rate per byte than
alternative mechanisms for rewarding contributions of users
uploaders are given credit. Related approaches to helping
can potentially offset the unmet demand due to the lack of
match supply and demand across torrents have been consid-
earning potential on older files. A second desideratum is
ered in Antfarm [11] and PACE [4].
thus to increase demand for older files, while balancing the
Price alterations should be done very carefully. If the cost
goal of maintaining a vibrant community.
decreases too much, too many users will wait to download
While system performance is important to DIME moder-
files and too few will seed them, thus realizing the mod-
ators, they look to reduce server costs when possible. More-
erators’ concerns about maintaining an active community.
over, moderators do get very busy, and given that they serve
If adopting asymmetric prices between seeders and leechers,
as a useful resource in the community, policies and mecha-
we need to be careful not to make it too easy to earn money.
nisms to reduce their workload can benefit the system.
Theoretical models [8] have shown that if it is too easy to
6.2 Potential Changes earn money, rational users feel “rich” and decrease their will-
Restricting access of new users to older files ingness to work, leading to a vicious cycle where fewer and
fewer users contribute.
Our measurements surprisingly found that that new users
have an increased tendency to download old files. Figure 12 Other methods from the economic “bag of tricks”
shows that the median user who registered within the last Economic mechanisms and incentives (credit-based or so-
0-14 days initiated download 96 hours into the torrent’s life- cial) can be useful for addressing many other issues in DIME’s
time, while the median veteran user (who had an account for economy. Below are some ideas:
more than 50 days) tended to join a torrent after only 11.3
hours. This effect may have several causes. First, users who • Distribution of wealth through progressive taxation: many
have just joined may find old files appealing (and they were users on DIME amass large amounts of credits they will
not around to download them when they were new). Second, never use, while others struggle to meet the ratio. Can
these users may be less aware of the pitfalls of downloading a redistribution of wealth help stimulate the economy?
old files (which may get them into debt very quickly). Fi- • Should we provide incentives to add new files to the
nally, some users may be joining the site to get a particular system? Currently, people who post new torrents are
file, and may not be interested in staying for the long run. granted respect, and the benefit of uploading them from
Newcomers who unwittingly end up in debt quickly may the very start – a humble reward considering the effort
be driven to create new accounts, or may turn to modera- involved in obtaining new recordings. Can additional
tors for a temporary extension of the ratio enforcement cy- incentives push people to publish even more files?
cle. An approach requiring less manual intervention would • Similarly, can we reward other forms of contributions by
be to limit the access of users to older files, e.g., until af- providing additional mechanisms for earning credit, e.g.,
ter they pass the first ratio enforcement cycle, or only when for reseeding a requested file?
they have enough absolute wealth to cover the entire cost of • Can we replace some of DIME’s ‘rigid’ rules (e.g., the
downloading the old file. This would both help new users number of torrents that are allowed to be seeded) with
avoid the potential mistake of getting into debt for down- economically-inspired rules that allow users to dynami-
loading a file they cannot later upload, and at the same time cally account for the system’s goals? For example, one
will also make free-riding less appealing as more effort will can place a tax on valuable resources such as server costs,
be needed to access many files. and have users ‘internalize’ this externality when making
seeding decisions that may affect server load. 8 Conclusion
7 Related Work We have presented an extensive study of DIME’s complex
economy. In it, we have shown that users react to economic
In a series of papers, Andrade et al. [2, 13, 3] studied traces
forces in numerous ways. While our focus is in understand-
seven BitTorrent communities, some of which use SRE. They
ing and improving DIME, it is important to note that even
found that peers contribute significantly more, particularly
without further intervention, DIME’s very survival in spite
by seeding for longer periods of time, in communities that
of changing conditions, such as increases in bandwidth and
use SRE. They also study the arrival rate of peers to tor-
file size and having a dynamic user population, is a tribute
rents, showing that it is initially high, but rapidly drops and
to both the community spirit that it maintains, and to the
then has a long, slowly-decaying tail. This arrival pattern
robustness of its economy. Over the years very few changes
fits with our observation that the greatest opportunities to
to its rules had to be made. That said, we believe further im-
gain upload as a seeder are early in the life of a torrent.
provements are possible, and have began to suggest potential
Hales et al. [7] report some basic statistics from a seven beneficial changes that can be implemented given the exist-
day trace of a community using SRE at a ratio of 0.67. ing infrastructure. Future work should aim to study other
They show that a majority of the uploading each day is aspects of DIME’s economy, and may make use of controlled
contributed by ten percent of peers, possibly starving oth- experiments via direct interventions.
ers of the opportunity to maintain an acceptable ratio while
downloading desired files. Using a theoretical model and
simulations, they demonstrate conditions under which this 9 References
occurs. Rahman et al. [12] build on this through additional
modeling and simulations and show how an adaptive pol-
[1] An estimate of infringing use of the internet. Technical report,
icy can help avoid credit crunches by instituting free leech Envisional, January 2011. http://documents.envisional.com/
periods when many peers are “stuck” at a low ratio. docs/Envisional-Internet_Usage-Jan2011.pdf.
Meulpolder et al. [10] studied five communities, three of [2] N. Andrade, M. Mowbray, A. Lima, G. Wagner, and
M. Ripeanu. Influences on cooperation in BitTorrent
which used SRE. In addition to community level data from communities. In Proc. of the ACM SIGCOMM Third
parsing reports from the community webpage, they also used Workshop on the Economics of Peer-to-Peer Systems
an instrumented client to gather data directly from peers. (P2PEcon), pages 111–115, 2005.
They found that more stringent ratio requirements lead to [3] N. Andrade, E. Santos-Neto, F. V. Brasileiro, and M. Ripeanu.
Resource demand and supply in BitTorrent content-sharing
higher download speeds, longer seeding time, and fewer un- communities. Computer Networks, 53(4):515–527, 2009.
connectable peers (those behind a NAT or firewall). [4] C. Aperjis, M. J. Freedman, and R. Johari. Peer-assisted
Zhang et al. [14] study the broad landscape of private content distribution with prices. In 2008 ACM Conference on
Emerging Network Experiment and Technology (CoNEXT
BitTorrent communities and estimated that over 800 pri- 2008), page 17, 2008.
vate communities combine to have approximately the same [5] X. Chen, X. Chu, and J. Liu. Unveiling popularity of
number of torrents as publicly available trackers and have BitTorrent darknets. In Proc. of the Global Communications
significantly more active users at any time. They performed Conf. (GLOBECOM), pages 1–5, 2010.
[6] X. Chen, Y. Jiang, and X. Chu. Measurements, analysis and
a two month trace of the identities of peers in four private modeling of private trackers. In Proc. of the IEEE Tenth Int.
communities and provide detailed analysis based on a single Conf. on Peer-to-Peer Computing (P2P), pages 1–10, 2010.
snapshot of a community that uses a ratio of 0.7. As this [7] D. Hales, R. Rahman, B. Zhang, M. Meulpolder, and J. A.
site has a policy of promptly removing users with low ratios Pouwelse. BitTorrent or BitCrunch: Evidence of a credit
squeeze in BitTorrent? In Proc. of the 18th IEEE Int.
rather than only after an extended period of inactivity, ap- Workshops on Enabling Technologies: Infrastructures for
proximately 90% of users have a ratio higher than 1. The Collaborative Enterprises (WETICE), pages 99–104, 2009.
users who are able to meet these stringent requirements are [8] I. A. Kash, E. J. Friedman, and J. Y. Halpern. Optimizing
scrip systems: Efficiency, crashes, hoarders and altruists. In
those who are quite active, with 50% online within 10 hours Eighth ACM Conference on Electronic Commerce (EC),
previous to their crawl and 95% online within 100 hours. pages 305–315, 2007.
This matches our observation that the most active users are [9] Z. Liu, P. Dhungel, D. Wu, C. Zhang, and K. W. Ross.
the most able to be big contributors. In a companion paper, Understanding and improving ratio incentives in private
communities. In Proc. of the 30th Int. Conf. on Distributed
Liu et al. [9] provide additional analysis of this snapshot. Computing Systems (ICDCS), pages 610–621, 2010.
They also provide a game theoretic model of how a sharing [10] M. Meulpolder, L. D’Acunto, M. Capota, W. Wojciechowski,
ratio affects user behavior in a single time period. However, J. A. Pouwelse, D. H. J. Epema, and H. J. Sips. Public and
this model essentially assumes the site has a single torrent private BitTorrent communities: A measurement study. In
Proc. of the 9th Int. Workshop on Peer-to-Peer Systems
in each period and does not capture any of the temporal (IPTPS), 2010.
effects we have discussed. They examine mechanisms that [11] R. S. Peterson and E. G. Sirer. Antfarm: Efficient content
can be used to prevent cheating by peers who falsely report distribution with managed swarms. In Networked Systems
Design and Implementation (NSDI), 2009.
their upload and download totals to the tracker or collude to
[12] R. Rahman, D. Hales, T. Vinko, J. A. Pouwelse, and H. J. Sips.
pretend to have performed uploads to each other that never No more crash or crunch: Sustainable credit dynamics in a p2p
occurred. community. In Proc. of the 2010 Int. Conf. on High
In a pair of papers, Chen et al. [6, 5] study 17 communities, Performance Computing & Simulation (HPCS), pages
332–340, 2010.
including a 68 day trace of DIME, and note that those that [13] M. Ripeanu, M. Mowbray, N. Andrade, and A. Lima. Gifting
use SRE have significantly greater user activity and seeding. technologies: A BitTorrent case study. First Monday, 11(11),
They comment on and model the tendency of peers to be 2006.
starved for opportunities to upload and discuss mechanisms [14] C. Zhang, P. Dhungel, D. Wu, Z. Liu, and K. W. Ross.
BitTorrent darknets. In Proc. 29th IEEE Int. Conf. on
such as free leech periods that communities use to ameliorate Computer Communications (INFOCOM), pages 1460–1468,
the problem. 2010.
APPENDIX some people to visit the site more frequently. Do you see
any problems with this, or do you see it as a good thing?
A Questions for DIME moderators 10. Sometimes, users download and seed torrents they are
DIME’s history, and goals not interested in to earn credit. Do you have a sense of how
1. We saw that DIME just had it’s 7th birthday – con- often this happens? This seems ‘wasteful’ from an economic
grats! Can you tell us a little bit about the history over perspective in that the user may not get a lot of value out
the years? Were there any major policy changes that took of the file, and if there were some other way to reward him
place? for effort on files he is interested in, it would improve the
2. The community seems very active on the mailing list, system’s efficiency. What are your thoughts on this?
and there are a number of polls on questions regarding site 11. Are there other known tips / strategies for raising
policies. Were there any particular things implemented based one’s ratio (other than the obvious of spending more time
on the community’s suggestions that you think are particu- online)? For instance, are there times of day during which it
larly interesting/useful in retrospect? is more valuable to be active? Are there certain categories
3. In running the DIME community, what are the mod- that are more beneficial to seed?
erators’ goals? Is there a sense in which the goal is to get 12. Have policies ever been considered that would give
the files to everyone as long as they contribute back a lit- additional rewards for seeding certain classes of torrents, for
tle bit, or is the motivation to get people to contribute more? example old or orphaned torrents?
13. Have any policies been considered for rewarding users
Questions about ratio enforcement for being connected as a seeder, even if they are not actively
4. While the site recommends users to maintain a ratio seeding?
of 1, the minimum ratio required is only 0.25. How did you
decide on 0.25 as the “right” ratio requirement for DIME? Questions about free leech
5. What do you view as some of the possible effects of 14. There was recently a “free leech day.” Why do you
changing the ratio requirement, whether up or down? What have them? How do you decide when to have them?
are some of the pros and cons?
6. How did the site decide on the length of the enforcement Questions about measurements
cycle? One thing we were intrigued by was that donating 15. When users are downloading a torrent, their overall
doesn’t actually improve one’s ratio — it just extends the up/dl speeds are reported, based on which a ’max ever’ is
enforcement cycle. Why is it designed this way? also reported. How are these rates computed?
Questions about how users earn credit on the site Additional comments
Some of the things we found in the data that really seem 16. Thanks for answering our questions! Do you have any
to affect one’s ability to earn credit on DIME are (a) how additional comments, or questions for us?
early a user joins a torrent, (b) the seeder/leecher ratio on a
torrent, (c) how often a user visits the site, and (d) the kind
B Answers from DIME moderators
of Internet connection a user has. These observations have 1. a) General History
also often been made by moderators and users, as things DIME went online as easytree.org (EZT) on January 11,
that affect one’s ability to earn credit on the site. Here are 2004. In the beginning, the site was meant mainly (but not
some graphs on each of these observations. Please let us exclusively for the so-called Vanatics (Van Morrison fans),
know if you are interested in seeing more graphs, or details who were looking for a simple and fast way to share boot-
on how we performed the measurements: leg recordings among them. At the same time the site
(a) graph on upload rate vs lifetime was meant to be an alternate approach to the then top
(b) graph on upload rate vs seeder/leecher ratio egroove.org (STG). STG had a nice forum, but their actual
(c) graph of ratio vs. days between visits sharing functionality was way over the hill.
[note: having 0 days between visits means a user was there In the course of trying to keep up with EZT’s sharing
every day] functionality and to migrate the tracker functionality to the
(d) graph of pairwise upload ratio vs pairwise bandwidth forum STG ate more than it could chew and went offline
ratio [we compare the ratios in the upload because of severe technical problems in late summer 2004.
rates of a pair of users over durations in which they are Many (all?) active STG members changed to EZT. Since
connected to the same torrent. A user’s bandwidth is termed then EZT became a tracker for all kinds of bootlegs and the
based on their max bandwidth as reported by the system] Van-Morrison portion became smaller.
The following questions are related to these observations. In the first week of April 2005 EZT was shut down by its
7. Do any of these results surprise you? former provider because of a copyright infringement claim
8. Joining a torrent early seems to be quite an effective and returned as dimeadozen.org (DIME) in the second week
way of earning credit. Do you see any potential issues with of April 2005. In May 2005 another shut down occurred
this strategy that may be harmful to the system, or do you which resulted in a complete rework of DIME’s ToS. Since
welcome it? then DIME was left alone in this regard.
9. Visiting the site often seems to help with one’s potential 1. b) Policy Changes
to earn credit, possibly because it allows a user to join on – Officially released material In the beginning there were
torrents he is interested in while they are still young. But almost no according rules, except for: no officially released
this also means that users who do not visit the site as often material. As long as the percentage of officially released ma-
are less likely to earn a lot of credit, which may be causing terial did not exceed 20for a live recording, a torrent was ok
for EZT. This was meant to keep the flow of a live perfor- mentation of a sophisticated search functionality, which was
mance intact. Officially released studio material was (and originally encouraged by the community.
still is) off-limits. 3. The moderators
After the April 2005 shut down the ToS were adapted for The moderators help with technical advice and try to ed-
the first time: officially released material, even if out-of-print ucate users about BitTorrent and BitTorrent etiquette in
(OOP) was not allowed to be included in a torrent any more. general.
After the May 2005 shut down another root-and-branch The moderators serve mainly as user-help-desk (UHD)
change of the ToS was introduced, which is, in principle, and keep the tracker clean from unwanted material. They
still valid today: anything iffy (smelling like officially re- also calm down so-called “flame-wars” which pop up in tor-
leased material) is not allowed to be included in torrent. rents’ threads from time to time.
Additionally, the NAB- and NAV-Lists were created. They In general, the moderators are to be contacted when what-
allow for artists and venues to “opt-out” from DIME. ever problems occur (forgotten passwords, problems with
In the course of the years there was some fine-tuning here seeding/leeching torrents...)
and there, but the basics are valid without major changes The moderators’ goals are to keep the site running as
since 2005. smoothly as possible. When an user has run their account
– Share Ratio Requirement and Enforcement into share ratio violation and is blocked from further down-
Mid of 2004 a method became available how a BitTorrent loading the moderators offer advice and help the user to get
tracker can identify its users/peers independent from their back on track. Users who abuse the system (“serial leeches”)
IP addresses. (Quite some time before the “key”-announce- by only taking without giving back and opening account af-
parameter was introduced to BitTorrent specifications). At ter account are banned from the site.
this time I had the idea that EZT could stop the hit-and-run 4.
leechers by using this method. The user identification and Nowhere on DIME or in the Wiki we recommend to main-
the share ratio enforcement with an enforcement cycle of 5 tain a share ratio of 1. We know that the request of a share
GB was developed and implemented within one month. The ratio of 1 is a widely spread bad habit of BitTorrent track-
functions went online end of September 2004. Before a poll ers. We also know that it is mathematically impossible that
among the Yahoo-Mailing-List members had ascertained a each member of a swarm reaches a share ratio of 1. That
reasonable share ratio. The majority decided for 0.25. This was the reason that in the aforementioned poll a share ratio
share ratio requirement is valid until today and turned out of 1 was not among the available choices. A share ratio of 1
to be ideal. Hence, there are no plans to change it. The was always out of the question for EZT/DIME.
enforcement cycle of 5 GB was then randomly decided by 5.
myself, however it should be sufficient for a DVD5, that’s Tough question. Please remind me in four weeks to answer
all I worried about. the question – then I may have an answer for you. Until now,
– Peers Limit I didn’t give this a thought and therefore, can not answer it
Mid of 2006 we found that the trackers were overloaded at the moment.
mostly because of “pointless announcements”. We at DIME Some thoughts by jupiter2101:
consider as “pointless announcements” announcements of one Torrent sizes increased considerably in the course of the
seeder for a torrent without leechers, or the announcement years. In the beginning, there were hardly any video torrents
of a leecher to a torrent, where the leecher is the only peer and if, they were in VCD quality. Nowadays there are not
in the swarm. Of course, the peer does not know in ad- only single-layer but also double-layer DVDs and even HD
vance that its annoucement is pointless, but we had to do videos. Therefore, an increase of the minimum share ratio
something about it nonetheless in the interest of all users, is out of the question. If at all, the enforcement cycle of 5
in order to prevent the trackers to be overloaded by suchlike GB would have to be increased as well. The downside would
announcements. Therefore the “peers limit” was introduced. be that the hit-and-run leechers would be able to take even
In the beginning it was a static limit and restricted the users’ more without giving.
number of peers. In 2007 the peers limit became dynamic The same goes for lowering the minimum share ratio. It
and tries to “predict”, if the known bandwidth of a new peer would only be of advantage for abusers.
is of advantage to a torrent’s distribution speed or not. If One also has to take into consideration that the modera-
it is of advantage, then an user is allowed to overdraw its tors can reset an user’s enforcement cycle when the user has
peers limit, if not, then not. not yet exceeded 10 GB downloaded. That means an user in
2. share ratio violation can be released temporarily from share
Until today, two major changes were done where the com- ratio violation and is allowed to download again – and can
munity was involved. use this chance to improve their share ratio until they’re
The minimum share ratio was determined by the commu- tested for their ratio again.
nity. There were several tries (especially after introduction 6. a) See answer “1. b) Share Ratio Enforcment”. 6. b) I
of the share ratio) by community members to change the always wanted to stay away from “selling downloads”. Cred-
minimum share ratio. However, those weren’t successful be- iting donations as direct SR improvements would be nothing
cause of the communities’ resistance. else than “selling downloads”. That was the reason for in-
The torrent ratings (1 to 5 stars) were disabled upon re- troducing the “enforcement cycle” and extending the cycle
quest of the community, as certain users were abusing the by donations: you can’t buy yourself out of your sharing
function (bad ratings for torrents of other users they dis- obligations by donating – you only can buy yourself a little
liked). bit more time to fullfil your sharing obligations, that’s all.
Of course there are always enhancements, which date more 7. No surprises in these graphs, except for one: the in-
or less from the community. Currently we are at the imple- creased upload rate at seeder/leecher ratios gt 105 in graph
(b) the user is a very active user, the intrinsic value is high. If
8. Well, that’s the nature of Bittorrent: the early bird the time frame is very small, there’s no intrinsic value in
wins the most. No, I don’t see anything harmful to the these figures.
site in this strategy/system. If it would be the other way 16. Well, some questions I’ve expected have not been
round (the late bird wins the most), no data would ever be asked. Feel free to ask them as soon as they come to your
distributed/shared! mind... ...it’s fun. ;-)
Some thoughts by jupiter2101:
Not necessarily. One can also maintain a healthy share ra-
tio by checking the torrents one has downloaded frequently
and open their window (respectively resuming to seed) when-
ever help is needed. IMO, the trick is to keep one’s down-
loads in the mix.
9. No, we want DIME to be a vital community. The more
often members visit the site, the better for the community!
10. a) No, sorry, no idea. 10. b) Downloading shows
an user doesn’t want just for increasing his ratio rewards
him with an increased SR value which allows him (earlier
or later) to download shows he/she wants. Why should we
need other ways for rewarding him for something he didn’t
do by heart?!
11. a) See jupiter’s remark on question 8, it pretty much
sums it up.
b) One may have better chances during the European day-
light hours when the North Americans are asleep. Certainly
the vast majority of members lives in America (North and
South). But that’s pure speculation, there is no hard proof.
c) Again maths. When you join an obscure torrent early
your chances to share may be better compared to a torrent
attracting many users where there is heavy competition be-
tween the peers. In the long run a torrent from a more
popular category/artist may be more rewarding. See notes
to question 8 respectively 11. a).
IMO, each user has to develop their own strategy, depend-
ing on their musical taste, their habits (not everyone can
be online 24/7), and internet connection (speed, bandwidth
limitations).
12. Nice idea, have to think about it. But no, never
considered before.
13. No, not considered yet, and will never happen.
14. “Freeloader days” are meant for those who struggle
hard but slipped a bit down minimum SR requirement and
only need a slight push to be back in “healthy state” again.
It was first introduced in 2008 – IIRC. Freeloader days are
usually on special occasions (holidays...).
15. a) Ul/dl speed of peers: The ul/dl speed of a peer is
computed at times of an announcements of this peer: () / . This is always a historical value,
valid only for the period between the last two announce-
ments.
b) Max ul/dl speeds of users: The max ul resp. dl speeds
of an user is always computed at times of an annoucements
of any of his peers. All available ul resp. dl speeds of all
current peers of this user are summed up and the results
are compared with the stored figures in the user profile. If
the new computed value is gt the stored one, the new value
gets stored to the user profile. The ’max speed’ figures say
nothing about the real bandwidth of an user, they say only
what value the tracker saw as a max for this user, yet. So,
the intrinsic value of the ’max speed’ figures depend on the
time frame an user is a DIME member and the amount of
his participations in swarms. If this time frame is large and