The Net Rat
Rethinking Connected Services for Increased Security
Bernd Pr
¨
unster
1
, Florian Reimair
2
and Andreas Reiter
2
1
Secure Information Technology Center – Austria (A-SIT), Graz, Austria
2
Institute of Applied Information Processing and Communications (IAIK), Graz University of Technology, Graz, Austria
Keywords:
Decentralisation, Serverless Services, P2P Networks, Personal Mesh Network, Distributed Services.
Abstract:
Traditional desktop computers have been outranked in terms of usage numbers by mobile devices. Still, many
popular mobile-first services rely on workflows designed decades ago for a different environment. When
relying on cloud-based services, privacy and data protection issues need to be considered. Mostly, however,
one can choose between either well-supported legacy applications or innovative niche solutions. In this paper,
we introduce the Net Rat, a framework enabling a seamless transition from existing centralised setups to
decentralised state-of-the-art services, increasing security while maintaining backwards compatibility to well-
established legacy services. We demonstrate the feasibility of our approach with a case study focusing on the
decentralisation of the e-mail service—until now, this failed due to missing backward compatibility. A security
analysis demonstrates how our approach reaches its goal of protecting user data through decentralisation. The
Net Rat is built on a solid foundation as result of a security-first design. The results of this work clearly show
the feasibility of decentralising existing services and highlight how well-established services can be improved.
Our approach also presents opportunities to develop new services based on a solid foundation.
1 INTRODUCTION
Examining today’s device landscape reveals that users
are connected to the Internet using multiple devices
1
.
In essence, mobile devices became ubiquitous and
so did connected, cloud-backed services. On the
one hand, the relatively new, mobile-first messenger
WhatsApp has over one billion active users
2
. On the
other hand, traditional services such as e-mail are still
highly relevant. Regardless of service and context,
users are accustomed to the fact that their data is avail-
able on all of their devices. Current solutions often
rely on cloud services to meet this demand and users
typically have no say in which entities are access-
ing their data.Service providers might even generate
added value out of the users’ data by showing targeted
advertisements (Google, Inc., 2014). Moreover, busi-
ness interests of service operators can change over
time, possibly leading to changes in the way sensi-
tive user data is handled. External factors such as
changes in government policies could also force ser-
vice providers to share sensitive data with public au-
1
https://www.statista.com/statistics/333861/connected-
devices-per-person-in-selected-countries/
2
https://blog.whatsapp.com/616/One-billion
thorities. This claim is supported by governments
having made statements that ways to break encryp-
tion are pursued to extract user data from cloud-based
services (Blum-Dumontet, 2017).
In this work we propose an innovative solution
to put the user back in control of their data using a
mesh-like organisation of all the user’s devices. We
introduce The Net Rat as a secure gateway for legacy
connected services on the one hand, and, on the other
hand, enable the development of new innovative ser-
vices, targeting the ubiquity of connected devices.
The Net Rat takes the characteristics of typical end-
user devices into account and delivers improved secu-
rity properties without sacrificing usability and com-
patibility. In essence, we propose to transform ex-
isting server-centric services into peer-to-peer (P2P)
applications running directly on users’ devices, with
each user operating their own P2P network to con-
nect their devices. However, compatibility to existing
(outside) services is imperative. Therefore, the Net
Rat was designed to be backwards compatible.
This paper is structured as follows: First, some
background and an analysis of the current situation
is provided. Next, our system’s architecture and its
security features are described in detail. Section 4
presents a case study illustrating how we used the Net
Prünster, B., Reimair, F. and Reiter, A.
The Net Rat - Rethinking Connected Services for Increased Security.
DOI: 10.5220/0006469103750383
In Proceedings of the 14th International Joint Conference on e-Business and Telecommunications (ICETE 2017) - Volume 4: SECRYPT, pages 375-383
ISBN: 978-989-758-259-2
Copyright © 2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
375
Rat under real-world conditions to provide a decen-
tralised e-mail service and replace existing implemen-
tations. Afterwards, we illustrate the security proper-
ties of our system by means of a security evaluation
comparing the Net Rat to traditional services. Hav-
ing elaborated on all aspects of our approach, we then
discuss related work. Section 7 finally concludes this
paper and outlines possible future work.
2 THE PRICE OF MAINTAINING
LEGACY WORKFLOWS
Observing some of today’s most popular connected
services, we have already identified three main is-
sues common to many (usually cloud-based) applica-
tions: data security concerns, significant overhead,
and legacy liabilities.
At its core, any centralised service processing data
in plain introduces some data security risks. Some-
times it is even part of a service operator’s business
model to have full access to all of its users’ data
(Dropbox, Inc., 2016). Moreover, everything happen-
ing on the back end of a service is typically out of the
user’s control. Even self-hosting services with lim-
ited complexity like e-mail typically involves a cen-
tral server. Compromising such a central instance is
typically enough to affect all devices using a service.
Times have changed since relying on central,
static instances was often the only way to provide a
service, when typical end-users had a single device,
and were only online during limited periods. The
server-centric, traditional e-mail service is the prime
example illustrating how forcing such rigid architec-
tures on today’s multi-device users introduces signif-
icant overhead: When starting to write a message on
one device, it is stored (as a draft) on a server. If a user
then chooses to continue composing this message on
another device, it first has to be downloaded. Now it is
located on three separate devices and has been routed
through the Internet at least twice. Once the message
is to be sent, it is transmitted to the server, which for-
wards it to some external server. In addition, it is also
stored as a sent message on the device, on the e-mail
server, and transferred through this server to all other
devices this e-mail account is used on.
Although it would be easy to dismiss such con-
cerns by arguing that legacy compatibility demands
certain workflows, this would be missing the point.
In fact, many popular Internet-based services have
been augmented to meet the needs of today’s multi-
device users but still operate based on paradigms de-
fined decades ago. For example, instant messengers
have been around since the nineties and even today’s
most popular messenger service, WhatsApp, is still
based on a server-centric architecture (Fiadino et al.,
2015). Consequently, the security properties of many
well-known and critical services have become more
complex over the past years. While the problem of di-
rectly connecting devices is still not fully solved, pro-
tocols likeICE (Rosenberg, 2010) have been proven
to be effective enough to enable even applications
with tight network constraints. The file sharing scene
has been dominated by BitTorrent (Cohen, 2013) for
years
3
, which demonstrates that peer-to-peer is a vi-
able way to go. We therefore believe that simply
adding new features on top of systems based on con-
cepts tailored to the environment of the 1980s will
lead to a dead end.
We show that there is another way to implement
even services demanding legacy compatibility by pro-
viding a working example of such a system based on
the Net Rat. The Net Rat also offers increased se-
curity and availability and utilises the full potential of
today’s connected device landscape by daring to ques-
tion established paradigms.
3 THE NET RAT
Our approach to harnessing the potential of current
devices comes down to creating per-user P2P net-
works. In essence, a user connects to local proxies,
which interface with a decentralised version of an ex-
isting service. The decentralised service is deployed
on each of the user’s devices and is interconnected
by a peer-to-peer network. The P2P network syn-
chronises the data among all nodes after translating
service-specific operations into service-independent
events. On each device connected to the network,
these events are transformed back and operations are
replicated. Users can therefore still use legacy client
applications while regaining control of their data.
As the Net Rat primarily targets multi-device
users, it will typically be deployed on few (< 10) de-
vices all controlled by a single user. Bootstrapping of
the P2P network is accomplished by simply (manu-
ally) registering each instance’s identifier at the other
instances.
3.1 Scope
The overall goal of the Net Rat is to offer a platform
to transparently decentralise existing services for in-
creased security. Most importantly, the applications
3
https://torrentfreak.com/bittorrent-still-dominates-
internets-upstream-traffic-151208/
SECRYPT 2017 - 14th International Conference on Security and Cryptography
376
themselves and service operations do not need to be
altered in any way. Regardless of whether an applica-
tion uses external infrastructure, the data a user op-
erates on needs to be present at the user’s devices.
The Net Rat simply eliminates the need for any exter-
nal services, thus eliminating all associated risks. By
keeping data close to the user, the Net Rat can also of-
fer increased availability at no cost. At the same time,
by only ever storing encrypted user data, it becomes
harder for attackers to extract data.
Figure 1: The Net Rat’s architecture.
Other aspects of decentralised, redundant storage
such as space-optimised data persistence or data re-
covery from lost devices are not the focus of the Net
Rat, as these issues can often be solved on the appli-
cation level. The same also holds true for remotely
wiping stolen or lost devices.
3.2 Architecture
The Net Rat combines concepts of a number of differ-
ent approaches in order to provide a scalable and flex-
ible solution to the challenges discussed initially. An
architectural overview on the Net Rat is depicted in
Figure 1. On an end-user device, the following com-
ponents are typically present: The client app, an un-
modified existing application, makes a service avail-
able to the user as usual. However, instead of using a
traditional remote service, the client app connects to
a local proxy service.
The proxy mocks a traditional service and there-
fore keeps existing client applications functional. In
order to provide its service, the proxy has to inter-
face with the storage, the authentication component
and the Event Transformation and Distribution Core
(ETDC).
Whenever the proxy service fetches data from the
storage, an authentication challenge needs to be
satisfied—this will typically involve user interaction.
However, Section 4 clearly illustrates that this does
not affect usability in real-world scenarios. In order
to distribute service-specific operations as events, the
proxy service has to interface with the ETDC.
3.3 Event Distribution
The Event Transformation and Distribution Core for-
wards incoming events to the storage component.
Events can originate from the local device, or from
other connected nodes. These are interconnected by a
peer-to-peer network. Events are the smallest unit of
data processed by the Net Rat since any functionality
of a service can be described as a series of events.
Figure 2: Event Tree.
The ETDC transforms service operations into
service-independent events, distributes, and replicates
them in-order. While transformation could also be in-
tegrated into proxies, it would then also need to be
integrated into gateways. With our approach, purely
client-side parts of proxy business logic can be re-
used for a variety of services using the same client-
applications. This does scale, as implementing event
transformation for a single protocol (such as HTTP,
for example) covers all services relying on this proto-
col. Consequently, scalability is much less of an issue
than it would seem at first glance.
Preconditions need to be met in order for each
event to be executable. By chaining events, the ex-
act flow of events can be replicated at any time. A
Net Rat event may have multiple successors, leading
to a structure similar to an unbalanced hash tree. By
relying on a tree-like event log, it is possible to start
multiple long-running events in parallel. Part of an
example event tree is depicted in Figure 2.
The Net Rat primarily aims for single-user scenar-
ios, where a user will typically only use one device at
a time. Therefore, conflict detection is not a prior-
ity. Instead, this basic event chaining and replication
ensures that events are executed in the correct order.
Detecting events lost during transmission is straight
forward, since gaps in the event tree can be detected
immediately.
With all of the Net Rat’s components in place,
events can be spread among all devices and users
can access event data using their client application of
choice. From the user’s point of view, it looks and
feels just like using a traditional service.
The Net Rat - Rethinking Connected Services for Increased Security
377
3.4 Connecting to the Outside World
In order to interface with external services and legacy
infrastructures, the Net Rat introduces a gateway. The
setup shown in Figure 1 uses a distinct device with
the sole purpose of connecting to the outside world.
No data is persisted and the only events produced on
the gateway device originate from external services.
These events are fed into the P2P network and are
therefore available at every client application. Only
those events from within the network addressed to ex-
ternal services are forwarded by the gateway.
Contrary to traditional client-server systems, the
Net Rat can even provide some functionality during
network downtimes, thanks to storing all data at end-
user devices.
3.5 Crypto-enhanced Storage
The storage is designed to write incoming event data
to disk. Incoming data is encrypted using public-
key cryptography and directly stored on disk while
accessing and decrypting data requires user-input in-
formation. This is necessary, since the private key
needed for decryption needs to be unwrapped. By re-
lying on key wrapping, no authentication information
needs to be stored. Ensuring transport security, on the
other hand, is up to the peer-to-peer network imple-
mentation.
The crypto-enhanced storage consists of the actual
storage and the following supporting building blocks:
A bulk encryption engine, a symmetric key encryp-
tion engine enabling key wrapping, and a key deriva-
tion function providing the secret used for key wrap-
ping.
During the setup of a Net Rat instance, the
following actions are performed for each device:
The bulk encryption engine creates a fresh key pair
(K
pub
, K
priv
, meta) (Eq. 1). Next, the user provides a
secret s in order to wrap the private key K
priv
follow-
ing Eq. 2. The public key K
pub
, the wrapped key wk,
and the key meta data meta are then stored and the
system is then ready for use.
(K
pub
, K
priv
, meta) KeyGen(rand, len) (1)
wk Enc(K
priv
, KDF(s)) (2)
Whenever data is fed into the storage, it is stored only
after being encrypted following Eq. 3. The resulting
ciphertext c is stored on disk for future use.
When the proxy service demands access to the
data, it has to be decrypted. First, the user secret s
is used to unwrap the wrapped key wk to recover the
private decryption key K
priv
(Eq. 4). Then, the de-
cryption can be performed following Eq. 5.
c Enc(d, K
pub
) (3)
K
priv
Dec(wk, KDF(s)) (4)
d Dec(c, K
priv
) (5)
Similar to traditional schemes, user authentication is
necessary prior to retrieving information. At the same
time, it is possible to notify the user about incoming
data since encryption only happens immediately be-
fore data is written to disk. Furthermore, no secrets
need to be stored and protected on disk. Receiving
data does not require user interaction. Yet, data is
enciphered before it is stored on disk. Therefore, an
attacker cannot gain access to any data even through
physical access to a device and its storage.
3.6 Summary
The Net Rat takes traditional client-server services
and provides them using a more secure, more robust,
and more flexible network of personal devices. By
using P2P networks, the Net Rat can operate in diffi-
cult network situations while drastically reducing the
attack surface and therefore bringing more security
and privacy to the user. By distributing data to all of
a user’s multiple devices, users experience increased
availability. In addition, the Net Rat’s foundation can
be used to develop new services, fully utilising the
connected nature of end-user devices.
4 CASE STUDY: E-MAIL
We evaluated the utility of our approach under real-
world conditions by building a decentralised e-mail
service on top of the Net Rat. From a technical point
of view, we adapted and split traditional Mail Transfer
Agents (MTAs) and Mail Delivery Agents (MDAs) to
work as gateways and proxy services. Users can thus
continue to use the Mail User Agents (MUAs) they are
familiar with and benefit from the Net Rat’s features.
Figure 3: Incoming Mail Processing at the Gateway.
Internally, during system setup, our prototype
creates an EC key pair for use with the Elliptic
SECRYPT 2017 - 14th International Conference on Security and Cryptography
378
Figure 4: Mail Replication at the Client.
Curve Integrated Encryption Scheme (ECIES). We
use the Password-Based Key Derivation Function
2 (PBKDF2) as a KDF for the key-wrapping key.
For key wrapping itself we rely on AES/GCM. Fur-
thermore, we defined IMAP and SMTP commands
for sending, fetching, and managing mail as ETDC
events.
In essence, users only connect to local services
(the proxy services) and the Net Rat abstracts away
distribution, synchronisation and message delivery
just as an ordinary e-mail server would. The ma-
jor difference here is that no central instances are
involved. Whenever a message addressed to a lo-
cal recipient arrives at the gateway-enabled device,
the message registers as an event and is distributed
through the P2P network as shown in Figure 3. When-
ever such an event is received by another device it
is replicated and transformed back into the original
message as shown in Figure 4. Only fetching mails
requires for the user to authenticate using an IMAP
password, just as it is the case with most traditional
IMAP servers.
5 SECURITY EVALUATION
The Net Rat is built on a solid foundation following
a security-first approach. Its primary goal is to pro-
vide increased data security through decentralisation.
To show that this was achieved, we performed a se-
curity analysis loosely based on the Common Criteria
for Information Technology Security Evaluation (In-
ternational Organization for Standardization, 2014).
Due to the flexibility of the Net Rat and the different
configurations it can be deployed in, we picked two
distinct scenarios for evaluation.
5.1 Threat Model and Assets
We based our threat model on traditional client-server
applications typically serving multi-device users. On
the one hand, successfully attacking end-user devices
can lead to extracting a huge amount of very personal
single-user data. Gaining access to central databases
of the service back-ends, on the other hand, usu-
ally reveals data of many users at once. Therefore,
we differentiate between low-volume targeted attacks
and high-volume bulk data extraction attacks. The
Net Rat is not concerned with protecting users from
targeted attacks, since well-crafted scams cannot be
countered by technology alone.
The primary asset to be protected is the data pro-
cessed by services (A1) running on top of the Net Rat.
All service-related user data stored on disk is consid-
ered sensitive and must therefore be protected. The
second significant asset which must also be kept se-
cret is the authentication information provided by the
user (A2). Internally, different services handle data in
a multitude of ways, possibly requiring different lev-
els of protection. However, the conservative assump-
tion that all user data and all authentication informa-
tion needs to be kept secret covers all cases.
5.2 General Assumptions
The main threats the Net Rat seeks to mitigate are
those targeted at back-ends. Even though any service
run on top of the Net Rat interfaces with the outside
world, we show how eliminating central storage can
effectively increase data security. We do not, how-
ever, deal with threats posed by attackers gaining root
access on end-user devices, as this is out of scope.
Due to memory isolation techniques employed by vir-
tually all modern operating systems, the Net Rat is
also not concerned with defending against threats in-
volving local attackers being able to read the memory
of other processes.
For the Net Rat to effectively provide its services
the following assumptions must hold:
AS1 Transport security is upheld by the P2P network
implementation.
AS2 Client applications only connect to proxies.
As a consequence of these assumptions, transport se-
curity becomes a non-issue, even if client applications
are not capable of securing connections.
AS3 The Net Rat runs in a sandboxed environment
provided by the isolation mechanisms present on
modern operating systems.
AS4 Client applications do not store authentication
information
Since users are free to continue using the client appli-
cations they are familiar with, evaluating the security
properties of such software is out of scope. In order to
evaluate the Net Rat’s potential in this context, it is as-
sumed that end-user applications do not store authen-
tication information. Due to (un-)wrapping decryp-
tion keys using authentication information, no au-
thentication secrets can be obtained by an attacker—
Asset A1 does not need to be protected as it is never
The Net Rat - Rethinking Connected Services for Increased Security
379
stored on disk.
Realistically, many users may choose to store authen-
tication information through their client applications.
However, scenarios violating Assumption AS4 are
considered out of scope.
AS5 Authenticated encryption schemes are used
wherever applicable.
These assumptions lead to some profound ramifi-
cations. An immediate consequence is the fact that lo-
cal attacks become the predominant threats to worry
about. However, as long as Assumption AS3 is up-
held, local attackers need to either overcome OS-level
isolation mechanisms or compromise the client appli-
cation used to interface with the local proxies—both
of which is out of scope.
Assumption AS4 is somewhat relaxed: Storing
authentication information in the client application
poses a threat to Asset A2, however, this has the same
consequences as in a traditional, centralised setup.
Therefore, the following corollary can be derived:
Corollary 1 Using the Net Rat with client software,
which stores authentication information, has no more
severe consequences than using the same client soft-
ware with an unmodified service.
In case Assumption AS4 is upheld, stronger security
properties can be observed.
5.3 Scenario 1: Baseline
This baseline scenario does not deal with the deploy-
ment details of the Net Rat. It is concerned with as-
pects independent of the setup:
S1 An attacker gains read/write access to the storage
of an end-user device running the Net Rat.
Taking into account the fact that data is only stored
encrypted, decrypted only on-demand, never per-
sisted in plain, and no authentication information is
stored anywhere, the following corollary can be de-
rived:
Corollary 2 As long as client applications do not
store authentication information, even an attacker
gaining full access to the device’s storage medium can
extract neither user data (Asset A1), nor authentica-
tion information (Asset A2).
Combining these observations with Assumption AS5
leads to an even stronger security guarantee:
Corollary 3 An attacker having full access to the stor-
age medium used by the Net Rat can at most carry
out denial-of-service attacks and delete existing data
without learning anything about its contents.
Due to the Net Rat’s flexibility, we have analysed two
concrete scenarios reflecting different use cases.
5.4 Scenario 2: Using Gateway-only
Devices
This scenario maps to the setup depicted in Figure 1.
It features two categories of devices configured dif-
ferently: End-user devices featuring local storage and
proxy services and gateway devices connecting to the
outside world and legacy services.
At first, having a single entity interfacing with the In-
ternet does not seem to differ much from traditional
client-server setups. In fact, many of the same threats
apply. The consequences of attacks, however, differ
significantly. Even successful attacks on the Net Rat
can only lead to data consumed and produced by a
single user. This scenario is defined as follows:
S2 An attacker compromises the gateway device and
has complete control of it, including full access to all
data passing through.
As no more sensitive data is processed on a gateway
device than on a traditional server, another corollary
can be derived:
Corollary 4 An attacker controlling a gateway device
can do no more harm than an attacker controlling a
traditional server.
However, due to the fact that no data is persisted at
a gateway device existing data cannot be extracted,
which leads to the following, stronger corollary:
Corollary 5 An attacker controlling a gateway device
can at the very most access data created after the at-
tack. Previously created data is out of reach.
The simple fact that data is only stored at end-user
devices clearly improves data security even without
fully decentralising a service. Of course, using a static
gateway device leads to some of the same weaknesses
present in traditional client-server setups. However,
such limitations are not inherent to the Net Rat, but
at most mandated to ensure full backwards compati-
bility. When using the Net Rat with modern services
allowing for dynamic gateways, the system’s proper-
ties change as described in Scenario 3.
5.5 Scenario 3: Symmetric Setup
This scenario consists only of devices featuring all of
the Net Rat’s components. No static or central gate-
way is present. Consequently, each device is interfac-
ing with the outside world and becomes a valid target
for attacks. More formally, we identified the follow-
ing attack scenario which shows the consequences of
fully decentralising a service and transforming end-
user devices into servers:
S3 An attacker gains control of an end-user device
and has full control of the Net Rat instance running
on this device.
SECRYPT 2017 - 14th International Conference on Security and Cryptography
380
When considering a traditional client-server setup,
Scenario S3 can be reduced to an attacker gaining ac-
cess to a client application in a client-server scenario.
This leads to the following corollary:
Corollary 6 From an individual user’s point of view,
an attacker gaining control of a Net Rat instance in
the context of a symmetric setup can do no more harm
than an attacker controlling a client application inter-
facing with a traditional server.
It is therefore reasonable to assume that attacks on
the Net Rat are infeasible, when it is also possible to
attack (unmodified) client applications. In addition,
as each Net Rat deployment only processes data of a
single user, another observation can be made, which
illustrates how our approach protects users from col-
lateral damage:
Corollary 7 Compromising a Net Rat deployment
only affects the data consumed and produced by a sin-
gle user.
As a consequence, other users do not suffer collateral
damage, when a service run on top of the Net Rat
gets compromised. Compared to centralised services
housing the data of potentially millions of users, the
impact of an unauthorised access is minimal.
5.6 Summary
Given a realistic environment and assumptions based
on real-life observations, the Net Rat matches the se-
curity of client-server setups (see Corollaries 1, 4, and
6). When taking into account the encryption mecha-
nisms incorporated by the Net Rat, data is protected
better compared to traditional systems. Neither user
data nor authentication information can be extracted
by an attacker gaining full disk access (see Corollar-
ies 2 and 3). The Net Rat also solves transport secu-
rity issues between client application and service as
users only connect to local proxies. This way, clients
not capable of securing data in transit do not prevent
using state-of-the-art transport security. Furthermore,
when deploying dedicated gateway-only devices, it is
impossible for an attacker to extract any information
previously processed form the gateway as discussed
in Corollary 5. Finally, we showed how using the Net
Rat over a traditional service can directly prevent col-
lateral damage.
6 RELATED WORK
The Net Rat uses mesh-like structures to organ-
ise users’ devices by making heavy use of the P2P
paradigm. Network-layer approaches to this prob-
lem have already been presented and refined over
the past decades. Distributed hash tables (DHTs)
like Chord (Stoica et al., 2001), and Kademlia (May-
mounkov and Mazi
`
eres, 2002), for example, provide
P2P connections. However, their primary goal is
a distributed key-value storage. Typically, a shared
global storage is created without control over which
nodes are in charge of storing particular pieces of
data. The Net Rat, however, requires fine-grained
control over which data is routed where.
Another category of related work consists of
designs which primarily aim at providing an
application-independent network layer. Anonymity
networks such as Tor (Dingledine et al., 2004) and the
I2P
4
are examples of concrete implementations of the
concept with a strong focus on keeping users’ traffic
private.
Concepts and applications more closely related to
what the Net Rat accomplishes include Bitmessage
5
,
and FlowingMail
6
. While these approaches imple-
ment decentralised communication frameworks, com-
patibility with legacy applications is simply not aimed
for. A framework for decentralising legacy applica-
tions has been proposed by (Hautakorpi et al., 2009).
However, it still tries to establish a closed system
without supporting a way of integrating the approach
with outside legacy infrastructure. Furthermore, be-
ing conceived in 2009, it does not reflect today’s
multi-device users and the capabilities of current de-
vices.
7 CONCLUSIONS AND
OUTLOOK
In this work, we introduced the Net Rat to transform
existing client-server setups into decentralised appli-
cations for increased security and availability. Our ap-
proach achieves backwards compatibility both to ex-
ternal legacy services and towards end-user applica-
tions. This way, it is possible for users to continue us-
ing familiar applications. Due to the modularity and
flexibility of our framework, it is possible to deploy
it in a heterogeneous set of devices including some-
what static servers and highly mobile smartphones.
In short, virtually any service can be transformed to
work on top of the Net Rat as long as its operations
can be represented as a series of events. In addition,
our framework can be used as a base for new ser-
vices which can be designed with the Net Rat’s fea-
tures in mind. If, for example, two users deploy a de-
4
https://geti2p.net/en/
5
https://bitmessage.org/
6
http://flowingmail.com/
The Net Rat - Rethinking Connected Services for Increased Security
381
centralised e-mail service on their devices, it is pos-
sible to devise next-generation, Net-Rat-aware gate-
ways without having to rely on legacy infrastructure.
An important consideration when transforming
applications to end-user devices are security aspects.
Firstly, users are effectively burdened with the re-
sponsibility of being service operators.This raises the
second important issue: Maintaining the integrity of
every user’s devices is imperative. Therefore, ways
of detecting Byzantine faults (Lamport et al., 1982)
also need to be considered. Approaches such as Peer-
Review (Haeberlen et al., 2007) deal with this issue,
however, the problem definition resulting from the
Net Rat’s architecture is simpler: Most importantly,
the impact of a compromised Net Rat instance is com-
parable to that of a single rogue user in a traditional,
centralised setup.
The security evaluation results illustrate how the
Net Rat achieves its security properties and how these
differ based on the chosen deployment. As our system
enforces encrypted storage and never stores authenti-
cation information, even attackers with full disk ac-
cess cannot extract any information. In combination
with a robust network layer taking care of transport
security, the Net Rat indeed provides increased secu-
rity compared to traditional services even for worst-
case scenarios. Most importantly, innocent users do
not suffer collateral damage from infrastructures out-
side their control being compromised. Furthermore,
due to all data being directly distributed among de-
vices, availability can be guaranteed to some degree
even during network downtimes.
Device enrolment has not been considered. How-
ever, leaving this issue open for future work does not
impact the properties of our system, as soon as de-
vices have been connected. In essence, enrolment is a
separate issue. When examining single, multi-device-
user scenarios, the burden of performing sophisticated
enrolment procedures does not outweigh its benefits.
Another area of future work includes supporting user
authentication methods which do not rely on a shared
secret like client certificates, or the Qualified Mobile
Server Signature (Orthacker et al., 2010).
The final area of future work concerns a more sophis-
ticated distributed storage. It must be possible to cater
to memory heterogeneity within a set of connected
devices for the Net Rat to be applicable to a wide
range. However, due to the fact that events are the
smallest unit of information, this issue is definitely
within reach. After all, event replication already lies
at the core of the Net Rat’s workflow. Therefore,
memory heterogeneity can be catered to by manag-
ing events in a clever way. By tweaking the ETDC,
this issue can be solved separately from the Net Rat’s
core functionality much like device enrolment.
The results of our case study and the outcome
of the security evaluation clearly show the feasibility
of decentralising services. This highlights how exist-
ing services can be improved through decentralisation
and presents opportunities to develop novel services
based on the solid foundation provided by the Net Rat.
REFERENCES
Blum-Dumontet, E. (2017). Defeating encryption:
the battle of governments against their people.
https://www.privacyinternational.org/node/1427, Re-
trieved: 2017-04-24.
Cohen, B. (2013). The BitTorrent Protocol Specifica-
tion. http://www.bittorrent.org/beps/bep 0003.html,
Retrieved: 2017-04-24.
Dingledine, R., Mathewson, N., and Syverson, P. (2004).
Tor: The Second-Generation Onion Router. In Pro-
ceedings of the 13th USENIX Security Symposium.
Dropbox, Inc. (2016). Dropbox Privacy Policy.
https://www.dropbox.com/privacy/, Retrieved: 2016-
09-27.
Fiadino, P., Schiavone, M., and Casas, P. (2015). Vivisect-
ing WhatsApp in Cellular Networks: Servers, Flows,
and Quality of Experience. In Traffic Monitoring and
Analysis: 7th International Workshop, pages 49–63.
Springer International Publishing, Cham.
Google, Inc. (2014). Google Terms of Service.
https://www.google.com/intl/en/policies/terms/, Re-
trieved: 2017-04-24.
Haeberlen, A., Kouznetsov, P., and Druschel, P. (2007).
PeerReview: Practical Accountability for Distributed
Systems. In Proceedings of Twenty-first ACM
SIGOPS Symposium on Operating Systems Princi-
ples, SOSP ’07, pages 175–188, New York, NY, USA.
ACM.
Hautakorpi, J., Camarillo, G., and Lopez, D. (2009). Frame-
work for Decentralizing Legacy Applications. In Pro-
ceedings of the 2009 9th IEEE/ACM International
Symposium on Cluster Computing and the Grid, pages
544–549, Washington, DC, USA. IEEE Computer So-
ciety.
International Organization for Standardization (2014).
ISO/IEC 15408-1:2008 Information technology
Security techniques Evaluation criteria for IT se-
curity — Part 1: Introduction and general model.
Lamport, L., Shostak, R., and Pease, M. (1982). The Byzan-
tine generals problem. ACM Transactions on Pro-
gramming Languages and Systems, 4(3):382–401.
Maymounkov, P. and Mazi
`
eres, D. (2002). Kademlia:
A peer-to-peer information system based on the xor
metric. In Revised Papers from the First Interna-
tional Workshop on Peer-to-Peer Systems, pages 53–
65, London, UK. Springer.
Orthacker, C., Centner, M., and Kittl, C. (2010). Qualified
Mobile Server Signature. In Security and Privacy
SECRYPT 2017 - 14th International Conference on Security and Cryptography
382
Silver Linings in the Cloud, volume 330 of IFIP Ad-
vances in Information and Communication Technol-
ogy, pages 103–111. Springer, Berlin, Germany.
Rosenberg, J. (2010). Interactive Connectivity Establish-
ment (ICE): A Protocol for Network Address Transla-
tor (NAT) Traversal for Offer/Answer Protocols. RFC
5245, Internet Engineering Task Force Request for
Comments. http://www.rfc-editor.org/rfc/rfc5245.txt,
Retrieved: 2017-03-17.
Stoica, I., Morris, R., Karger, D., Kaashoek, M. F., and
Balakrishnan, H. (2001). Chord: A scalable peer-
to-peer lookup service for internet applications. In
Proceedings of the 2001 Conference on Applications,
Technologies, Architectures, and Protocols for Com-
puter Communications, pages 149–160, New York,
NY, USA. ACM.
The Net Rat - Rethinking Connected Services for Increased Security
383