Dynamic Scheduling and Division of Labor in Social Insects

Market-based algorithms have been introduced several years ago (Kurose & Simha, 1989; Waldspurger et al., 1992; Huberman & Hogg, 1995; Clearwater & Huberman, 1994; Clearwater,1995; Wellman, 1993, 1995; Ykke et al., 1997; Ykke, 1998) as a new paradigm for controlling complex, unpredictable systems. In a market-based algorithm, resources or tasks are allocated efficiently through a market-clearing mechanism. Agents bid for resources and the agent with the highest bid gets the resource. The market-clearing mechanism ensures that all tasks or resources have been allocated. Agents adjust their bids according to prior successes or failures in getting the resource. For instance, Waldspurger et al. (1992) have developed Spawn, a computational system where each task, starting with a certain amount of “money” that corresponds to its relative priority, bids for the use of machines on a network. Examples of tasks include searching through a database for an item, printing a paper, etc. Through bidding for the use of machines, each task can allocate its budget to those resources that are most important for it (Waldspurger et al., 1992; Huberman & Hogg, 1995). When prices are low enough, some tasks may use several machines in parallel; the number of machines used by, and the computation time devoted to, each task are adjusted to the demand from other tasks. In another example (Clearwater & Huberman, 1994), the problem of thermal resource distribution in a building was solved using market-based control: computational agents that represent individual temperature controllers bid to buy or sell cool or warm air via a double-blind computerized auction moderated by a central computer auctioneer. Clearwater and Huberman (1994) have shown that this system results in an equitable temperature distribution throughout the building at a reduced cost. Ygge et al. (1997), who analyzed the performance of Clearwater and Huberman’s (1994) approach and carefully compared it with other approaches, argue that this problem lends itself to a treatment with a market-based approach in that the goal state can be formulated as a market equilibrium; they also indicate that other multi-agent-based approaches might work equally well or even better. Other, more sophisticated, market-based approaches have been developed in the recent years: for example, Walsh et al. (1998) have formulated distributed scheduling as a resource allocation problem, applied general equilibrium theory to show the existence of, and characterize, equilibrium solutions, and devised auction protocols for implementing solutions. Social insects –ants, bees, termites and wasps–provide us with another metaphor for controlling complex systems (Bonabeau et al., 1999). A social insect colony is a complex system (Wilson, 1971) often characterized by division of labor (Oster & Wilson, 1978; Robinson, 1992): workers tend to be specialized in certain tasks. But this specialization is flexible. Adaptive Behavior 2001, 8(2), 83-92