Anonymous Communications : A survey on I2P

Gildas Nya Tchabe and Yinhua Xu

Technische University of Darmstadt, Theoretische Informatik ,Kryptographie

and Computeralgebra Cryptography, Privacy and Security
Karolinenplatz 5, 64289 Darmstadt, Germany
https://www.cdc.informatik.tu-darmstadt.de/cdc/

Abstract. Since some latest revelation from some secret organizations,

the interest in anonymous communication have been gaining more and
more attention from internet users as privacy and anonymity problem
have come out. to achieve online anonymity, several software or platform
such as Invisible Internet project(I2P), Tor, and many others have been
developed to help internet users to protect their anonymity. In this pa-
per we provide a brief overview of I2P, then gives some weaknesses and
advantages and finally gives a brief comparison between I2P and Tor
network.

1 Introduction
The rapid growth of the Internet applications has made communication privacy
an increasingly important security requirement. Although encryption aims at
preserving some modification of data, it is still possible for the adversaries to get
significant information about the traffic carried on the network packets and the
physical entities, such as the network addresses of the sender and the receiver of
the message.
One of the main problem is the exposure of the network address which might
result in severe consequences: Adversaries can easily overhear all the messages
and perform traffic analysis; even if the communication content is encrypted,
routing information is still sent in the clear because routers need packets desti-
nations in order to route them in the right direction.
Several platforms were developed to help increase anonymity in Internet, as
an example, we have Tor network which is one of the first solution to provide
anonymous communication[1]. But Tor network later came up to have some lim-
itations in a sense that it is built on a centralized system. To resolve those con-
straints, developers have came up with a distributed alternative for file-sharing
which was then followed by the appearance of P2P networking. And then later
followed by a new project called the Invisible Internet Project (I2P)[2].
I2P has been developed by a group of anonymous people and supporters,
whereby the main developer and person responsible for this project is known by
the nickname jrandom. The I2P developers’ concept was to implement a great
and unique idea for distributed P2P anonymous systems, which provide its users
a better anonymity and security[3].
2 Gildas Nya Tchabe and Yinhua Xu

2 Description of I2P Network

I2P is described as a self-organizing network, providing better scalability, and

has a resilient packet switched anonymous network overlay over which a certain
number of anonymity and security can operate. I2P is based on what is called
I2P-router. The router is a virtual software that runs on the host and provides
a bridge to the local application. I2P application has two types of capability. It
could either access the darknet services and playing the role of client, or access
the host services and playing the role of the server. Both approaches are linked
together using a peer-to-peer approach(completely decentralized). This link runs
as overlay on top of IP. I2P to be able to exchanged messages on internet, needs
protocols as well. There are two kind of protocols which I2P is able to use :
either a TCP(NTCP) or UDP protocol(SSU) (see figure 1). The connection to
the tunnels are mapped by the router[4].

figure 1 : layer overview[12]

As many other software that are designed to protect user’s anonymity, I2P
also allows the implementation of an additional layer of encryption model and
 Anonymous Communications : A survey on I2P 3

a particular routing system known as ” Garlic routing”. Garlic routing is quite

identical to Tor’s onion routing. The difference with onion routing is that Garlic
routing extends onion by bundling different messages together. All informations
that are transmitted through the tunnels are fully encrypted and only the re-
ceiver’s router is able to decrypt the message.

2.1 I2P Components

The fundamental components in I2P network are: router, tunnels, NetDB.

Router: As mentioned before, routers are just simple software that partici-
pate in the network.
Tunnel: They are unidirectional path through several routers, which means
that the sending path and the receiving path are different. (example for bob
and alice want to communicate via I2P, they actually require 4 tunnels). After a
certain amount of time, tunnels get expired. Tunnels are checked every time to
remove failing tunnels. The default lifetime is set to 10 minutes. The length of a
tunnel hop usually varies. Tunnels could be exploratory or client. there are two
type of tunnels (inbound and outbound tunnels) see figure 2 that is proposed by
I2P[5].

figure 2 : I2P tunnels overview

4 Gildas Nya Tchabe and Yinhua Xu

Inbound and outbound tunnels are automatically built when I2P is started.
It is also important to notice that connections to tunnels are only valid for
nodes(systems) over which I2P has an installed paths. As shown in figure 2, the
first tunnel is a tunnel gateway and the last tunnel is the the tunnel endpoint.
NetDB : One of the key concept of I2P is the network database called
NetDB. it is composed of a pair of algorithms(DHT Kademlia and floodfill) that
make the sharing of metadata possible[13].

2.2 I2P routing

I2P practically faces the same problem as in some other peers-to-peers systems,
which is to find available peers and the services offered by those peers. In I2P
network, peers can only be identified by a data structure which is known as
routerInfo. RouterInfo contains all the necessary information about the available
peers. It also contains the encryption information such as 2048-Bit Elgamal
encryption key, 1024-Bit DSA, AES 256/CBC Mode, MAC, (HMAC-MD5-128)
and valid certificates as well. I2P to discover others peers, use a non-anonymous
HTTP to download list of routerInfo of available peers from a fixed location.
I2P’s DHT: After getting a list of peers, I2P uses a super-peer DHT to
build the NetDB. the NetDB contains all informations about the peers and
services that are available in the network. The super-peer in the DHT are peers
having more responsibilities(exploratory) than the other peers. One of their main
responsibility is to maintain the databases. There are called floodfill. Floodfill
peers is also responsible for getting the informations about closest ID. In order
to get the information the distance has to be calculated. It is calculated by
using Kademlia’s XOR distance metric[8]. For a peer to become a floodfill, some
conditions have to be respected:
- Have a sufficient bandwidth
- Check its configuration
After completing those steps, the peer seeking to become a floodfill also
has to check whether the active floodfill drops below a certain threshold before
promoting itself to floodfill peers.
Information about how to contact peers are kept in the so-called leaseSet.
The leaseSet are stored in the same NetDB that also contains routerInfos. But
the LeaseSet and the routerInfo are completely independent entities. They are
just stored in the same database. A leaseSet main job is to specify a set of entry
points(called leases) to the service. As mentioned before the entry point of a
tunnel is the peer currently serving as inbound gateway to the service[6,7].
For the LeaseSet and the RouterInfo to be stored in the NetDB, a request
has to be sent to super-peer. After the floodfill peer receives the request, it
forwards the request to seven other super-peers and then sends the confirmation
to the initiator. To retrieve data from the NetDB, first we have to mention
that the retrieving of Data into the NetDB are performed via tunnels. The
retrieving procedure is done this way: firstly the request is transmitted to the
(with respect to the destination address) two super-peers ( both super-peers
have to been known by the requester). In case a floodfill does not have the
 Anonymous Communications : A survey on I2P 5

requested Information, a list of other close floodfill is sent back. The requesting
peers will continue to query others floodfills peers until all known floodfills have
been contacted. Then the floodfill sends the response to the initiator through
the inbound tunnels[15] .

3 Thread Models and I2Ps Weaknesses

Although a detailed introduction has been expanded about the property of pro-
tecting users anonymity, it is still a hard question to answer, whether I2P satisfies
our particular or special demands on anonymity appropriately. The website[16]
has mentioned the 18 thread models which will lead to various results and we
will find the most effective ways from them to show the shortcomings of I2P. In
reference to [18], we will unfold the DoS and deanonymization attacks on the
basis of [16].

3.1 Brute Force Attack

The firstly considered attack always goes in a brute force way, in which the
attacker tries to observe all data streams and associate the possible sending
paths with possible endpoints. If such a try against I2P is trivial, the attacker
will be confused by the behaviour of frequent data transmission by all peers of
the network and the changing size of content-varying traffics. In addition, it is
difficult for the external observer to access the messages because of encoded and
streamed communication among the intern routers. However, an attacker, with
plentiful resource and also occupying a large ISP (Internet Service Provider) or
an Internet exchange point, concentrates a huge amount of data, like 10 GB, on
one inbound endpoint in a short period and excludes those peers not receiving
the 10 GB data, which guides the attacker to find possible routes to the real
destination.
In order to fight against the attack, one will on one hand set lower bandwidth
threshold, nontrivial latency and restricted routes (not yet implemented) as some
countermeasures, but at the same time the quality of service cannot be guaran-
teed. On the other hand one can think about unpublished or encrypted leaseSets
for eepsites or just handle a peer selection algorithm with high reliability.

3.2 Intersection Attack

We give two assumptions, one is that the attacker takes control of both ends of
the tunnel simultaneously, and the other that the size of the network is not large
and expands only slightly, so that the monitor can regularly go in touch with
his quarries and keep track of as many peers of the network as possible. Over
quite a long period, during which node churn happens, the exact location of the
quarry will be indirectly exposed by doing the intersection of the routes where
data are successfully passing, i.e. ignoring the non-participants in contact with
the target and narrowing down the number of candidate routes.
6 Gildas Nya Tchabe and Yinhua Xu

It is said in [17] such an attack would be more effective in combination with

other types of attacks as Sybil attacks, which we will introduce later, and timing
attacks. If the attack do not launch a combinatorial attack mentioned above,
some succeeding steps are taken against the attack: as a partial defense, ordering
the peers in a strict plan, altering the peer profiling and selection in the network
in slow pace, decreasing the number of tunnels via each peer and supporting
hosting on multiple routers at one time; for further fixing of the vulnerability,
abandoning guard nodes with low bandwidth, traffic shifting among tunnels,
tunnels with a short life cycle, configurable tunnel lengths will come to help as
well.

3.3 Partitioning Attack

Just as the name implies, in a partitioning attack the peers of a network are to
be divided into separate smaller zones. Here the partitioning attack is classified
into two categories: the technical and the analytical one. In regard to the techni-
cal partitioning, the connections between peers are cut to fragment network into
pieces with the help of the information I2Ps built-in network database, which
stores peers statistics and records and hence decides whether an existent link
to other sections will be utilized to heal the disrupted network. Under the ter-
rible assumption that the attacker disconnects all contacts with peers, and as
a result the target zone will be left alone, the network database will not help
fix it, because the zone is unreachable in accordance with routing-table. The
best choice for the router in this case is to alert the user that all connections
are temporarily lost, if it can eventually notice the disappearance of a gigantic
number of previously available peers. Of course, the network size is a key factor
in protecting the anonymity of the target.
Another powerful kind of partitioning is characterized to be analytical. Under
this circumstances one records different behaviours of routers and endpoints, and
segregates them in terms of their behaviour, i.e. an attacker gets hold of a list
of peers inside the I2P network by planting a peer in it and then harvesting
information about the unknown neighbours, so that it will get well acquainted
with the exact number of inbound tunnels extended to the target, and segment
the peers in the light of criterion, e.g. number of tunnels. An attacker has to
remember: before the trick he needs to make preparations, like booting its own
routers into a separate network after taking over a significant part of the network,
or cheating the developers to trust it.

3.4 Denial of Service Attack

Like brute force attack, the DoS will also produce a large amount of messages,
but quite differently, the DoS degenerates or even terminates the network ser-
vice in an abnormal way instead of identifying one single peer. In one case, the
attacker overloads the peer processors by flooding a tremendous number of re-
quests for some cryptographically complex operations. Attempting to mitigate
the trouble, only one with good-formed network engineering skills can echo the
 Anonymous Communications : A survey on I2P 7

expensive requirements, nevertheless, the attacker will exploit some bugs in the
implementation of countermeasures all the same. The bugs of the applications
will not be discussed in this paper.

Floodfill Takeover One of the most efficient DoS attack is floodfill takeover,
where an attacker creates multiple fraud identities as bases to control the total
network. In fact, the participants will estimate the performance of the known
peers and measure them in connection. It turns out definitely, that multiple
identities on the same host limits the performance of each malicious peer, in
another word, no sufficient resource will be distributed to those additionally cre-
ated identities. Nonetheless, the attacker still has a opportunity to makes the
participation of its peers in the NetDB possible by following the next instruc-
tions.
The availability of resources is weighted in regard to two available data rate,
statically and administratively configured for peers, and job lag, the average
delay between continuous scheduled tasks during operation. The attacker can
generate more scheduled assignments to deploy the job lag, which is significantly
dependent on the number of open tasks. For the network load changes and
routers are possible to be rebooted from time to time, the attacker would like to
wait for the decrement of the legal floodfill participants number so as to make
additional malicious identities alive in the network, especially when a churn
happens.
Furthermore, the attacker can even lengthen the job lag in cooperation with
DoS attacks against a legal floodfill participant to raise update frequency of
floodfill nodes. He not only turns many legitimate nodes down, but builds many
new tunnels through them as well to order more searching jobs, which can over-
load the victims in the tunnel and build up a job lag. It should be noticed that
the emission of large amounts of data through the victim will be avoided by the
data limitation mechanism, which forces the victim to drop job requests and
lower its burden, and a DoS attack is launched on one victim at a time only
after the last victim quits from the floodfill set.

Sybil Attack A Sybil attack is doubtlessly an evolutional version from floodfill

takeover, which allows the attacker to predominate just a limited part of the
keyspace, consuming fewer resources than a complete takeover. Despite that the
attacks on multiple I2P nodes cannot be performed in parallel due to the peer
profiling, it is still possible to take down a large quantity of healthy identities
and activate the malicious ones around the victim. In this condition an attacker
will completely take over all intended positions in the keyspace within a couple
of minutes.
It takes a long setup time, about an hour, for a new node to get access into the
network, before the Sybil attack can really be in effe. The mapping from leaseSets
and routerInfos to netDB keys, in which the attacker is actually interested,
contains the current date with respect to UTC. As a result the keyspace is
not kept stable, which results in that the attacking nodes have to hold flexible
8 Gildas Nya Tchabe and Yinhua Xu

positions in the keyspace to prevent to conceal themselves, whereas the substitute

attacking nodes will be in right position thanks to the predictable behaviour and
no random nonce before a new day begins and the attack will be run with merely
few additional resources. In [10] the former attacks are described on occasion of
full keyspace, and later attacks of partial keyspace.

Eclipse Attack The final goal of this attack is to make netDB unavailable to
legitimate net- work participants like other DoS attacks. The attacker should
isolate the victim to prevent it to find a key in the database by holding all its
neighbours, otherwise a lookup-failing message from an uncontrolled neighbor
will tell the victim that the attacker is intruding. If the attacker builds up a
firewall in this region, he blocks all the access requests to items and responds
to them with the answer ”information not found . Gradually the legal ones will
not be served and even not have no interaction with the nodes which should be
visible to them.

3.5 Deanonymization Attack

This attack allows the attacker to correlate the service a legitimate participant
uses to his IP address. For the sake of successful achievements, we will arrange
malicious nodes in the netDB, as we have described earlier in the last section,
to observe the relationship of the events in the network.
As peers store data they gather on the closest known floodfill node, they
often verify the proper storage of these information by operating a lookup in
other close floodfill nodes. If both storage node and verification node are con-
trolled by the same person, the manipulator will set up connection between them
and monitor both communications. For the peer information is stored without
a tunnel, the peer will be exposed by the content of the netDB entry for all
aspects; for the verification can be figured out through an exploratory tunnel,
the tunnel endpoint will be exposed anyway. The attacker can take advantage
of the combination of both to create a logical mapping from exploratory tunnel
endpoints to the tunnel owners and for sure effectively deanonymizes the user.
If service information expires some time later, every user has right to fetch it
before starting a communication with a service and updates it regularly during
the communication. Then the attacker gains the opportunity to identify which
of the monitored peers communicate with each of the observed resources and
its precise time point. The regular update of service information additionally
reveals how long the service can be enjoyed by a certain peer. In this way the
attacker is also able to deanonymize users concerning their utilization of certain
services.

4 I2P vs Tor
As mentioned at the beginning of our report, there are a lot of software that are
deployed to help to provide anonymity of internet users. One of the first system
is Tor. Here we compare the Tor network to I2P network.
 Anonymous Communications : A survey on I2P 9

4.1 Tor
Tor network is a free software and open source that helps to provide anonymity
of users. Tor is composed with 3 different types of node: directory servers, exit
points, and internal relays. The directory server contains a list of relays which
are to be obtained by the client. The directory servers do not only send the list
of relays to the client but also his address and some basic configuration files. In
order for the client to get valid relays, he needs to choose a trusted server. Once
the client has received a list of valid and operational relays, he could choose an
optional route for his traffic across the Tor network and later transmit it to the
exit node. Tor components are basically the PC to which you are connected,
relays, routers, and exit node[9,10].
In Tor network all traffic follow the same route and exit at the same point.
This approach is quite different to way I2P operates. During the traffic all mes-
sage are fully encrypted as in I2P network. The routing protocol in Tor is known
as ”Onion routing” which are used to repeatedly encrypt the original data in-
cluding destination IP address, and send it through several network nodes called
onion routers. It is also important to notice that exit node is only aware of it
intermediate node to send/ receive data. He does not know what is actually in
the message because it is fully encrypted . The difference between the exit and
entry node is that: The entry node knows ” who you are” but not ”what you
are doing ” while the exit node knows ” what you are doing” but not ”who you
are”. The relays in between only forward encrypted messages.
In addition to Tor capability, it is possible for the users to access what it
called ”hidden services” such as emails servers, forums etc.

4.2 I2P
Tor and I2P both have a lot of benefits in common since they are both designed
to help provide better anonymity. But from the beginning Tor was designed to
help users to access public network anonymously . while I2P first function was
to be a network between the network ”being a true darknet”[10].
In contrast to Tor, all nodes in I2P act as routers. As mentioned before I2P
uses different route than Tor network: I2P traffic are unidirectional (sending and
receiving route are different) which help to increase reliability and redundancy
to the network while Tor uses a simple duplex circuit (sending and receiving
messages go through the same path).
In I2P, informations about peers and en/decryption keys are stored in the
NetDB while in Tor those informations are stored in a central directory servers.
Concerning the application-level, The users to be able to use Tor, the must
configure the proxy services which are stored on his computer. While in I2P,
special applications have already been written and must only be used in I2P
network (i2pSnark, eepsites, and i2p-messenger ).
Except that I2P and Tor distinguish themselves from each other in some other
details, which will not be discussed in this paper, the main part of which was
summarized briefly by in a table[11]. Most of them have already been mentioned
earlier.
10 Gildas Nya Tchabe and Yinhua Xu

I2P Tor
Communication Method Packet-Switched Circuit-Switched
Communication Protocol TCP or UDP TCP
Type of Tunnels Unidirectional Biddrectional
Routing Algorithm Garlic Onion
Lifetime of Tunnels Short-Lived Long-Lived
Encryption Type Link-, Layered- and End to End encryption Link -and Layer Encryption
Storage of Peer Info Floodfill Peers 7 Directory Servers
Service Provider Build-In Servers External TCP Servers
Number of Services Many Integrated Services Few Hidden Services
Peer Selection Performance-Based Bandwidth-Based
Hop Number in the Tunnel User-Configurable, Random Number of Hops 3-Hop
Number of Exit Nodes One A large Number
Network Dimension SMall Large
Implementation Code Java C
Table 1. Main Differences between I2P and Tor Network
Some advantages of Tor Over I2P: Tor offers better resistance, performance[14],
documentations, low bandwidth overhead, more translations. While I2P on the
other side is faster, completely distributed, self organized, selection of peers
through ranking performance[12].

5 Conclusion
In this paper we firstly presented a brief description of I2Ps components then
we commented on some weaknesses such as DoS etc. We compared the routing
aspects of I2P system to Tor and saw that both Tor and I2P provide crypto-
graphically methods to anonymously access information and communicate on-
line. Tor provides service with better anonymity for Internet and higher QoS
for any network, while I2P provides a more robust and reliable network within
the network. Of course, when implementing either of these two tools, one must
always be aware of that ones ISP can see that he or she is using Tor or I2P
(though they cannot determine the content of the traffic itself). In order to hide
this knowledge from ones ISP, one should make use of a high-quality VPN ser-
vice to act as an entry point to either ones anonymous network arbitrarily or
to the Internet at large. One of the key elements that annoy anonymous sys-
tem researchers is QoS for the bandwidth utilized by peers on the systems and
the overall network performance. Although this has been slightly commented
on, more research in QoS and a bandwidth-choking approach is required while
concentrating on security and functionality implications.

References
1. Abdelberi Chaabane, Pere Manils, and Mohamed Ali Kaafar. Digging into Anony-
mous Traffic: A Deep Analysis of the Tor Anonymous Network. In Proceedings of
 Anonymous Communications : A survey on I2P 11

the 2010 4th International Conference on Network and System Security, NSS ’10,
Washington, DC, USA, September 2010. IEEE Computer Society.
2. J. Jrandom, I2P Anonymous Network: Technical Introduction, Retrievedon Decem-
ber 13, 2010, from Anonymous Network.,http://www.i2p2.de/techintro.html
3. Privacy-Implications of Performance-Based Peer Selection by Onion-Routers:A
Real-World Case Study using I2P. Michael Herrmann and Christian Grothoff.
http://grothoff.org/christian/i2p.pdf
4. ”Introducing I2P”., http://geti2p.net/en/docs/how/tech-intro
5. wiki.ubuntuusers.de/i2p
6. Goldschlag, D. M., Reed, M. G. and Syverson, P. F.: Hiding Routing Information. In:
Information Hiding. Lecture Notes in Computer Science. Cambridge, UK: Springer-
Verlag. May 30 - June 1, 1996: 137 - 150.
7. Maymounkov, P, Mazieres, D.: Kademlia: A peer-to-peer information system based
on the xor metric. p. 53-65 (2002)
8. Dingledine, R., Mathewson, N., Syverson, P.: Tor: The second-generation onion
router. In: Proceedings of the 13th USENIX Security Symposium (August 2004)
9. Ehlert M.: I2P Usability vs. Tor Usability: A Bandwidth and Latency Comparison.
Seminar. Humboldt University of Berlin. Berlin, Germany, November 2011.
10. Herrmann, M. and Grothoff, C.: Privacy Implications of Performance-Based Peer
Selection by Onion Routers: A Real-World Case Study Using I2P. In the Proceedings
of the 11th Privacy Enhancing Technologies Symposium (PETS 11). Waterloo, ON,
Canada, July 27 - 29, 2011.
11. http://i2hq.srv.i2p2.de/de/docs/protocol
12. I2P Compared to Tor., https://geti2p.net/de/comparison/tor
13. TMA2012-LNCS.pdf. http://hal.archives-ouvertes.fr/docs/00/63/22/59/
PDF/TMA2012-LNCS.pdf
14. Panchenko, A.; Lanze, F.; Engel, T. ”Improving performance and anonymity in the
Tor network”, Performance Computing and Communications Conference (IPCCC),
2012 IEEE 31st International, On page(s): 1 - 10
15. B. Zantout and R. Haraty, I2P data communication system, in ICN 2011, The
Tenth International Conference on Networks, 2011, pp. 401409.
16. I2Ps Threat Model.https://geti2p.net/en/docs/how/threat-model.
17. Zantout, B.C. and Haraty, R.A.: I2P Communication System. In the Proceed-
ings of 10th International Conference on Networks (ICN 11). Saint Maarten, The
Netherlands Antilles, January 23 - 28, 2011: 401 - 409.
18. Egger, C., Schlumberger, J., Kr gel, C. and Vigna, G.: Practical Attacks against the
I2P Network. In the Proceedings of the 16th International Symposium on Research
in Attacks, Intrusions and Defenses (RAID 13). Rodney Bay, Saint Lucia, October
23 - 25, 2013