Click here to view PowerPoint slides

Data Information, Resource Discovery and Information Taxonomy for Presentation, Selection, and Design

Visualisation for Defensive Information Warfare

Report on Combined Syndicates 2 and 5 Discussions

Peter Clark, Rashaad Jones, Mark Nixon, Martin Taylor


This report contains some ideas shared during our syndicate sessions. We tried to use these few hours of brainstorming to come up with guidelines about how to improve future visualisation techniques to address the challenges of analytic support to defensive information warfare (D/IW) operations. Our discussions focussed on a number of topics especially pertinent to visualisation in this domain:

Visualisation in D/IW is an emerging topic of intensive research throughout the NATO community. It is one of the key enablers of sound Information Assurance in our modern era of communications and control dominated by networked computer systems.


There are several different stakeholders (presentees, human perceivers) in networked information systems with different interests in visualising the conduct of D/IW:

The service provider, whose primary interest is in the costs, profits, schedule and performance achieved in operating networked information server hardware and software, will want an effective visualisation of the current and cumulative financial state of his/her enterprise.

The system subscriber, on the other hand is the ultimate end user of the service provided and will want an effective visualisation of the security and integrity of his/her information products (files, programs) stored in and maintained by the networked information service to which he/she subscribes.

Finally, the system administrator, whose primary purpose is continuity of operations, will want visualisations of current capacities, transactions (and their rates) and levels of resource utilisation. As the front-line engager in D/IW the system administrator has the greatest interest in his/her system’s effectiveness at

Much of what follows, then, is aimed at the purpose, role and interests of the networked system administrator in the conduct of D/IW.

2.1. Good COP, bad COP for D/IW

A Common Operating Picture (COP) for D/IW depends on the presentee’s (human perceiver’s) purpose, role and cognitive style. Different COPs for D/IW are distinguished by different:

With IW, just as with other analytic support challenges, different ways of indexing visualised information elements include indexing by labels, e.g., “Köln-Bonn”, by descriptors, e.g., “Clear flight ceiling” and by locations, e.g., geo-spatial co-ordinates as shown below in figure 2.1 for a conventional air-traffic control visualisation:

Figure 2.1. Indexing Visualised Information by Labels, Descriptors and Locations

For IW visualisation, we should also expect that differences in the granularity and dimensionality of information include continuous (analogue, scalar, vector) and discrete (categorial, symbolic, linguistic, tabular) as shown below in figure 2.2 for the lately mentioned air-traffic control challenge:

Figure 2.2. Indexing Visualised Information by Linguistic and Tabular Presentation

Finally, we note that different modes of information access (data, timing, etc.) may be imposed by the information (e.g., real-time, signal-based), chosen by the presentee explicitly (e.g., by variation in perspective, attribute, etc.) and may be static or, again, streamed (audio and/or video).

2.2. Fidelity Follows Purpose

In D/IW as in other problem domains, a key to effective visualisation will be the degree to which the information is presented with appropriate fidelity. Unless summaries or details are “right-sized” to the role of the presentee, effective (correct and timely) decision making on his/her part will not be enabled.

The “objectness” of a visualisation also serves its fidelity to the presentee’s purpose. This notion speaks to the interactiveness of the visualisation and the degree to which the presentee is able to manipulate it so as to draw out important problem domain features with relative ease. Objectness depends on presentee control and affords the constructed presentation its intuitiveness.


Collaboration and teamwork for distributed D/IW operations are vital to their effectiveness through implementation of a divide-and-conquer strategy. Similarly, through collaboration and teamwork, countermeasures to IW attacks will be disseminated to vulnerable sites sooner and more readily.

Collaboration, however, raises challenges of its own in the D/IW domain if different team members are to divide their efforts effectively so as to maximise continuity of operations. Role-distinguished displays will be required in order to discourage interference of team members with each other’s work. At the same time, teamwork will also require that distributed operators visualise each other’s behaviour.

In order for the existence and progress of an attack to be identified most readily, a phased detection involving many “pairs of eyes” is called for while, at the same time, the security of the system, communications and individual team-member information stores are maintained.


In this section, we discuss some of the more salient characteristics if IW and the intelligence information by means of which it is countered defensively.

4.1. The Pace and Tempo of IW

IW operations are significantly different from those of conventional warfare. Adversaries in IW are not restricted or paced by the rate at which physical objects of conventional warfare (aircraft, missiles, armoured vehicles, etc.) can be developed and manufactured – they have only to be perceptive of system vulnerabilities and imaginative in ways of exploiting them. IW is rather like an accelerated “arms race”. The key to effectiveness in D/IW is in implementing and disseminating countermeasures to his attacks. This taxes the adversary’s main resource, viz., his creativity with respect to new, unprecedented attacks. Here teamwork and collaboration are vital to winning by attrition at a numbers game with the IW adversary.

NB Countermeasures could well play into an adversary’s hand by establishing new vulnerabilities.

Below we discuss the perceptual modes involved in D/IW. Underlying all of these is the desirability of understanding the plans and strategies of an IW adversary. To the extent that the IW adversary’s intentions can be anticipated and his plans and strategies understood, there is the opportunity to engage in “look ahead” evaluation the potential of new vulnerabilities that can arise from any given D/IW countermeasure.

NB The IW adversary’s intentions and strategies can reveal new vulnerabilities due to countermeasures.

The desirability of understanding the IW adversary’s intentions, plans and strategies for mounting attacks calls for problem-solving capabilities in the domains of planning, analogical reasoning and decision/risk analysis in Artificial Intelligence (AI).

The common thread in all such AI problem-solving capabilities is the ability to recognize, declare and apply a rich variety of hypotheses regarding potential attacks against identified vulnerabilities. Indeed, such hypotheses may expose the perpetrator of an offensive IW (O/IW) attack to the risk of being “caught in the act” by contributing, ahead of time, to the formation and anticipation of a possible plan of attack by the D/IW defender before that plan is actually attempted by the O/IW attacker. In this way, higher-level problem-solving techniques contribute to more robust D/IW.

However, it must be emphasised that higher-level AI problem-solving techniques alone, absent effective visualisations for

are of little use in themselves since such techniques do not signal the need to respond as much as they suggest potential ways of responding to O/IW attacks. And, in D/IW, appropriate response is the key to success.

4.2. The Identification of System Risk and Dimensions of Essential Information in D/IW

Identification of system risk is the raison d’être for any of the data acquired in D/IW operations. For, it is only by monitoring the status of vulnerable system assets that D/IW operations can anticipate potential O/IW attacks and establish alerts for their occurrence. All other concerns, such as identifying potential attackers, are secondary.

There are, however, only a limited number of parameters in distributed information systems whereby an O/IW adversary can mount an attack and through which system assets can be exposed to risk. These parameters define dimensions of IW information used for determining anomalous vs. baseline behaviour of network message traffic which is, after all, the means of perpetrating O/IW attacks:

The key to supporting effective D/IW operations, then, will be the effective (co-)visualisation of system behaviour along these dimensions for purposes of detecting anomalous occurrences. However, dimensionality greater than 2 or 3, as in our case, presents important design challenges for effective visualisations. This is where the importance of “objectness” enters with a vengeance. For, it is by the choice of at most 2 or 3 dimensions at a time from among those listed above that candidate visualisations can be formed. The key, then, rests in affording the D/IW operator a simple means of choosing, manipulating and switching among the dimensions this information in controlling the content of his/her visualisations.


In this section, we apply the concepts of NATO RTO Technical Report 30, Visualisation of Massive Military Datasets: Human Factors, Applications, and Technologies, (hereinafter, the HAT Report) to the challenges of visualisation in D/IW. Our aim is to draw out the fundamental perceptual modes required of D/IW, the HAT taxonomic classification of applicable data types for D/IW and the taxonomic classification of effective D/IW displays (visualisations).

We begin with the HAT Report notion of the “dataspace” inside the computer supporting D/IW visualisation. This dataspace, shown in figure 5.1, represents some aspect of the world the human seeks to understand and influence.

Figure 5.1. The IST-05 (VisTG) Reference Model

Display “engines” inside the computer enable human visualisation of the relevant aspect of the world by accessing and presenting dataspace content. In terms of the IST-05 (VisTG) Reference Model, Figure 5.1 exhibits a reciprocal relationship between the human’s understanding and the dataspace in the computer on the one hand, and the human’s visualisation and the engines in the computer that operate on the dataspace on the other.

Figure 5.2. The Emergence of Understanding from Effective Visualisation

As shown in Figure 5.2, the effectiveness of visualisation systems depends on how well they accommodate human cognitive and task performance limitations and on how well they engage and enable human cognitive, sensory and motor capacities.

5.1 Modes of Perception in D/IW

Human purposes in using D/IW visualisations are categorised by four modes of perception distinguished in the HAT Report:

These four modes of perception in D/IW are not of a piece: Alerting and exploring are the primary human purposes in these quarters. For, it is by exploring the vulnerabilities of the system that an O/IW adversary might seek to exploit that alerts may be established to signal critical events requiring immediate response. Interestingly, exploring vulnerabilities and hypotheses regarding an O/IW adversary’s potential lines of attack should consume much of the D/IW operators planning efforts since this leads to the explicit declaration of events that can be detected by alerts.

5.2. The HAT Report’s Six-Dimensional Taxonomy of Data Types in D/IW

The HAT Report prescribes six dimensions along which data types can be distinguished for any given visualisation challenge in support of the intelligence analyst:

We note that D/IW operations for data acquisition, data are acquired in real time for purposes of alerting and off-line for purposes of exploring the characterisation of potential attacks. Data sources are typically singular insofar as assessing immediate vulnerabilities are concerned and plural insofar as disseminated countermeasures (solutions) are published across potential D/IW target systems. Data choice is fundamentally chosen insofar as the D/IW operator needs to select and manipulate the many dimensions of D/IW information available at any one time. Data identification in D/IW is by label according to the dimensions of information chosen from the list above for visualisation. Data values in D/IW are typically discrete and categorical insofar as they involve distinct values from the dimensions indicated above – time being the exception. Finally, data inter-relations in D/IW typically involve the dimension of time with the others as a critical relationship, e.g., when port scans are compared at different access rates to determine the pattern of a potential attack.

5.3 The HAT Report’s Four-Dimensional Taxonomy of Data Displays in D/IW

The HAT Report prescribes six dimensions along which displays (presentations) can be distinguished for any given visualisation challenge in support of the intelligence analyst:

In discussing the challenges of D/IW, we benefited enormously by the participation of Mr. Peter Clark who is an active D/IW operator. We surmised the following points following examples of visualisations Mr. Clark found especially useful. First, if display timing is static, it is useful only if the D/IW operator can manipulate the refreshing of its content. Otherwise, dynamic displays of time vs. some other important dimension of D/IW information (listed above) are called for. Second, it is crucial given the high dimensionality (7+) of the dataspace for D/IW and the variability of threats to which a response must be defined that the operator be capable of directing the selection of data for display. Third, temporal location is critical for many displays such as pattern recognition in port scanning. Nonetheless, labelled data placement is certainly also called for insofar as many of the other dimensions are useful typically for singling out specific values of interest in detecting O/IW adversary operations. Finally, the data values themselves, apart from time, are categorical insofar as they do not admit of degree or uncertainty.

5.3. Summary of Key HAT Report Findings

The perceptual modes in D/IW operations are alerting, exploring and searching. Alerting is the primary mode insofar as it reveals known (components of) earlier attacks and anomalous behaviour relative to standards on message protocols. It is technologically easy to “prime for” many such alerts – NB a real attack may be a subtle one among many simultaneously alerted. Exploring is the key to averting future attacks. This involves understanding vulnerabilities and risks afforded by external accessibility and determining “normal” network user behaviour. Exploring supports priming for alerts and for planning alternative detection, “baiting” schemes. Finally, searching supports determination of the meaning, source and malevolence of a detected attack in progress. Monitoring/Controlling enter only indirectly into the surreptitious tracking of intruders and their resource accesses. The remedy here is a change of resource accesses and performance so as frustrate the attacker.

The taxonomic classification of D/IW data types is as follows: Data acquisition is streamed in real-time. Data sources are multi-source, from redundant router tables, etc. Data choice is chosen through selection of messaging attributes. Data identification is typically labelled except for time. Data values except for time are categorical. Data inter-relations exhibit strong constructional inter-relationships between message components (packets, messages, sessions).

The taxonomic classification of D/IW displays is as follows: Display timing is dynamic in support of real-time alerting, searching or monitoring/controlling but static in exploring potential future attacks and current vulnerabilities. Data selection for display is user-directed to account for the ways in which attacks can exploit many dimensions of message traffic. Data placement is located in 2D or 3D displays, labelled in tables. Data values are categorical in displays


Although much work in the design of effective visualisations for D/IW has already been accomplished, we believe that there is much room for exploring the effectiveness of new or existing D/IW visualisations and prospects for new and different ones.

Effectiveness estimates for new or existing D/IW visualisations require the definition of many scenarios for performance measures against known or standard attacks. For timing-sensitive scenarios, streamed IW data require time-to-detection tests. In any case, the variety and number of IW threat scenarios required for valid performance measures should be rich.

Correctness criteria for performance measures can be established for cases in which measures are not suitable as, e.g., when the question is simply whether an attack was detected at all. Here, risks are ranked qualitatively to some extent for detection purposes. Scoring weights on different risks are nonetheless required given their different costs to continuity of system operations.

For purposes of exploring new visualisations, we recommend the development of “software-less demonstrations” – literally, PowerPoint-slide mock-ups of D/IW visualisations distinguished by purpose, role and cognitive style. Such software-less demos serve the purpose of rapid requirements capture for D/IW visualisation by encouraging broad analysis of the insights they afford. Throughout, a primary goal of the demos would be to characterise data types and displays for D/IW according to the HAT taxonomies.


We mention but a few of many fruitful areas of further research that should be undertaken in D/IW. We note that DARPA has recently published a new Broad Agency Announcement (BAA) for research on effective visualisations in Information Assurance (a nearly identical field) as an indication of interest in this quarter of the NATO community.

Exploration vs. Search in D/IW visualisations – This topic seeks to support the prevention of O/IW attacks by complementing visualisation with hypothesis formation and elaboration regarding system vulnerabilities and adversary capabilities for exploiting them. Here the combination of visualisation with hypothesis formation is the key to anticipating unprecedented attacks

Countermeasures to O/IW – We mentioned monitoring/controlling as an activity useful in tracking and identifying perpetrators. Ideally, this would be but a first step in “taking the fight to” the IW adversary by, in turn, identifying his/her system vulnerabilities and exploiting them as a follow-on to the tracking and identification function afforded by the monitoring/controlling mode of D/IW perception.

Revisiting the theoretical foundations of distributed system security – This topic follows from the observation that system vulnerabilities are not all of a general nature and that some systems are subject to attack by virtue of the user base that they support, the information that these users store in the system and its sensitivity. Here the locus of research addresses unauthorized access attacks (i.e., attacks other than denials-of-service). It involves not so much the system as it does the users (problem solvers). Vulnerability in this case is problem-centric vs. system-centric authorization of access. The promise is that, by assuming a problem-centric stance, techniques may be evolved that may ease the general difficulty of enforcing access control by forcing the adversary to follow problem solver (team) from system to system (2nd order user content security). The problem-solving users engage in a “spread spectrum” use of different systems as infrastructure for problem solving in ways that the adversary cannot readily follow.