DARPA ULTRALOG Final Report - Industrial and Manufacturing ...

More documents

Recommendations

Info

O, U Stresses Physical/Infrastructure Layer X F,T Se TechSpecs P DB MP OS Model Builder Q P,X C DMAS Controller S,P,O Control Set Evaluator O',X' Queueing Model P' Stresses Application / User U*,O* U' Sa Utility Calculator Figure 1: Architecture Overview Evaluator knows the current operating mode (opmode) set O as well as the set of possible opmodes, OS from TechSpecs. To evaluate the performance due to a candidate opmode set O ′ , the Control Set Evaluator uses the Queueing Model with a scaled set of operating conditions X ′ . Once the performance P ′ is estimated by the Queueing Model it can be cached in the performance database PDB and then sent to the Utility Calculator. The Utility Calculator computes the domain utility due to (O ′ ,P ′ ) and caches it in the utility database, UDB. Subsequently, the optimal opmode set O ∗ is identified and sent to the DMAS Controller. The functional units of the architecture are distributed but for each community that forms part of a MAS society, O ∗ will be calculated by a single agent. We now examine the capabilities of the framework. 2.1.1 Self-Monitoring Capability TechSpecs acts as a distributed structure that contains meta-data about all variables, X, that have to be monitored in different portions of the community. The data/statistics collected in a distributed way, is then aggregated to assist in control alternatives by the toplevel controller that each community possesses. Each agent can look up its own TechSpecs and from time-to-time forward a measurement to its superior. The superior can analyse this information (eg. calculate statistics such as delay, delay-jitter) and/or add to this information and forward it again. 2.1.2 Self-Modeling Capability One of the key features of this framework is that it has the capability to choose a type of model for representing itself for the purpose of performance evaluation. The system is equipped with several queueing model templates that it can utilize to analyze the system configuration with. The inputs to the Model Builder are the flows that traverse the network (F ), the types of packets (T ) and the current configuration of the network. Given we know that there are n agents interconnected in a hierarchical fashion, this unit represents the information in the required template format (Q) which is subsequently used to analyze the current performance. O* U DB 2.1.3 Self-Evaluating Capability The evaluation capability allows the MAS to examine its own performance under a given set of plausible conditions. This prediction of performance is used for the elimination of control alternatives that may lead to instabilities. Given that a variety of tasks traverse the heterogeneous network of agents in predefined routes (called flows), the processing and wait times of tasks at various points in the network are not alike because of dissimilar configurations, resource availabilities and/or environmental stresses. Under these conditions, performance is evaluated in terms of end-to-end delays for the “sense-plan-respond” cycles. 2.1.4 Self-Controlling Capability Since tasks can be processed at various pre-defined qualities, opmodes allow for trading-off quality of service (task quality) for performance (end-to-end response time). The available resources fluctuate depending on stresses S = S e +S a, where S e are the stresses from the environment (i.e. multiple contending applications) and S a are the application stresses (i.e. increased tasks). Using current measured performance P and the measured stress S the DMAS Controller relates the overall utility (U) asU(P, S) = P w nx n where x n is the actual utility component and w n is the associated weight specified by the user. To adjust P to get the best achievable utility under S, the following is done. Since P depends on O, which is a vector of opmodes collected from the community, we can use the QNM to find O ∗ and hence P ∗ that maximizes U(P, S) for a given S from within the set OS. In words, we find the vector of opmodes (O ∗ ) that maximizes domain utility at current S. The utility computation is performed in the Utility Calculator module using a learned utility model based on UDB. 3. CONCLUSIONS We combined queueing analysis and application-level control to engineer a generic framework that is capable of self-optimizing its domain-specific utility to assure application-level survivability. While application-level adaptivity yields improvement in utility further gains are possible by leveraging underlying layers. 4. ADDITIONAL AUTHORS Additional Authors: Natarajan Gautam (Pennsylvania State University, email: ngautam@psu.edu), Wilbur Peng and Vikram Manikonda (IAI Inc., email: wpeng,vikram@i-a-i.com), Marshall Brinn (BBN Technologies, email: mbrinn@bbn.com) and Mark Greaves (DARPA IXO, email: mgreaves@darpa.mil). 5. REFERENCES [1] M. Brinn and M. Greaves. Leveraging agent properties to assure survivability of distributed multi-agent systems. Proceedings of the Second Joint Conference on Autonomous Agents and Multi-Agent Systems, 2003. [2] N. Gnanasambandam, S. Lee, N. Gautam, S. R. T. Kumara, W. Peng, V. Manikonda, M. Brinn, and M. Greaves. Reliable mas performance prediction using queueing models. IEEE Multi-agent Security and Survivabilty Symposium, 2004. [3] H. Jung and M. Tambe. Performance models for large scale multi-agent systems: Using distributed pomdp building blocks. Proceedings of the Second Joint Conference on Autonomous Agents and Multi-Agent Systems, July 2003. [4] O. F. Rana and K. Stout. What is scalabilty in multi-agent systems? Proceedings of the Fourth International Conference on Autonomous Agents, 2000. [5] G. Tesauro, D. M. Chess, W. E. Walsh, R. Das, I. Whalley, J. O. Kephart, and S. R. White. A multi-agent systems approach to autonomic computing. Autonomous Agents and Multi-Agent Systems, 2004.
Manuscript for IEEE Transactions on Automatic Control 1 ADAPTIVE CONTROL FOR LARGE-SCALE INFORMATION NETWORKS THROUGH ALTERNATIVE ALGORITHMS TO SUPPORT SURVIVABILITY * Seokcheon Lee † and Soundar Kumara ‡ †‡ Department of Industrial & Manufacturing Engineering, The Pennsylvania State University, University Park, PA 16802 † Phone: 814-863-4799; Fax: 814-863-4745; E-mail: stonesky@psu.edu ‡ Corresponding author. Phone: 814-863-2359; Fax: 814-863-4745; E-mail: skumara@psu.edu ABSTRACT As modern networks can be easily exposed to various adverse events such as malicious attacks and accidental failures, there is a need to study their survivability. We study a large-scale information network composed of distributed software components linked together through a task flow structure. The service provided by the network is to produce a global solution to a given problem, which is an aggregate solution of partial solutions of individual tasks. Quality of service of the network is determined by the value of global solution and the time taken for generating global solution. In this paper we design an adaptive control mechanism along the lines of model predictive control to support the survivability of such networks by utilizing alternative algorithms. To address adaptivity we model stress environment by quantifying resource availability through sensors. We build a mathematical programming model with the resource availability incorporated, which predicts quality of service as a function of alternative algorithms. The programming model is decentralized through an auction market without any degradation of the solution optimality. By periodically opening the auction market, the system can achieve desirable performance adaptive to changing stress environments while assuring scalability property. We verify the designed control mechanism empirically. Key Words: Adaptive control, survivability, alternative algorithms, scalability * This work is supported in part by DARPA under Grant MDA 972-01-1-0038.
Page 1 and 2:
Ultra*Log PSU/IAI Final Report for
Page 3 and 4:
Contents Contents .................
Page 5:
Executive Summary Ultra*Log is a De
Page 8 and 9:
2.3 Gnanasambandam, N., Lee, S., Ku
Page 10 and 11:
6 Characterization and analysis of
Page 12 and 13:
timizing simultaneously the link de
Page 14 and 15:
where ∆ 1 (j) ≥ 0 and ∆ 2 (i,
Page 16 and 17:
Table 2. GA Results Agent N A D max
Page 18 and 19:
connected component, in which a pat
Page 20 and 21:
Random Small-world Scale-free with
Page 22 and 23:
Growth mechanisms Start with a smal
Page 24 and 25:
Table 2. The proposed network’s c
Page 26 and 27: ¡ ¡ ¢ £ ¤ ¥ ¦ £ § ¨ © ¥
Page 28 and 29: © © ¤ ¢ ¨ ¤ £ ¦ ¨ © §
Page 30 and 31: DMAS Controller. The functional uni
Page 32 and 33: 1000 500 0 -500 -1000 0.2 0.4 0.6 0
Page 34 and 35: tems. Proceedings of the Second Joi
Page 36 and 37: Proceedings of the 1st Open Cougaar
Page 44 and 45: 1 SITUATION IDENTIFICATION USING DY
Page 46 and 47: 3 3.2 Behavior In SSC society an ag
Page 48 and 49: 5 All the behavior parameters may n
Page 50 and 51: Estimating Global Stress Environmen
Page 52 and 53: chaotic deterministic time series.
Page 54 and 55: 100% 1 95% 2 14 15 64% TAO 4 64% 62
Page 56 and 57: interactions as a dynamical system
Page 58 and 59: ehavior states under varying system
Page 60 and 61: Extensive testing and validation of
Page 62 and 63: 5 Conclusions and Future Research T
Page 64 and 65: 2 to function critically even under
Page 66 and 67: 4 points into a corresponding inner
Page 68 and 69: 6 stresses well. The Warehouse 1 ag
Page 70 and 71: © £ ¨ ¥ ¤ § ¢ ¥ ©
Page 72 and 73: Table 1: Notation Symbol Descriptio
Page 74 and 75: 4 Conclusions and Future Work 4.1 C
Page 78 and 79: Manuscript for IEEE Transactions on
Page 106 and 107: 2 discuss problem domain and in Sec
Page 108 and 109: 4 t Si t ∫ + ( ) i ) t When RA i
Page 110 and 111: 6 S UB i ∑ = P LI / LI . (16) i p
Page 112 and 113: 8 policies while underutilizing in
Page 114 and 115: architecture based on both the Grid
Page 116 and 117: to machines (network topology) and
Page 118 and 119: component is a member of one of the
Page 120 and 121: denotes the immediate predecessors
Page 122 and 123: limit of large number of tasks. If
Page 124 and 125: Min-min heuristic algorithm Step 1:
Page 126 and 127:
We set up eight different experimen
Page 128 and 129:
0. 4 8057 8237 0.5 6439 6635 0.6 53
Page 130 and 131:
pp. 191-200. [3] I. Foster, C. Kess
Page 132 and 133:
Technology, Cambridge, MA, 1995. [2
Page 134 and 135:
Manuscript for IEEE TRANSACTIONS ON
Page 136 and 137:
Page 138 and 139:
Page 140 and 141:
Page 142 and 143:
Page 144 and 145:
Page 146 and 147:
Page 148 and 149:
Page 150 and 151:
Page 152 and 153:
Page 154 and 155:
Page 156 and 157:
Coordinating Control Decisions of S
Page 158 and 159:
directly. It has coarser scale. It
Page 160 and 161:
Understanding Agent Societies Using
Page 162 and 163:
• For tasks, the UID of the direc
Page 164 and 165:
for tracking the dependencies betwe
Page 166 and 167:
within the monitored enclave. With
Page 168 and 169:
Figure 1. Agent Hierarchy in CPE So
Page 170 and 171:
Figure 3. The World Model CPY Agent
Page 172 and 173:
Table 2. TechSpecs: Infrastructure
Page 174 and 175:
terms of the average waiting times
Page 176 and 177:
Acknowledgements The work described
Page 178 and 179:
key realization to tackle this prob
Page 180 and 181:
are or ignored perturbations. The e
Page 182 and 183:
sustained oscillations. If the inst
Page 184 and 185:
solved by biological systems for li
Page 186 and 187:
non-linear and non-stationarity. Ne
Page 188 and 189:
D k : Outflow from high priority qu
Page 190 and 191:
the system starts to perform self-s
Page 192 and 193:
sustained in a supply chain. Resour
Page 194 and 195:
which are necessary to make good mo
Page 196 and 197:
focusing on the computational compl
Page 198 and 199:
line of research that deals with ne
Page 200 and 201:
ealizable. But the inherent complex
Page 202 and 203:
Min, H. and Zhou, G., 2002, Supply
Page 204 and 205:
In the rest of this paper we report
Page 206:
shows relatively irregular behavior
show all

DARPA ULTRALOG Final Report - Industrial and Manufacturing ...

Create successful ePaper yourself

Delete template?

Save as template?