Network-based Intrusion Prevention System Prototype
with Multi-Detection
A Position Paper
Daniel Kavan, Klára Škodová, and Martin Klíma
Department of Applied Research, CertiCon, a. s, Václavská 12, 120 00 Prague 2, Czech Republic
Keywords: Intrusion Prevention System, Intrusion Detection System, Network Behavior Analysis, Signature-based
Detection, CAMNEP, Snort, IDMEF, Iptables, Ipset, PCAP, Netflow, Traffic Generation.
Abstract: The ongoing need to protect key nodes of network infrastructure has been a pressing issue since the outburst
of modern Internet threats. This paper presents ideas on building a novel network-based intrusion prevention
system combining the advantages of different types of latest intrusion detection systems. Special attention is
also given to means of traffic data acquisition as well as security policy decision and enforcement
possibilities. With regard to recent trends in PaaS and SaaS, common deployment specific for private and
public cloud platforms is considered.
1 INTRODUCTION
The protection of local networks is a key goal for the
current network administrators. The cyber war is in
progress and it becomes still more and more difficult
to keep up with the modern threats. We are focusing
on network security, namely on new generation
intrusion prevention systems (IPS) that combine two
major approaches of intrusion detection.
To prevent an intrusion, it has to be identified
first, therefore the problem solution comprises of
two main parts – detection and prevention, together
forming an IDPS
1
. This paper describes a prototype
of such a system.
1.1 Motivation
There are two major branches of intrusion detection
systems (IDS). One builds on signature based detec-
tion, the second on behavior analysis. Each has
strengths and weaknesses and our challenge is to
combine them into one IPS. (Kazienko & Dorosz,
2003) and (Talawat, 2008) show that:
1
Intrusion detection and prevention system. The term and an
analysis of the principles used are covered by (Scarfone &
Mell, 2007).
Signature-based detection works reliably,
but cannot respond to unknown threats.
Behavior analyzers and anomaly detectors
can reveal previously unseen problems, but
they do not work as fast as signatures and
for known threats they may guess what has
already been proven (signatures exist).
Bare deployment of detectors of both types
would provide a fair chance to capture problematic
behavior from both categories. (Gómez, et al., 2009)
demonstrates it is not uncommon to combine known
detection approaches to achieve better results.
However, our ambitions go beyond such simple
solution by incorporating different detection tech-
niques, consuming the knowledge they provide
about the traffic, and attempting to deduce further
information based on the combination of data
evaluation. Such combination can yield additional
knowledge about actual threats and means to react.
1.2 Objectives
Our aim is to create an intrusion prevention system
composed of complementary detection modules.
Such system should function in regular (enterprise)
networks as well as private/public cloud environ-
ments.
Our solution is based on sets of rules to be en-
forced within the network environment.
417
Kavan D., Škodová K. and Klíma M..
Network-based Intrusion Prevention System Prototype with Multi-Detection - A Position Paper.
DOI: 10.5220/0005102204170425
In Proceedings of the 11th International Conference on Security and Cryptography (SECRYPT-2014), pages 417-425
ISBN: 978-989-758-045-1
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Figure 1: Protected segment schema.
An important objective was to research ways to use
the derived enforcement actions, considering the
limits of particular enforcement-point platforms.
We want to combine the detection capabilities of
IDS and based on this, to introduce the prevention,
e.g. to enforce network traffic shaping.
We strive to demonstrate that the additional logic
added to the used detection modules will yield bene-
fits compared to simple combined deployment of
detectors unaware of each other.
2 RELATED WORK
We build on existing ideas from fields of both signa-
ture-based and behavior detection. The advantages
and drawbacks of both these approaches are outlined
in Section 1.1. For our prototype purposes, these two
groups are represented by Snort and CAMNEP,
respectively.
(Roesch, 1999) published an initial form of his
lightweight NIDS Snort. Since then the design was
improved heavily while the key concept was kept –
network traffic is checked against a database of
known signatures in order to identify a threat.
(Rehák, et al., 2008) introduced the project
CAMNEP as a representative of behavior network
intrusion detection system (A-NIDS) strongly
relying on agent technology based on A-Globe
platform (Šišlák, et al., 2005).
According to the taxonomy by (García-Teodoro,
et al., 2009), CAMNEP can be described as statisti-
cal and machine-learning-based detector featuring
fuzzy logic techniques, clustering and outlier detec-
tion. (Lim & Jones, 2008) would identify CAMNEP
as ‘learned-model-based’ as it automatically creates
its model of normal network behavior.
Even though CAMNEP is based on previously
described detection techniques, due to the employ-
ment of agent-based collective trust aggregation or
history integration it is able to function with signifi-
cantly lower rate of false positives than common
anomaly detectors, (Rehák, et al., 2008).
3 PROPOSED SOLUTION
An intrusion prevention system consists of three
basic modules: a sensor providing the input data ac-
quisition, a data processor evaluating the input and
deciding about actions to be performed, and an
executor responsible for the results enforcement.
For the data evaluation, we choose a combina-
tion of signature-based detection tool Snort (that can
work as a simple IPS as well) and CAMNEP, NBA
2
anomaly detector developed by Cognitive Security /
Cisco Systems. Conceptually, other similar tools can
be used; specific conditions when and how this can
be done are described in section 5.1.
The initial IPS prototype relied on network beha-
vior analysis provided by CAMNEP based on the
traffic coming from a NetFlow
3
sensor
4
. Following
engagement of Snort's output allowed building a
versatile network-based intrusion prevention system.
Figure 1 depicts the resulting schema: Traffic
coming from the external network enters the protect-
ed network via a gateway where the passing data can
be collected for an analysis within the NIPS. Various
aspects that can affect traffic data gathering by sen-
sors are introduced in section 3.2. A behavioral and
a signature-based detection system are engaged to
compensate weaknesses of each other as discussed
in section 3.1. Section 3.4 explains how alerts
2
network behavior analysis
3
metadata format described by (B. Claise, 2004)
4
also known as „probe“
SECRYPT2014-InternationalConferenceonSecurityandCryptography
418
exported from both IDS are processed by the IPS
core (configurable via a web service). The resulting
actions are enforced by a specialized module at the
gateway as described in section 3.5. Additionally,
also other ways of intrusion prevention can be
employed (see section 3.3).
Event data from detectors are preferably ob-
tained as XML files in standard Intrusion Detection
Message Exchange Format (IDMEF)
5
supported by
the CAMNEP IDS. Alternative security alert retriev-
al methods can be used – e.g. IDS Snort alerts are
obtained via shared MySQL database access.
3.1 Symbiosis of Signature and
Behavior Detection
The NBA-only IDS core is not sufficient on its own
to provide reliable information for actions to control
the network traffic. Many known attacks do not
display any statistical anomalies and therefore can-
not be revealed by a behavior detector. Moreover,
even a very specific output of an NBA analyzer
points always to only a suspicious behavior and
without information context, significant risk of false
positive detection and thus an encroachment might
occur. False alarms are considered even more harm-
ful than false negatives (especially when used for
automated responses).
6
However, the necessary context information
related to the detected event can be delivered to the
IPS by the other detector, the signature IDS core.
The advantage of the signature approach is its
promptness allowing real-time interventions, high
specificity and full sensitivity
7
. These qualities are
redeemed by the dependency on the signatures de-
veloper and frequent regular database updates. In
comparison, the behavior analysis is not only com-
putationally expensive but it also needs a time to
collect traffic flow sets and to compare the observed
data with a previously built statistical model (further
referred to as “model”). CAMNEP is capable to
provide output in 5min intervals at the best, which
can become critical for the intrusion prevention.
A considerable property of the NBA is concealed
in its strength – the ability of the model to learn over
time can cause a creeping adaptation to an anoma-
lous traffic that becomes a part of the norm.
5
standardized by (Debar, et al., 2007) in RFC 4765
6
this is the prevailing opinion supported not only by (Rehák, et
al., 2008)
7
This is true for all threats previously described in the signature
database. Sensitivity is also known as recall.
In spite of all its drawbacks, the NBA proves to
be irreplaceable in case of long-term enterprise
network issues, important data leaks or suspicious
events where no signature detectors can discover any
problem. Employment of an NBA-based detector
usually assumes a human-expert participation to
investigate the conditions and circumstances. Such
manual intervention could be replaced by a rule
system based on the previous knowledge in order to
provide the protective decision supporting informa-
tion.
Symbiosis of both detection types (compared in
Table 1) leads to an increase of overall capabilities
of the newly developed IPS.
Table 1: Behavior and signature analysis comparison.
Property NBA Signature-
based
determinism anomaly
detection
reliable attack
detection
specificity suspicious
activity
precise threat
identification
efficiency time-
expensive
real-time
autonomy independent updates
necessary
adaptability ability to learn
new threats
only known
signatures
3.2 Traffic Data Acquisition
All analyzers are directly dependent on the incoming
data extracted from the network traffic using a sen-
sor. The sensor must either be placed in the course
of the data packets (an in-line sensor) or be sent a
copy.
An example is a network tap or a spanning port
8
directing all data coming through the network device
also towards the listening hardware or software sen-
sors. However, this solution requires a special hard-
ware that may not be available in the target network.
Another option for the information gathering is
offered by the network-card promiscuous mode that
enables “sniffing” of packets destined for other
nodes in the local network. This way, both the inner
traffic and the communication with external hosts
can be evaluated. Nevertheless, many architectures
do not allow promiscuous mode for security reasons.
8
differences between these two are illustrated in (Network
Instruments, 2013)
Network-basedIntrusionPreventionSystemPrototypewithMulti-Detection-APositionPaper
419
Alternative means of data acquisition can be
provided by promiscuity emulation. The iptables
9
tool on the entry node of the observed network seg-
ment can be instructed to duplicate all the incoming
traffic and forward it to a specified host via the non-
terminating TEE target. This solution assumes
administrative privileges in the gateway.
An important decision to be made about the
position of the sensor is related to the network ad-
dress translation (NAT). If NAT is performed on the
entry point, placing the sensor in front of it can lead
to a considerable loss of available data as the pro-
tected nodes are not individually observable and the
internal communication is not visible at all. On the
other hand, gathering data from inside of the protect-
ed segment causes missing the information about the
traffic already blocked on the gateway. The previ-
ously mentioned iptables TEE target at the very
beginning of the forward chain of the mangle table
is the most appropriate solution (see Figure 2).
Figure 2: Simplified iptables traversal schema.
3.3 Dual Usage of Snort
Snort can be used either purely as an intrusion-de-
tection product or also its prevention capabilities can
be leveraged.
An active sensor, located directly in the flow of
all data incoming and outgoing to/from the protected
network, enables immediate control over the passing
packets. The control actions include dropping, band-
width limiting, and modification in form of normali-
zation or sanitation. However, severe bottleneck is-
9
a user-space application for Linux allowing kernel firewall
management
sues are imminent using the in-line sensor at inten-
sive traffic load. When the sensor is flooded with
data and not able to accomplish their processing in
real-time, high network latency or even loss of pack-
ets may occur or unchecked flows may pass through.
In our IPS prototype, a trade-off between such
unintentional traffic limiting and a delayed interven-
tion caused by the analyzer overhead and by internal
processing was inevitable. We use an active sensor
for the fast signature-based Snort in-line enforce-
ment combined with passive collecting and/or
thorough examination of data by both IDS modules
(CAMNEP and Snort).
The optional protection feature of the Snort de-
tector is available in several data acquisition modes
(DAQ). The “NFQ”
10
was chosen as the most appro-
priate one for our purposes because it collaborates
with the iptables allowing for user-space packets
processing
11
. The active Snort sensor provides fast
and reliable signature-based decision on blocking
the malicious packet. In order to prevent unneces-
sary workload, this step can be prepended to the
more complex evaluation by the passively listening
analyzers controlled by the IPS-configuration rules.
A way to achieve this setup is enqueuing the Snort’s
NFQ into the forward chain of the mangle table
while placing the enforcement of the IPS system in
the forward chain of the filter table – see Figure 2.
3.4 Prevention Logic
The prototype was constructed as a rule system con-
sisting of dynamic and static rules (see Figure 3).
Dynamic rules sequentially process incoming events
(alerts from IDS of various types, e.g. “port scan”,
“DNS tunnel”, “attempted reconnaissance”) and
respond with primary actions (traffic limiting and
modification) and supplementary secondary actions
(logging or user notification). Static rules setting is
provided by whitelist and blacklist definitions and a
default policy (packets “deny” or “allow”).
The 3-dimensional space of network connection
endpoints (see Figure 4) is described by “tokens”,
the substituting symbols for a range of IP addresses,
set of L3/L4 protocols and a range of port numbers
in case the protocol supports ports. As an example,
10
an inline module for Linux leveraging the QUEUE target in
netfilter to move packets from the kernel-space to user-space
for evaluation – see (Snort Team, 2013)
11
(Leblond, 2013) refers to this act as “issuing a verdict”.
SECRYPT2014-InternationalConferenceonSecurityandCryptography
420
Figure 3: Event processing into actions.
the depicted token t1 defines a set of hosts within the
IP range ip1ip2 communicating via the UDP proto-
col on port p1; the token t2 defines a node with the
ip3 address at the TCP protocol on the port p2. Such
tokens are used to describe the hosts or host groups
identified as carrying on the suspicious communica-
tion reported in an IDS alert. Another application of
tokens is to determine the enforcement-action target
as explained later.
Figure 4: 3D network space described by tokens.
The “rule-engine” uses a set of dynamic rules. A
rule includes one or more triggers and a single
enforcement action. The enforcement action, such as
packet dropping, rejecting, or limiting, is applied to
a set of tokens provided by the rule after its activa-
tion. So e.g., a “drop” action applied to the t2 token
from Figure 4 directs the gateway to drop all packets
from or to a host with the ip3 address sent via the
TCP protocol on the host’s port p2.
A “trigger” can be thought of as a definition of
the rule’s area of interest and if activated, it provides
output tokens that are eventually used for the rule’s
enforcement action as documented in Listing 1.
process_alert(alert_type,alert_tokens)
foreachrule
foreachtrigger
iftrigger_regular
ifalert_typetrigger_types
 trigger_ignore_alert//continue
trigger_output
:=alert_tokenstrigger_tokens
iftrigger_outputempty
 trigger_ignore_alert//continue
activate_trigger
iftrigger_rule
ifnotdependee_rule_acivated
 trigger_ignore_alert//continue
trigger_output
:=get_tokens_from_dependee_rule
activate_trigger
if
notall_triggers_activated
rule_ignore_alert//continue
enforcement_action_input
:=
all_triggers
trigger_output
submit_enforcement_action
Listing 1: General incoming alert processing.
A regular trigger consists of a set of alert types
and a set of tokens that the trigger should react to if
a corresponding alert occurs. Such trigger is activat-
ed by an incoming alert if the alert type matches a
type defined within the trigger and if the intersection
of both the alert’s and the trigger’s tokens is not
empty; this intersection constitutes the activated trig-
ger’s output.
A special kind of trigger, a “ruletrigger”, repre-
sents a link to another rule, so (de)activation of the
referenced rule implies (de)activation of such rule-
trigger. The presence of ruletriggers allows creating
trees of dependent rules.
In order for the rule enforcement action to be
applied, all triggers within the rule must be activat-
ed. In case of activation of a trivial rule (with only
one trigger), the enforcement action is applied on the
network space defined by the trigger’s output. In
case of a more complex rule (with multiple triggers),
the logic works alike – the individual triggers’ out-
puts are further intersected with each other and only
the result is applied via enforcement.
A rule with only some triggers activated is in a
partially excited state, which is remembered for a
defined time interval. If also the rest of triggers get
activated within this period then the whole rule is ac-
tivated; otherwise the activation of triggers expires.
As a demonstration of this mechanism, consider
a rule with two triggers defined by the “port scan”
and “reconnaissance” alert type, respectively, and a
common IP range 198.51.100.0/24. The former trig-
ger is activated by an incoming “port scan” alert
identified in the traffic from the 198.51.100.101 and
198.51.100.102 addresses. A following “reconnais-
sance” alert warning against the 198.51.100.102 and
198.51.100.103 hosts activates the latter trigger.
IPaddress
p
ort
protocol
255.255.255.255
65535
0
0.0.0.0
UDP
TCP
t1
t2
ip1 ip2
ip3
p2
p1
Network-basedIntrusionPreventionSystemPrototypewithMulti-Detection-APositionPaper
421
This causes activation of the whole rule. The activat-
ed rule uses the token intersection of both the
triggers’ outputs, i.e. the 198.51.100.102 IP address,
to enforce the “reject” action on the gateway of the
protected network.
For cases when the rule’s action is being already
enforced, the system configuration allows an exten-
sion of the rule impact. In this way, other incoming
relevant alerts can cause (a) an extension of the set
of tokens, subject to the enforcement action, by a
newly detected attacker address, (b) action amplifi-
cation or action duration extension (e.g. packets
being rejected by now become being dropped), or (c)
a combination of both.
3.5 Enforcement
Network nodes protection can be reached by dif-
ferent means depending on the point of intervention
in the data flow or the level of ISO/OSI model
12
to
be affected. Network-based protection constitutes
the only available solution for securing hosts of an
undefined platform, which the protected nodes do
not need to know of at all.
Access to L3/L4 (our target level) can be
configured to a great extent, thus it is possible by
various administration tools. Probably the most well-
known of these tools is iptables which has also been
employed within this IPS.
Network protection settings must be modified
atomically in order to avoid exposing the internal
network to the outside world while not missing any
packets. Iptables itself is capable of such behavior,
but a reasonable and common practice is using the
ipset
13
tool in cooperation with iptables to simplify
IP addresses and/or ports definitions. For such com-
bination, a specific approach has to be applied to
preserve the atomicity of the key operation. This
approach is shown in Figure 5:
12
defined in (International Organization for Standardization,
1996)
13
user-space tool for organizing (mostly) IP addresses for
efficient lookup, integrated with iptables
Figure 5: Atomic iptables and ipsets workaround.
The newly needed ipsets (thought of as “red” in
an alternating red-black sequence) have to be created
first (non-atomically one by one as they are not used
yet – state transition 1-2). Then the iptables chains
referencing the old ipsets (called “black”) have to be
flushed and simultaneously the new chains with re-
ferences to the new red ipsets imported (which hap-
pens atomically via the “iptables-restore” command;
2-3). The old black ipsets can be non-atomically
deleted after that as they are no longer referenced
and used (3-4). In the next enforcement turn, new
black sets can be created and applied analogically.
4 EXPERIMENTS
The prototype behavior has been evaluated in seve-
ral environments emulating different network setups
from cloud to a local deployment in a small to
middle-sized enterprise network. Experiments were
performed in a dedicated network segment repre-
senting a set of running services in cloud or in a cor-
porate network (possibly a demilitarized zone). Ex-
perimental intrusion attempts were conducted from
the outside of NAT to simulate an external attacker.
All traffic for this model was generated artificial-
ly. The traffic consisted of generated attacks and ar-
tificial background. Legitimate UDP communication
between individual hosts was simulated with the
RUDE and CRUDE
14
tools. A source for massive
traffic to camouflage malicious actions was the Li-
nux netcat utility. Netcat CLI was used to emulate a
web server, to run as a port scanner or to be used as
an HTTP client on both UDP and TCP protocols.
15
The IDPS-testing framework Pytbull
16
provided
prepared experiments covering a range of paramete-
rized attacks (its modules comprise fragmented
packets, evasion techniques, shell codes, denial of
service and other types of exploits). We also used
the Python API to extend the tests and adjust them to
our needs.
Hydra
17
was used for multi-threaded cracking of
various services authentication. A DoS attack effect
was reached by flooding the target using LOIC
18
, the
14
http://rude.sourceforge.net, “Collector for RUDE”, resp.
15
http://www.howtoforge.com/useful-uses-of-netcat
16
http://pytbull.sourceforge.net
17
http://thc.org/thc-hydra
18
http://sourceforge.net/projects/loic
SECRYPT2014-InternationalConferenceonSecurityandCryptography
422
ping tool, hping
19
or by target resource exhaustion
with Slowloris
20
). Network reconnaissance attempts
including horizontal and vertical scanning were
accomplished by the Nmap security scanner
21
.
The tcpdump
22
and tcpreplay
23
provided a way of
capturing and emitting network traffic in the form of
PCAP files. Contagio Malware Dump
24
served as a
public source of current intrusion traffic recordings
for our experiments.
Finally, two penetration-testing Linux distribu-
tions, BackTrack and Kali, were engaged for ad-
vanced simulation of threats.
A different setup was designed for real-world
traffic observations – e.g. a network tap placed on
our company’s outer gateway to the Internet.
4.1 Testing Results
Using Pytbull test-sets, that are close to real attack
schemas, we were able to induce all possible combi-
nations of our analyzers detection, i.e. situations
where only signatures indicate a threat, where only
behavior analysis signals an alert, and where both or
none of them succeed. This result demonstrates the
added value of the detection methods combination.
The Hydra experiments confirmed that a massive
clutter with legitimate connection requests on a ser-
vice could not be caught by signature detector while
its quantitative anomality and time distribution allow
safe identification of an attack. The DoS and Nmap
related tests confirmed shortcomings of a signature-
based approach as well.
Data obtained from the Contagio Malware Dump
database split up into an undetected part, almost
52 % detected by Snort, and roughly 9 % which both
types of IDS warned about. Thus for the malware
communication typically displaying statistical insig-
nificance of traffic features, the necessity of the
signature-based detection was demonstrated.
5 CONCLUSION
This paper presents ideas and technological methods
for building a network-based intrusion prevention
system with an ample potential.
19
http://hping.org
20
http://ha.ckers.org/slowloris
21
http://nmap.org
22
http://www.tcpdump.org
23
http://tcpreplay.synfin.net
24
Security database; http://contagiodump.blogspot.com
Approaches to employ multiple detection en-
gines have been outlined and prototyped, with
emphasis on signature-based and anomaly-based
(NBA) detection composition, specifically using
Snort and CAMNEP, respectively. Projected bene-
fits of such arrangement have been explained in the
manner where the resulting system should be able to
protect key nodes or subnets against threats that
neither of the detection subsystems would reliably
react to on their own.
Configuration logic of the prototype provides
means for modeling of arbitrary dependency trees
including alerts combined from the different detec-
tion subsystems, their evaluation and enforcement
actions for entry-point firewall application.
At the moment, enforcement can be performed
on Linux gateways via iptables and ipset which safe
ways of operating have been found for.
5.1 Arbitrary Detection Integration
During the design and implementation phases of the
project, we have always had detection extensibility
in mind. We chose IDMEF (as discussed in section
3) to be the cornerstone exchange format as it is de-
signed exactly for this purpose – to exchange intru-
sion detection information – and serves this function
better than its counterparts (e.g. in comparison with
CEF
25
).
New detection modules can be used to enhance
capabilities of our prototype and depending on their
output details (whether they implement IDMEF and
how), changes to the prototype, needed to be done in
order to integrate them, may even be little to none.
Employment of an IDS using different export
structure and/or format should be possible as well,
but the complexity of such integration process can
vary dramatically. A successful proof-of-concept of
such approach is the Snort integration we have con-
ducted as described in section 3. In that case, the
prototype is able to read intrusion event data from a
shared database.
When the detection composition schema is
changed (because of a new detection unit), the IPS
configuration must be brought up-to-date. Configu-
ration of the complete system needs to respect new
types of events the joint IDSs produce. The XML-
based rules configuration is extendable, hence this is
possible. However, the logic of the events should be
revised and tested in order for the combination of
25
Common Event Format (ArcSight, Inc., 2009)
Network-basedIntrusionPreventionSystemPrototypewithMulti-Detection-APositionPaper
423
chosen detection subsystems to interact sensibly and
offer practical benefits.
5.2 Future Work
There are matters that need further work and re-
search. The prototype is at the point of being further
tested. Experiments outlined in section 4 yielded
promising results regarding the combined detection
capabilities of the prototype. Real-world employ-
ment, result evaluation and tuning are a next logical
step along our efforts. We have to prove to what ex-
tent the solution resists against new, still unknown
attacks.
Current rulesets used for our testing deployments
need improving, as they may be too naive for real-
world malware and threats. Based on detection and
environment knowledge, extended sets of rules must
be developed for particular cases tailored to given
production environments to fully utilize the project-
ed advantages of these and other detection combina-
tions. This is also a key requirement for demonstra-
ting potential capabilities of the prototype. General
rulesets may also be created, but the full extent of
the beneficial impact is yet to be determined.
Areas regarding deployment and enforcement
possibilities have been considered and a noticeable
amount of ground research has been done, however
there is also still much to work on:
Cloud platforms are a prime candidate for our
solution. However, inquiries done in this field show
that along with them striving to deliver all-around
secure environment, they also severely limit the
execution possibilities for employment of different
intrusion detection methods compared to classic
LAN cases (promiscuous mode, iptables kernel
support, custom routing and network management as
analyzed in section 3.2).
As for enforcement, our current implementation
leverages iptables and ipset for Linux. There are
other similar tools available on different platforms
(access-lists on Cisco network devices, PF
26
for
BSD, and RRAS
27
in Windows Server), their en-
gagement is potentially possible (we have briefly
experimented with a few in order to confirm this),
but whether such usage would be entirely practical is
yet to be confirmed.
Considering the near-future course of our IPS
prototype development, some of the main subjects of
26
stateful packet filter comparable to iptables
27
Routing and Remote Access Service – http://
technet.microsoft.com/en-us/library/dd469714.aspx
the work to be done are large data measurements in
both artificial and real enterprise networks with
actual imminent threats present.
ACKNOWLEDGEMENTS
This research has been supported by Czech Ministry
of the Interior within the project named “Applied
research of a method of an efficient management of
network security of data centers, servers and cloud
services verified by practical measurements”, grant
number VG20122014086, PID VG2VS/242.
REFERENCES
ArcSight, Inc., 2009. Common Event Format. [Online]
Available at: http://mita-tac.wikispaces.com/file/
view/CEF+White+Paper+071709.pdf [Accessed
December 2013].
B. Claise, E., 2004. Cisco Systems NetFlow Services
Export Version 9. [Online] Available at:
http://tools.ietf.org/html/rfc3954
Debar, H., Curry, D. & Feinstein, B., 2007. The Intrusion
Detection Message Exchange Format. [Online]
Available at: http://tools.ietf.org/html/rfc4765
Fielding, R. T., 2000. Representational State Transfer. In:
Architectural Styles and the Design of Network-based
Software Architectures. Irvine: University of Califor-
nia, pp. 76-97.
García-Teodoro, P., Diaz-Verdejo, J., Maciá-Fernández,
G. & Vázquez, E., 2009. Anomaly-based network in-
trusion detection: Techniques, systems and challenges.
Computers & Security, 28(1-2), pp. 18-28.
Gómez, J. et al., 2009. Design of a snort-based hybrid
intrusion detection system. In: Distributed Computing,
Artificial Intelligence, Bioinformatics, Soft Computing,
and Ambient Assisted Living. Berlin: Springer, pp.
515-522.
International Organization for Standardization, 1996.
ISO/IEC standard 7498-1:1994. [Online].
Kazienko, P. & Dorosz, P., 2003. Intrusion Detection
Systems (IDS) Part I. [Online] Available at:
http://www.systemcomputing.org/ssm10/intrusion_det
ection_systems_architecture.htm [Accessed February
2014].
Leblond, E., 2013. Using NFQUEUE and
libnetfilter_queue. [Online] Available at:
https://home.regit.org/netfilter-en/using-nfqueue-and-
libnetfilter_queue/ [Accessed November 2013].
Lim, S. Y. & Jones, A., 2008. Network anomaly detection
system: The state of art of network behaviour analy-
sis.. s.l., s.n., pp. 459-465.
Network Instruments, 2013. TAP vs SPAN. [Online]
Available at:
SECRYPT2014-InternationalConferenceonSecurityandCryptography
424
http://www.networkinstruments.com/includes/popups/
taps/tap-vs-span.php [Accessed February 2014].
Rehák, M. et al., 2008. CAMNEP: An intrusion detection
system for high-speed networks. In Progress in Infor-
matics, number 5, pp. 65-74.
Roesch, M., 1999. Snort: Lightweight Intrusion Detection
for Networks. Seattle, WA, s.n., pp. 229-238.
Scarfone, K. & Mell, P., 2007. Guide to intrusion detec-
tion and prevention systems (idps). NIST special pub-
lication, Volume 800, p. 94.
Snort Team, 2013. SNORT Users Manual 2.9.6. [Online]
Available at: http://s3.amazonaws.com/snort-
org/www/assets/166/snort_manual.pdf
Šišlák, D. et al., 2005. A-globe: Agent development plat-
form with inaccessibility and mobility support. Soft-
ware Agent-Based Applications, Platforms and Devel-
opment Kits, pp. 21-46.
Talawat, T., 2008. Five major types of IDS. [Online]
Available at: http://advanced-network-
security.blogspot.com/2008/04/three-major-types-of-
ids.html [Accessed November 2013].
Network-basedIntrusionPreventionSystemPrototypewithMulti-Detection-APositionPaper
425