T. Hogg and B. A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November/December 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of self organisation and emergent behaviour. analysing the stability of the ecosystem. analysing the adaptability to the environment. getting hints in order to design stable ecosystems. 3. Ecosystem outline Our ecosystem has been built on the ecosystem proposed by Hogg and Huberman [1], in which they define the interaction among large groups of homogeneous and heterogeneous agents. Our ecosystem species are made up of physical heterogeneous agents with different physical abilities. These agents cooperate, through a consensus process, and compete for using limited resources. ....

Hogg T. and Huberman B. A., Controlling Chaos in Distributed Systems, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 21, No. 6, November/December 1991.

....Montaner, Josep Antoni Ramon Institut d Informtica i Aplicacions, Universitat de Girona LEA SICA Lluis Santal s n, E 17071 Girona, Catalonia peplluis, imunoz, bianca, figueras, montaner, jar eia.udg.es Abstract. This paper is a step forward from the agent ecosystems that Hogg studied [6]. We plan to extend these agents ecosystems to physical agents that interact with the physical world. The aim is to conceive algorithms for the choice of resources and to expand this work. Dynamics of choice in such ecosystems depends on pay off functions that contain information about the ....

....role in realising intelligent behaviour using RoboCup framework [7] Our research focuses in dynamical physical agents, in which the term dynamics is a way to model the transient and steady behaviours of the physical bodies of agents [8] 1. 3 Universal Information Ecosystems Hogg and Hubermann [6] studied the dynamics of choice in communities of agents, with different delays to access information. This work defined the resource choice as a preference function of resource 1 vs. resource 2. It is defined in terms of difference of the performance obtained by applying the information from ....

[Article contains additional citation context not shown here]

Hogg T. and Huberman B. A., Controlling Chaos in Distributed Systems, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 21, No. 6, November/December 1991.

....of AppLeS problem [Berman 97] is not clearly understood. This is indeed a very important issue because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [Mitzenmacher 97] and even chaotic behavior [Hogg 91] Sadly, it has been very difficult to investigate the Bushel of AppLeS question under realistic scenarios be cause application scheduling is only in its infancy. There aren t that many application schedulers in production use for any emergent behavior to have appeared in current systems. 8 2. ....

....[Walsh 98] Walsh 99] Computational markets seem to be a natural scenario to explore the stability and performance of systems with multiple independent decision makers. Although small, there is some literature on the Bushel of AppLeS problem. Some fundamental work has already been done by [Hogg 91] and [Mitzenmacher 97] They highlight the importance of diversity for the stability and performance of systems with independent decision makers. Intuitively, when all decision makers employ very similar strategies, there is a greater the chance for herd behavior to happen. Diverse systems are ....

[Article contains additional citation context not shown here]

Tad Hogg and Bernardo Huberman. Controlling Chaos in Distributed Systems. IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, November/December 1991. 55

....behavior created by having multiple instances of SA in the system. This is indeed a very important issue because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [71] and even chaotic behavior [59]. As tr SA SA we shall see in Chapter 6, one emergent behavior resultant of using SA with many jobs is that the system as a whole becomes more competitive, making it harder for each instance of SA to improve the performance of the job it schedules. On the other hand, the emergent behavior ....

....behavior of the system thus comes from the emergent behavior of all SAs. This is a very important issue because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [71] and even chaotic behavior [59]. There are two basic concerns about a system in which many entities make decisions independently. First: Is the system as a whole stable, or does it oscillate in some thrashing cycle Second: What is the impact of multiple SAs on the performance attained by each of them In our environment, the ....

[Article contains additional citation context not shown here]

Tad Hogg and Bernardo Huberman. Controlling Chaos in Distributed Systems. IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, November /December 1991.

....attained by each SA individually, and the system as a whole. This is an important issue because there is theoretical evidence that systems in which resource allocation is t SA SA performed by many independent entities can exhibit performance degradation [24] and even chaotic behavior [18]. As one can expect, one aggregate behavior resultant of using SA with many jobs is that the system as a whole becomes more competitive, making it harder for each instance of SA to improve the performance of the job it schedules. On the other hand, other aggregate behaviors generated by SA seem to ....

....global behavior of the system thus comes from the aggregate behavior of all SAs. This is an important issue because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [24] and even chaotic behavior [18]. Therefore, there are two basic concerns about a system in which many entities make decisions independently. First: Is the system as a whole stable, or does it oscillate in some thrashing cycle Second: What is the impact of multiple SAs on the performance attained by each of them In our ....

[Article contains additional citation context not shown here]

T. Hogg and B. Huberman. "Controlling Chaos in Distributed Systems". IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, Nov/Dec 1991.

....behavior caused by multiple application schedulers in the same system. This is indeed very important matter because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [26] and even chaotic behavior [21]. Hopefully, that is not the case for SA, as fully discussed in [9] 3. SA: The Supercomputer AppLeS SA, the Supercomputer AppLeS, is an application scheduler that adaptively selects the request that submits a moldable job to the supercomputer (as depicted in Figure 2) Recall that a moldable ....

T. Hogg and B. Huberman. Controlling Chaos in Distributed Systems. IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, November/December 1991. 28

....where AppLeS is the shortening for Application Level Scheduler, the application schedulers developed by Fran Berman s group at UCSD and Rich Wolski at the University of Tennessee. n tr SA SA SA 4 many independent entities can exhibit performance degradation [23] and even chaotic behavior [19]. As one can expect, one emergent behavior resultant of using SA with many jobs is that the system as a whole becomes more competitive, making it harder for each instance of SA to improve the performance of the job it schedules. On the other hand, other emergent behaviors generated by SA seem to ....

....behavior of the system thus comes from the emergent behavior of all SAs. This is a very important issue because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [23] and even chaotic behavior [19]. There are two basic concerns about a system in which many entities make decisions independently. First: Is the system as a whole stable, or does it oscillate in some thrashing cycle Second: What is the impact of multiple SAs on the performance attained by each of them In our environment, the ....

[Article contains additional citation context not shown here]

Tad Hogg and Bernardo Huberman. Controlling Chaos in Distributed Systems. IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, November/December 1991.

....tasks across parallel processors to maximize throughput, and section 7 summarizes and discusses outstanding questions. 2TheBasicModel For empirical modelling, we turned to a simplified version of the computational ecosystem model pioneered by Huberman, Hogg, and their colleagues at Xerox Parc [4, 5]. We created a simple model of resource utilization by agents. Theagentswant to maximize their payoffs, and decidewhich resource to utilize using publicly available informationabout each resource s payoffs. The resources payoffs are decreasing functions in the number of agents utilizing them, so ....

....improved, but as the proportion of these predicting agents grew, the system grew unstable again. Using heterogeneity for stability has the advantage of not imposing an additional computational burden on the agents, and potentially stabilizing the system near the true equilibrium. Hogg Huberman [4] approach this problem by increasing the heterogeneity of the system: by manipulating payoffs, they effectively increase the heterogeneity in agents lag times, which is enough to stabilize the system. In addition, they briefly explore the idea of introducing classes of agents that systematically ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November /December 1991.

....behavior caused by multiple application schedulers in the same system. This is indeed a very important matter because there is theoretical evidence that systems in which resource allocation is performed by many independent entities can exhibit performance degradation [23] and even chaotic behavior [18]. Hopefully, that is not the case for SA, as fully discussed in [7] 28 5.2. Future Work The results presented in this paper were obtained using rigorous scientific methodologies. Care was taken to model the intricacies of real life supercomputer usage, and we are confident SA will improve the ....

T. Hogg and B. Huberman. Controlling Chaos in Distributed Systems. IEEE Transactions on Systems, Man, and Cybernetics, vol. 21, no. 6, November/December 1991.

....While many of the details of this model are different from our own, it serves to inform us that agents which are capable of reasoning about and learning from their performance are one potential mechanism for finding a stable (and hopefully optimal) set of congregations. Hogg and Huberman (Hogg Huberman 1991) take a different approach to the problem of controlling the global behavior of a large scale multiagent system. They show that properly applied external rewards and penalties can be used to stabilize an otherwise chaotic system. We take this as encouragement that external pressures can indeed be ....

Hogg, T., and Huberman, B. A. 1991. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man and Cybernetics 21(6):1325--1332.

....concerning convergence are possible [17, 18, 19] Also note that forcing the algorithm to generate averaged behavior does not necessarily lead to good results; for an example in the context of departure time choice see Ref. 20] 1 This fact has at least intuitively been known for a long time [21, 22, 23]. ffl From a behavioral perspective, people in reality do not have access to as much information (travel time on all links) as they have in the simulations. It is actually possible to make the iterations more realistic with respect to the second aspect [23] also, the Intelligent Transportation ....

T. Hogg and B.A. Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21(6):1, 1991.

....under central control. Multicommodity flow algorithms depend on conceptualizing flow as passive . instead of an aggregation of autonomous agents. An example that highlights this difficulty is the computational ecosystem model developed by Huberman, Hogg, and their colleagues at Xerox Parc [6] [4], a simple model of distrubted resource allocation, where individual agents choose between resources on the basis of which resource has a higher payoff. Equilibrium is found when the payoffs from the two resources are equal. If the agents were under central control, this would be a simple root ....

....The essential problem is one of coordination; it is difficult to get the number of agents using a given resource to adjust incrementally, since once they see that the other resource is paying more, they all want to switch to it. This leads to instability and failure to converge. Huberman Hogg [4] first explored the idea of exploiting agent heterogeneity to stabilize the system. Thomas Sycara [9] expanded on that insight. By slightly simplifying the model, they were able to prove that while some forms of agent heterogeneity guaranteed stability at the cost of converging to a sub optimal ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November /December 1991.

.... falls under the rubric of market based control , in which economic transactions are used to bring about some predefined, desired end [Birmingham et al. 1996, Wellman, 1993, Stonebraker and others, 1994, Clearwater, 1995] Agents may be designed to cooperate [Huberman et al. 1996] or to compete [Hogg and Huberman, 1991], but so long as the aggregate evolves 1 toward a globally defined optimum, the system as a whole is deemed successful. But in an open system like the Web, there is no global purpose being served by the collective of agents; in a sense, there is no collective. Agents goals may be harmonious, ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21:1325, 1991.

....the number of agents sharing it (for example, congestion on traffic lanes) Assume that initially the agents are randomly assigned to one of two identical resources. Now, if every agent opts for the resource with the least current usage, the overall system cost (cost incurred per person) increases [12]. So, the dilemma for each agent is whether or not to make the greedy choice. We will now briefly review some of the work by Glance and Hogg [4] on social dilemmas in groups of computational agents. Glance and Hogg [4] study a version of the social dilemma problem known as the Braess Paradox [14] ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21(6):1325--1332, December 1991. (Special Issue on Distributed AI).

.... during this period include Katia Sycara s work on using case based reasoning to allow bargaining parties to arrive at a compromise deal [41] and Hubermann and colleagues work on analyzing the dynamics of interactions between a large number of interacting agents following simple behavioral rules [42, 43]. In addition, two DAI testbeds developed during this period gave researchers a common platform to evaluate new coordination schemes and was also used as pedagogical tools in courses on DAI offered around the world [44, 45] The research issues that came to the forefront over the last five years ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21(6), December 1991. (Special Issue on Distributed AI).

....new agents or clones to perform excess tasks using the unused resources on the system. To decide when to clone, a stochastic model of decision making based on dynamic programming is used. Results by Hogg Huberman indicate the potential benefits of introducing heterogeneity of different forms [5]. These agree with the intuition that in homogeneous settings, the sharing of knowledge may have an undesirable effect on coordination. This is especially so when the agents must make complementary decisions so as to coordinate, i.e. move to different locations. This problem is closely related to ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21(6):1325--1332, 1991.

....tasks across parallel processors to maximize throughput, and section 7 summarizes and discusses outstanding questions. 2 The Basic Model For empirical modelling, we turned to a simplified version of the computational ecosystem model pioneered by Huberman, Hogg, and their colleagues at Xerox Parc [4, 5]. We created a simple model of resource utilization by agents. The agents want to maximize their payoffs, and decide which resource to utilize using publicly available informationabout each resource s payoffs. The resources payoffs are decreasing functions in the number of agents utilizing them, ....

....improved, but as the proportion of these predicting agents grew, the system grew unstable again. Using heterogeneity for stability has the advantage of not imposing an additional computational burden on the agents, and potentially stabilizing the system near the true equilibrium. Hogg Huberman [4] approach this problem by increasing the heterogeneity of the system: by manipulating payoffs, they effectively increase the heterogeneity in agents lag times, which is enough to stabilize the system. In addition, they briefly explore the idea of introducing classes of agents that systematically ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November /December 1991.

....algorithms depend on conceptualizing flow as passive, divisible entity under centralized control instead of as an aggregation of autonomous agents. An example that highlights this difficulty is the computational ecosystem model developed by Huberman, Hogg, and their colleagues at Xerox Parc [6] [4], a simple model of distributed resource allocation, where individual agents choose between two resources on the basis of which resource has a higher payoff. Equilibrium is found when the payoffs from the two resources are equal. If the agents were under central control, this would be a simple ....

....The essential problem is one of coordination; it is difficult to get the number of agents using a given resource to adjust incrementally, since once they see that the other resource is paying more, they all want to switch to it. This leads to instability and failure to converge. Huberman Hogg [4] first explored the idea of exploiting agent heterogeneity to stabilize the system. Thomas Sycara [9] expanded on that insight. By slightly simplifying the model, they were able to prove that while some forms of agent heterogeneity guaranteed stability at the cost of converging to a sub optimal ....

Tad Hogg and Bernardo A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November /December 1991.

No context found.

T. Hogg and B. A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November/December 1991.

No context found.

T. Hogg and B. A. Huberman. Controlling chaos in distributed systems. IEEE Trans. on Systems, Man and Cybernetics, 21(6):1325--1332, November/December 1991.

No context found.

T.Hogg and B.Huberman. Controlling chaos in distributed systems. In IEEE Transactions on Systems, Man, and Cybernetics, pages 1325--1332, 1991.

No context found.

T. Hogg and B. A. Huberman, "Controlling Chaos in Distributed Systems", IEEE Transactions on Systems, Man,and Cybernetics, 21, No. 6 (1991) p9 1325--1332.

No context found.

Hogg, T., Huberman, B.A.: Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21:1325, 1991. http://www.parc.xerox.com/istl/ groups/iea/www/controllingChaos.html

No context found.

T. Hogg and B. A. Huberman, "Controlling chaos in distributed systems," IEEE Transactions on Systems, Man and Cybernetics, vol. 21, no. 6, pp. 1325--1332, November/December 1991.

No context found.

Tad Hogg and Bernardo Huberman. Controlling chaos in distributed systems. IEEE Transactions on Systems, Man, and Cybernetics, 21(6):1325--1332, 1991.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC