EXAMINING THE MYTHS AND FACTS CONCERNING BITCOMET BEHAVIOR
By Robb Topolski, July 31, 2007

FORWARD

The purpose of this paper is to provide independent data to the BitTorrent
community about several persistent concerns regarding the BitComet client.

I am posting this information to the BitComet forum, and I will be
notifying certain purveyors of P2P news and information about it.

My name is Robb Topolski, and I’ve been on the Internet since before the
World-Wide-Web! I am a networking and protocol expert, with more than 25
years of experience. I have been recognized as a Certified Software Quality
Engineer by the American Society for Quality, and I have been recognized as
a Microsoft Most-Valued Professional in Networking. I consider myself an
ever-learning expert in software quality, security, testing, and
networking.

I was not asked by anyone to do this testing and analysis. I have
approached the managers of the BitComet forum in an attempt to get
information, but they have not been notified in advance of my testing
results. This work is independent and is without the permission or direction
of anyone on the BitComet team.

I have been studying the BitTorrent protocol for over a year, and grew
curious about these accusations concerning BitComet. I decided to test some
of the most popular “accusations.” In some cases, I applied direct
testing, using popular analysis tools such as Wireshark to observe the
“wire” behaviors of BitComet. In some cases, I determined that the
protocol itself lent either credit or discredit to a particular accusation.

I found that my ISP was interfering with my tests -- even showing some of
the “quick disconnect” behaviors frequently associated with BitComet
(although the behaviors were seen across all clients). My ISP frequently
“shapes” or redirects P2P traffic using several methods. To avoid this,
all of my testing was either performed on my own LAN or on the public
Internet using a secure VPN tunnel. Through the VPN, these interferences
disappeared.

I am currently having long-term medical issues that leave me homebound, and
a full day’s work is impossible. But it has given me the time to be able
to study this closely, and keep my skills current, and this paper has slowly
come together as a result.

My examination of these accusations and of the behavior of the BitComet
client indicates that the raised concerns are consistently false.
BitComet’s reputation has suffered due to a December 2005 DHT “leak”
bug and a false perception that response to that bug was slow. Banned on
many trackers since that bug, several other observed or imagined issues have
been “piled on” to further tarnish the reputation of BitComet. Tracker
admins reading this paper should conclude that BitComet is not the
detestable client it is perceived to be.

Note: I did not go back to early versions -- so my results are limited only
to the versions that I tested (which are listed below).





ISSUE #1: BitComet abuses or ignores the DHT flag on private torrents.

CONCLUSION: TRUE ONLY FOR VERSION 0.60

SUMMATION: A bug existed that allowed the addition of DHT to existing
“private” download tasks. The bug was found in version 0.60 in early
December 2005, and the version was withdrawn two weeks later. The bug fixed
in version 0.61 in January 2006 (http://www.slyck.com/story1094.html).
Several private trackers banned BitComet as a result of this bug.




ISSUE #2: BitComet’s developer, RnySmile, failed to acknowledge or act
upon the DHT bug (ISSUE #1) for several months.

CONCLUSION: FALSE (WAS NEVER TRUE)

SUMMATION: The timeline of the DHT bug shows that the buggy version (0.60)
was replaced with the previous stable version of BitComet (0.59) within two
weeks of the initial report. Three weeks later, version 0.61 was released
which did not contain the DHT bug.





ISSUE #3: BitComet disconnects and immediately reconnects (approximately 10
times per second) in an attempt to be unchoked by the peer.

CONCLUSION: FALSE IN TESTED VERSIONS 0.70 (JUNE 2006) AND LATER

SUMMATION: If true, such reconnecting might give BitComet a 33% better
chance in being unchoked sooner than if it remained connected. This
provision in the BitTorrent protocol is intended to give new clients in a
swarm some piece data to share. It is not intended for clients that return
to a swarm. In numerous test runs, several versions of BitComet were
observed. Indeed, they do frequently disconnect from their peers. However,
it is clear that such disconnections are not an attempt to gain an
advantage. Specifically, on the 0.70 through the current 0.90 versions,
BitComet does stay connected for a short while (60-90 seconds) after being
CHOKED by a peer. After it disconnects, it DOES NOT immediately try to
reconnect. In all but one occasion, it waited more than a minute before
trying to reconnect to the swarm. The delay before disconnecting and the
delay before reconnecting would erase any advantage that BitComet would get
if it intended to exploit the BitTorrent protocol’s unchoke rules.

NOTE: I was unable to get a specific answer from anyone as to exactly why
BitComet chooses to disconnect from peers after being choked for 60-90
seconds. This behavior is allowed by the protocol, but the reason for doing
so is not clear. It may help users with weak routers as such disconnections
from idle clients would reduce the number of simultaneous connections. Other
than that, there is very little disadvantage that I can see to remaining
connected and idle. However, this is clearly a design choice and is simply a
question of efficiency. BitComet’s behavior is neither incorrect nor
damaging.





ISSUE #4: BitComet abuses Super-Seeding clients by failing to inform the
sending peer that it successfully downloaded a piece.

CONCLUSION: FALSE (WAS NEVER TRUE)

SUMMATION: Many different BitTorrent client developers have been
experimenting with withholding HAVE messages from seeding clients. The
theory is that such messages are unnecessary overhead, and that debatable
theory still holds even if the seeding client is using the Super-Seeding
(unofficial extension to the BitTorrent protocol) feature. Super-Seeders are
looking for confirmation that a unique piece that was sent to a specific
peer has been shared with a third peer in the swarm. For the unique piece
that was just sent to a client, Super-Seeders DO NOT depend on the HAVE
message from that receiving client to confirm receipt. On the contrary, the
Super-Seeder only needs the HAVE message for the unique piece from ANY OTHER
CLIENT as an indicator that the peer has shared the piece to others. (Note:
I did not test whether BitComet sent HAVE messages for received pieces or
not. This accusation is confirmed as false based on the Seeding and
Super-Seeding behaviors themselves, and the fact that the seeder never
relies on the HAVE message of the receiver.)





ISSUE #5: BitComet abuses Super-Seeding clients by immediately
disconnecting from and reconnecting to the Super-Seeder rather than sharing
the pieces it has received.

CONCLUSION: FALSE IN TESTED VERSIONS 0.70 (JUNE 2006) AND LATER

SUMMATION: By Super-Seeding a torrent and observing BitComet client
behavior in the swarm, there is no trend toward BitComet clients
“holding” unique pieces – which would be the case if this issue was
true. Furthermore, there is no BitComet feature, or combination of settings
unique to BitComet, that would produce this behavior. As noted in ISSUE #3,
BitComet does disconnect from peers more frequently than most clients.
However, in the versions tested and observed, BitComet waits 60-90 seconds
after being choked to disconnect and then waits 60 seconds or more to
reconnect
(http://forums.bitcomet.com/index.php?s=&showtopic=12787782&view=findpost&p=31159).

NOTE: Even though this is verified as false for BitComet specifically, most
clients (including BitComet) and network stacks can be hacked or adjusted to
create a poor sharer. BitTorrent client developers implementing
Super-Seeding should take care to prevent a possible exploit to
Super-Seeding by poor-sharing clients that may attempt to evade accounting
for unique pieces by disconnecting and reconnecting.






ISSUE #6: BitComet announces too frequently, “hammers” the tracker, and
ignores the Tracker’s "reannounce" interval.

CONCLUSION: FALSE IN TESTED VERSIONS 0.90 and 0.91

SUMMATION: It is common practice among BitTorrent clients to reannounce
when the client deems that it has too-few peers. In my own experience using
several BitTorrent clients, such reannouncing generally occurs between two
and five minutes. BitComet seems to wait 20 minutes. In using BitComet, I
found myself using the “Manual Connect” (manual reannounce) feature
because I felt like BitComet was waiting too long before asking the tracker
for more peers.





ISSUE #7: BitComet has an abusive multi-tracker implementation (announces
to all trackers in all tier always)

CONCLUSION: FALSE IN TESTED VERSIONS 0.90 and 0.91

SUMMATION: BitComet follows the recommended practices for multi-tracker
torrents. One tracker in each tier is contacted, and other trackers within a
tier are not contacted unless the originally-tried tracker failed to
respond.





ISSUE #8: BitComet deliberately misreports upload and download amounts to
trackers and seeds in order to get the more upload bandwidth from seeders.

CONCLUSION: FALSE ACCUSATION

SUMMATION: Simply put, the BitTorrent protocol does not work that way. The
tracker does not inform seeders that certain peers have preference or
priority. BitComet user XSTREM suggested to me that BitComet may send a
“Completed” announce message when the user elects to download only a few
files from a torrent. He suggests that some tracker managers may be using
that message in performing user-accounting. If that is true, then the
tracker managers are applying meaning to the “Completed” message that is
not supported by the specification
(http://forums.bitcomet.com/index.php?s=&showtopic=12787870&view=findpost&p=32101
and http://wiki.theory.org/BitTorrentSpecifica...est_Parameters). The
Completed message cannot be used as some kind of “checksum” to look for
malicious users.





ISSUE #9: After reconnecting, BitComet reports having fewer pieces of the
file than it had before disconnecting.

CONCLUSION: FALSE IN TESTED VERSIONS 0.70 (JUNE 2006) AND LATER

SUMMATION: This concern was raised by TheSHAD0W, the developer of the
BitTornado client and the inventor of the popular Super-Seeding behavior
used by many BitTorrent clients. After several carefully observed sessions
and controlled test runs spanning many versions of BitComet, this reported
behavior was still not seen
(http://forums.bitcomet.com/index.php?s=&showtopic=12787870&view=findpost&p=30879).
It is very possible that TheSHAD0W’s testing or development tools are
faulty, or that there is some network problem that caused him to see this
behavior. (Indeed, my own ISP is doing P2P “management” that has raised
some problems in my own testing – I had to eliminate those problems by
testing through a VPN).





ISSUE #10: BitComet seems to favor uploading to other BitComet clients,
even when getting faster download speeds from other clients.

CONCLUSION: FALSE IN TESTED VERSIONS 0.90 and 0.91

SUMMATION: In dozens of observed sessions using BitComet, I see no such
preference being given to BitComet peers.





ISSUE #11: BitComet is a poor peer due to no upload slot control. Upload
bandwidth is stretched too thin.

CONCLUSION: TRUE WHEN ONLY INITIAL SEEDING, FALSE WHEN DOWNLOADING,
NON-ISSUE WHEN SEEDING AN ALREADY-SEEDED TORRENT (IN TESTED VERSIONS 0.90
and 0.91)

SUMMATION:

This concern was raised by TheSHAD0W, as an issue affecting Super-Seeding.
However, BitComet’s slot control is only too thin if the BitComet client
is a passive seeder. As a downloader in the “Tit-for-Tat” mode,
BitComet’s behavior does not spread upload slots too thin. As a result,
TheSHAD0W’s concerns (which I also initially shared) are not warranted
after testing and observation.

When BitComet is simply seeding, nearly every peer in the peer list is
UNCHOKED and the amount of upload bandwidth given to each is often less than
1 KB/s. This is inefficient because it takes an exceptionally long time to
send any one complete piece to a peer (in BitTorrent, incomplete pieces held
by a peer cannot be spread). This becomes an issue when the BitComet client
is holding unique pieces (such as acting as the first seeder or initial
seeder to a swarm).

If the BitComet user is the initial seeder, that user will take more time
and bandwidth to seed a torrent than any other BitTorrent client I have ever
used. (Tests: BitComet 200% to 255%, MainLine 145% to 175%, uTorrent with
Super-Seeding 105% to 115%).

However, when BitComet is a non-seeding peer, it has exceptionally
intelligent slot control. BitComet adjusts the speed of each upload slot
individually, providing more upload bandwidth to peers that reciprocate with
more upload bandwidth of their own.

This practice tends to reward more good peers in a swarm than the
traditional “slot and unchoke” BitTorrent behavior. It also allows
BitComet to automatically and efficiently handle more download tasks at
once. (Available in other typical BitTorrent clients only by knowledgeable
use of the “Force” functions.)

BitComet uses all surplus upload bandwith first to try to find either other
additional reciprocating peers or to get additional download speed from
existing reciprocating peers; and second to seed to non-reciprocating peers.
As a result, BitComet downloaders will perform somewhat better in swarms
that have poor peers, and they may complete downloads with somewhat better
ratios. (Because all available upload bandwidth is always used, BitComet is
not considered a “greedy” client as described by the BitTyrant paper.)



CONCLUSIONS AND RECOMMENDATIONS:

BitComet is a worthy download client, providing some advantageous features
not found in any other current BitTorrent client. Some of these features are
confusing and are poorly implemented, but they are not detrimental to a
BitTorrent swarm, nor do they take unfair advantage. It is a moderate user
of resources. BitComet provides very little technical feedback (such as logs
and other real-time indications of activity. As a result, it is less
“Geek-friendly” as most other clients, it seems to attempt to be more
“user-friendly” instead).

BitComet is an exceptionally poor upload client and should be avoided if
the user will be the initial uploader to a swarm (see ISSUE #11 above). This
is not an issue if the BitComet user is a seeder in an already-seeded
swarm.

None of the typical accusations against BitComet, those that are provided
as reasons for trackers or users to “Ban BitComet,” have held true. It
is my professional opinion that the bans of BitComet are based on
misunderstandings and falsehoods, and not on good data. Now that tracker
admins are equipped with my objective and independent data and analysis, it
is my hope that they reconsider their bans against this client.

Respectfully submitted to the BitTorrent community.

/s/ Robb Topolski
robb@funchords.com