Administering a network means keeping track of copious amounts of information. Today's network monitoring tools are "log-based"; they produce large logs of information through which the system administrator is expected to sift and discover problems.
The PEEP approach is to eliminate the need to search through large amounts of text by representing network information in real-time. PEEP uses sound to represent the vast amount of available information about network status. With PEEP, a system administrator can tell what activity is occurring in his network in real-time and isolate where the problem lies.
Original Project Proposal:
It's implied in Star Trek that Engineer Scott can tell whether the ship is running well by the sound it makes. Instead of looking at just a computer screen, Commander Scott uses all his senses, including touch, hearing, and smell to know whether things are going smoothly. Even the computers hum and beep appreciatively when running smoothly, and acknowledge all commands with an enthusiastic response of beeps and gurgles. A network is very much like that starship; it is large, complex, and difficult to describe in pictures. Why shouldn't we incorporate our other senses into the process? If we can create a semiology, in which individual sounds distinguish activities, and sonorities represent states, we can define normality in terms of the gestalt sonority it generates.
Over the last 6 years, since Joan Francioni and Mark Brown first sonified parallel computations[2,3], little has been done to exploit the representational qualities of sound in describing network behavior. Partially this arises from the difficulty of using musical representations that remain pleasing regardless of the data they represent. This, in turn, came from an insistence upon using pure tones as a representational medium. We consider the problems of tonal combinatorics too difficult to solve in general, and instead take the very radical approach of layering natural sounds to represent an event stream. In this way, the resulting sonorities can be made both representative and naturally metaphorical for the network states they represent.
The Semiology of Sound
In Jacques Bertin's work "Semiology of Graphics", he distinguishes between the representational qualities of graphical elements, creating a hierarchy of representational qualities, including:
a) Presence: The viewer can tell something is present.
b) Distinction: The viewer can tell the difference between two things.
c) Order: Viewer can tell which one comes first in order.
d) Quantity: The viewer can tell exactly how big something is.
Bertin argues that although one can learn to infer all of these from any graphical element, certain graphical elements are naturally suited to represent particular kinds of information. Size is the most versatile attribute, able to represent all kinds of information, while shape is among the poorest, able to represent only presence and perhaps distinction.
In sound, the similar classifications can be applied to describe how each sonic attribute can be used representationally:
a) Loudness: order, distinction, presence.
b) Pitch: order, distinction, presence.
c) Timbre: distinction, presence.
Sound differs from light, in that there's no truly quantitative attribute. However, unlike light, individual sounds do not necessarily lose their identity in cluttered sonic environments.
The choice to use natural sounds instead of pure tones comes from several motivations:
a) Years of experience in trying to map pure tones with no luck.
b) The distracting nature of music in the work environment.
c) The soothing effect natural sounds can have in the work environment.
d) The lack of dissonant qualities when adding several natural sounds together to create a sonic gestalt.
Here is one simple sonic mapping with very interesting properties:
- Login = Redwing Blackbird Chirp
- Logout = Wing Flap/Takeoff
- Su = Other Bird Chirp
- Web Hit = Spring Peeper Peep
- Problems Requiring Operator attention = Frog Croak
- Load Average = Wind
- Shell Traffic = Rain
Pitch and aural location can be varied to indicate the source of the data, while loudness indicates the severity of a condition. The key is that regardless of what's happening:
- The sonority indicates roughly what's going on.
- The resulting sonic environment is, on average, soothing rather than distracting.
- Major changes in the computing environments are immediately apparent.
- The person monitoring the environment can be doing something else while listening.
We propose to build a prototype system for network sonification based upon this model, that would run on a standard PC and be driven from logfile, rstat, and snmp data. The system would provide a standard map from events to sounds that preserves representational properties of the sounds according to Bertin's theories. For research purposes, the map from state to sounds will remain flexible.
 J. Bertin, "Semiology of Graphics", University of Wisconsin Press, 1987 (out of print).
 M. Brown, "An Introduction to Zeus: Audiovisualization of Some Elementary Sorting Algorithms", in CHI'92 proceedings, Addision-Wesley, Inc, 1992.
 J. Francioni and J. A. Jackson, "Breaking the Silence: Auralization of Parallel Program Behavior", Journal of Parallel and Distributed Computing, June 1993.
 G. Kramer, Ed, "Auditory Display: Sonification, Audification, and Auditory Interfaces", Addison-Wesley, Inc 1994.