Socio-Adaptive Systems

Socio-Adaptive Systems are systems in which human and computational elements interact as peers. The behavior of the system arises from the properties of both types of elements and the nature of how they collectively react to changes in their environment including mission, and the availability of the resources they use. These properties and interactions give rise to new quality attributes that will influence the structure of these systems.

One environment where socio-adaptive systems are needed is that of first responders operating in disaster relief situations. In those settings, first responders rely on wireless networks that are quickly assembled in the field, called ad hoc networks. These networks provide needed information and directions, and they support responders through situational awareness, using messaging-, voice- and video-based applications. The needs of first responders for using such information changes throughout the mission, and the network resources must be used to best satisfy those needs. At the same time, the capacity of the network changes as the network nodes move through the environment, resulting in periods of insufficient capacity. This situation sometimes leads users to overstate their needs in their requests to obtain resources, without considering how this might impair availability for other users. This, in turn, can result in an inefficient use of resources, detrimental to the overall mission.

The SEI created the Socio-Adaptive Systems Project to help enable effective, adaptive, mission-aware use of resources. The goal is to allocate—automatically, continuously, and effectively—scarce network capacity to users based on their own accurately reported needs. The SEI aims to establish a new approach for designing adaptive socio-technical systems in which people, networks, and computer applications can determine locally how to respond when the demand for resources outstrips supply, while guaranteeing the best use of available capacity.

Challenges

This research combines the adaptability of human social institutions, in particular those based on market institutions, with automated network-resource optimization. Highly specialized workers, such as first responders, are trained to act independently and to make complex decisions when confronted with unexpected circumstances. But ever faster operational tempos and the expanded importance of digitally networked resources, combined with a proliferation of new uses for those resources, require substantial automation of resource allocation and optimization procedures. We must address significant challenges to create the desired socio-adaptive combination:

Overstated needs affect resource allocation.
Decision makers do not have access to timely, reliable estimates of network capacity.

Innovations

We focus on two interrelated tasks to address the challenges of allocating scarce network resources in environments of uncertainty and change:

Use computational mechanism design (CMD) to elicit changing mission needs and compute an optimal allocation of resources.
Use decentralized quality-of-service (QoS) optimization to optimally respond to changes in emergency resource capacity.

Research Approach

We apply microeconomic foundations known as CMD to address the decentralized and dynamic nature of network resource allocation in emergency settings. There is significant literature on using incentive-compatible market mechanisms (i.e., mechanisms that cause participants to truthfully reveal relevant information) for allocating computational resources, but little work has considered the dynamics and uncertainty that typify emergency operations. Our research builds on previous work by generalizing promising mechanisms that model the effects of uncertainty on current and future allocation decisions.

We are developing a distributed version of the QoS resource allocation model (Q-RAM)* to allocate emergency resources to applications requiring them. Distributed Q-RAM (D-Q-RAM) is novel in that it does not require a centralized aggregator of application QoS information, nor does it require explicit knowledge of the capacity of emergency resources. D-Q-RAM provides a vehicle for expressing mission needs incrementally: different levels of QoS are associated with expressions of mission utility or value. CMD will rely on D-Q-RAM to compute an optimal lower level allocation based on responder input.

Unlike previous work, the SEI approach makes explicit and quantifiable the relation between responders' needs as they perceive these needs and the ability of the emergency network to satisfy these needs. Further, this approach provides adaptation mechanisms that allow application performance to degrade gracefully in a way that maximizes mission value. Finally, the results of this work can be applied to other scarce emergency resources, such as emergency vehicles or cloud capacity. This approach is also largely agnostic about the underlying mobile ad hoc network protocol and can therefore be combined with a variety of networking infrastructures.

Objectives

This capability will improve current practice in several ways:

Incentive-compatible allocation procedures: The best outcome for each responder is obtained when responders report their true needs. Responders will be able to continuously report their changing needs for digital resources without having to speculate on how their reports will influence allocation decisions.
Effective bandwidth allocation: The proliferation of digital resources, the growing diversity of their uses in emergency networked operations, and the quickened tempo of these operations demand a high degree of automation in managing scarce resources. This capability will reduce reliance on manual intervention to deliver reliable network capabilities where they are most needed, and in time to meet those needs.
Gracefully degrading QoS when network demand outstrips capacity and automatically recovering when capacity becomes available: Our research allows the network infrastructure and the applications it supports to agree on a range of actions that the application can take to exploit increased network capacity and to accommodate reduced network capacity.