Shortest Path Using Antnet Routing
The increasing reliance of mobile IP-enabled and GPS-capable devices is creating a groundswell of support for optimization algorithms that are capable of quickly interpreting and optimizing paths throughout mobile networks. At the intersection of these evolving Mobile Ad-Hoc Networks (MANETs) and optimization of algorithms for mobile networks that resemble in both patterns of localization, adaptation, resiliency and environmental awareness of organic networks, with antnet routing algorithms and approaches to network optimization becoming more prevalent in research. Also inherent in organic network composition is the ability to create regenerative cells or components; the ability of MANETS to do this through growth of subscribers is evident from the origins of the antnet protocol research, based on the mobility assumptions and the growth pattern characteristics of ant colonies, often called ACS (Ant Colony Systems) a network optimization method where artificial ants move around a graph and each represents an instance of a given problem. The movement of these “ants” or instances of a problem creates solution clusters or associations based on the collective agent response to a given problem.
Antnet routing refers to a series of network optimization algorithms that seek to define the most efficient path for network routing of message packets through a series of agents; increasingly these agents are mobile network nodes, where optimization path algorithms seek out the most time-efficient path of both discovery, identification, validation and communication phases of a typical agent interaction cycle of an Antnet-based network.
The intent of this paper is to define the how increasing software agent intelligence has the potential to increase system-wide reliability, scalability across mobile-based networks including the ability to “learn” how optimize routing throughout organic networks. Antnet routing optimization is heavily dependent on both the methodologies of how the antnet agents are deployed, followed by attributes each agent has to environmentally sense problems and formulate solution sets based on interrelationships with other agents. Antnet routing and algorithm optimization theorists have remarked that the development of optimization approaches needs to move beyond merely optimizing on cost alone and its associated drivers including link delay, constraining to minimize the costs of access times, and the reduction of distance through interrelationship routing through agents. Network optimization is possible in antnet algorithms due to the fact that each agent stores intelligence on the current state of the network itself, including the recent inclusion, exclusion, and status of nodes on the network. The researchers contend that antnet routing is still constrained not by problem optimization but by cost reduction strategies, and as a result both Routing Information Protocol (RIP) which defines optimization through distance-vector methods and the Open Shortest Path First (OSPF) optimization approach which is focused on the optimization of the shortest path using the link-state method. Both methods in fact search for the path that is the lowest cost and generally also the shortest path between nodes.
The extensive work of Dorigo and Di Caro, seek to define network optimization over and above path optimization, enabling antnet agents store data on the entire network’s optimization node path, allowing for the creation of synchronized strategies for traversing the network for the configuration of solution sets, relying on least-cost algorithms including RIP and OSPF. Dorigo and Di Caro are also the authors of Antnet 1.0 and 1.1, with the latter C-language based approach to defining the antnet protocol was tested by researchers who found that in comparing performance of RIP, OSPF, and LBR. LBR (Load Balancing Routing) was introduced into the testing of Antnet 1.1 to see if there would be any significant difference in antnet algorithm performance using the three approaches of traditional network routing. LBR as the name suggests seeks to equalize network loading across all network routes and nodes, thereby alleviating congestion and also working towards the goal of ant destruction. The results of the tests showed that using LBR to route traffic through a network. Comparing testing results of RIP, OSPF, LBR, A1.0 and A1.1 is shown in Figure 1, showing the impact on packet delay and average throughput measured in milliseconds (ms). From Baran and Sosa’s article and research, Figure 1, shown below, yields the following insights:
Figure 1: Comparing Performance of optimization approaches
Antnet 1.1’s greatest performance increase is in the area of Transient Hotspot discovery and routing, as measured with a 112.58ms response time, an improvement of antnet 1.0 response of 116.25 ms and more efficient than LBR with 116.76.
Antnet 1.0 overall performance on throughput in responding to a random failure of a node is significantly better than Antnet 1.1, showing that over time agents can “learn” network optimization. This implies that antnet networks have a pronounced learning curve or experience effect in accumulating knowledge of networks.
Antnet 1.0 and 1.1’s greatest weaknesses are in finding transient hot spots, as the algorithm needs to be optimized to first find these specific instances and report back to them. The combinational affects of least-distance algorithms (RIP, OSBF and LBR) all point to the fact that when the distance between nodes on a network is minimized and learned the effects on learning accumulation are at present more significant than using an algorithm to encompass and in effect “learn” an entire network at once.
In this significant finding from the work of Baran and Sosa the unmet requirements of antnet routing become clear. What is needed is a more effective approach at creating constraint-based learning logic within antnet agents, so the network becomes traversed and optimized through a series of constraint-based approaches.
Constraint-based logic and constraint-based modeling is defined as aligning agents and their characteristics with specific goals, objectives, even if that goal is optimization of the network. Constraints define the objective through a series of optimization approaches that alleviate the need for a series of trail-and-error-based path approaches that typify the structure of least-cost and least-distance algorithms today. Constraint-based agents in conjunction with “learning” agents, all inherent in the structure of the agent itself, can significantly increase the performance of an antnet routing algorithm.
Composition of the Antnet Agent
At a minimum an antnet agent needs to have the ability to “learn” of nodes throughout the network, including the definition of type of node, distance, location as defined by network interpolation of location and for larger nodes, IP address, intensity, confidence in the node, and timestamp which taken collectively define where a node is in relation to another, its distance and reliability as measured by both confidence and intensity. Using the concepts of Directed Diffusion applied to an antnet routing agent definition sets the foundation for creating a specific configuration of antnet agent that both “learns” the structure of the network yet also optimizes its own performance relative to the constraints of the given objective (or in the case of a network request) query being created. For any agent to specifically create a constraint-based view of a network, a taxonomy, over and above the database pointers and linkages as advocated by other researchers, first needs to be created through very rapid exploration and mapping of a given network. In effect an antnet-based agent in constructing an ACS which could very well be in the context of a global MANET-driven network, would parse and quickly ascertain the taxonomy and characteristics of the network, then instead of using pointers or any other highly inefficient means to communicate back into databases and other central repositories, create in effect a stored and shared taxonomy across all agents in the network. A taxonomy of the entire network or grid emerges, where each antnet agent contributes to the entire mapping of the network. it’s important to realize that the two-dimensional approach of many of the least-cost and most efficient routing node-based algorithms (RIP, OSBF and LBR) don’t keep a record of interactions with and characteristics of nodes, so the creation of a taxonomy of the network at higher levels overall is not possible. Speed, not intelligence and optimization based on multi-layered taxonomies, are all that RIP, OSBF and LBR look to. What’s needed then is the ability to create a taxonomy that becomes in effect service-oriented – the taxonomy becomes the field of constraints that antnet agents use to navigate and seek optimization of the entire network as opposed to a single connection between nodes. The learning of a network and the creation of a taxonomy is related to the work of MIT on the semantic web, yet at a much more atomized state given the complexities of networks.
What then is needed in an antnet agent to optimize networks given the background in this area as presented in this paper and the significant unmet needs in MANET optimization are the following characteristics, elements, and constraints:
Agent-based state engine. This is critical if the goals of creating a higher-order taxonomy of the network is going to be possible in addition to the creation of an agent-based state engine that acts as the coordination point in the creation, maintenance, decay and regeneration of network taxonomies. The role of the agent-based state engine is also critical for the coordination with directed diffusion components (more on this in a latter part of this section) and the synchronizing of sensory data from directed diffusion elements with goal-seeking constraint-based logic, which is also critical for inclusion in antnet agents. An agent-based state engine also alleviates the need for frequent database queries and the use of time-consuming pointers that drastically drag down ms access times and erase any optimization gains from first defining the network. The antnet agent-based engine only on exception writes back to a database and instead keeps its own table-based approach to mapping the network while synchronizing the key elements suggested for inclusion to antnet agents within this section.
Taxonomy creation algorithms and shared intelligence approaches to ensuring all ants have perfect knowledge of the network’s structure (taxonomy). This is critical as antnet routing needs to include the ability not just map, but learn specific networks’ characteristics and either equate the network structure and behavior to previously-learned models, or quickly create one through a series of network definition routines that scope, classify and optimize the network structure.
Support for Directed Diffusion data elements. Included within an antnet agent there also needs to be descriptors of each node to further add intelligence to the definition of the taxonomy. This includes the type of node, distance, location (also IP address if available), intensity and confidence (measures of reliability of the node) and timestamps. Directed diffusion also assumes that networks’ sensor-based definition assist in continual environmental scanning and the creation of environmental definition of the entire network definition. Combining directed diffusion characteristics in the context of antnet agents brings in the necessary environmental sensing aspects critical for the continued growth and support of a taxonomy of the network that moves beyond being simply dimensional of distance between nodes. Antnet agents then become part of the learning algorithm which is inherent in the optimization of a complete network. Taking this concept further and applying it to the unmet needs on MANETs the pervasive use of GPS positioning in all mobile devices gives each node in any mobile-based ad hoc network a signature and identity which can be sensed throughout the network and then used in the development of a unique network taxonomy. For mobile ad-hoc networks specifically, the creation of taxonomies is essential for the development of increased performance routing and the use of antnet algorithms in the creation of greater efficiencies in Transient Hotspot discovery and as Table 1 has shown, the critical need for being able to capture Failure Nodes throughout a network. Today antnet routing is struggling specifically in this area of performance yet with a taxonomy of the entire network created, those nodes exhibiting characteristics that indicate they have a higher than average level of probability of failing could be flagged and routed around. That way optimization and network traffic could be routed around any node showing probabilistic tendencies to fail and decay.
Constraint-based logic that interpolates the taxonomy and creates an optimized set of node-based strategies for goal attainment. By incorporating constraint-based logic into the antnet agent, goal optimization can be used as the master constraint, followed by the agent’s state engine providing a series of goal attainment alternatives through the network’s taxonomy. This constraint-based logic is essential for inclusion in antnet agents as it capitalizes on directed diffusion data elements in terms of sensory-based reading of the network, and the extrapolation of this data into a taxonomy which acts as the total constraint base for this goal-seeking logic flow. Constraint-based approaches to goal attainment in the context of the taxonomy-defined network also give antnet routing algorithms the potential of finding and reacting to failed nodes and increasing performance in finding Transient Hotspots as well. This is possible given the continual streaming of goal-based queries throughout the network on the one hand and the use of antnet agents to sense and report back to all other agents the current environmental, directed diffusion and ultimately constraint-based structure of the network.
Future Developments to Antnet Agents
The need for bringing together state engines, directed diffusion, constraint-based logic and most critically, the ability to quickly interpret and classify the taxonomy of a network is critical for antnet agent-based technologies to attain the optimal performance levels possible. The next steps beyond these design objectives are the ability of antnet agents to create their own composite applications driven by constraint logic for a specific objective. Composite applications comprised of a series of antnet agents would be defined through a series of constraint-based strategies that would seek to optimize the performance of a given process. An example would be the use of antnet optimization to re-route and manage cell phone service for text messaging through a large mobile ad-hoc network when one specific node becomes inoperable. Another approach to using antnet-based agents is in the definition of new add-on services which could be incrementally added into existing MANETs. The objective would be the optimization of a new communications service using the ability of antnet agents to first create a taxonomy and increase the performance of a new service due to network-level as opposed to node-level optimizations.
Summary
Antnet-based routing algorithms today are not performing at the level they have the potential to due to many factors. The intent of this paper is to recommend the inclusion of antnet-based agent components provide a higher level of optimization performance through the definition of taxonomies, execution of constraint-based logic to navigate and use these taxonomies, and lead to the eventual definition of composite applications that align with common process areas or in the case of accentuating performance over a MANETs, the definition and delivery of a new high speed communications service. Ideally this would be possible in amore controlled environment, such as a company-based Intranet. Ultimately however the creation of taxonomies over and above the use of database tables provides a solution for antnet agent decay, in addition to monitoring and optimizing the patterns of generation. Antnet routing today is in need of these agent-based improvements to fulfill its theoretical potential of being efficient across a network-wide set of nodes and performing faster than node-to-node based algorithms.
Dhakan and Menezes (2005) – the Role of Social Structures in Mobile Ad-Hoc Networks. Presented at the 43rd ACMSoutheast Conference March 18-20, 2005, Kennesaw, GA. Accessed through the ACM Portal.
George, Evans, Marchette (2003) – a Biological Model for Self-Healing.
SSRS ’03, October 31, 2003, Fairfax, VA Copyright 2003 ACM 1-58113-784
Almir6n M., Banfn B. & Chaparro E., “Ant Distributed System for Solving the Traveling Salesman Problem,” XXV lnformatic Latinoamerican Conf.-CLE1, Paraguay, pp.779-789, 1999.
Baran and Sosa (2000) – a New Approach to Antnet Routing.. Proceedings Ninth International Conference on Computer Communications and Networks. October, 20000 Sponsored by the Nat. Univ. Of Asuncion, San Lorenzo; Paraguay
Dorigo M. & Di Caro G., “AntNet: A Mobile Agents Approach to Adaptive Routing,” Tech. Report, IRIDIA- Free Brussels University, Belgium, 1997. http://iridia.ulb.ac.be/dorigo/ACO
Dorigo M. & Di Caro G., “Ant Colonies for Adaptive Routing in Packet-switched Comm. Networks,” Technical Report, IRIDIA Free Brussels University, Belgium, 1998.
Dorigo M. & Di Caro G., “AntNet: Distributed Stigmergetic Control for Communications Networks,” Journal of Artificial Intelligence Research, Number 9, pp. 317-365, 1999.
Baran and Sosa (2000), ibid
Kalra, Devendra (1990) a Unified Framework for Constraint-Based Modeling. Technical Report. California Institute of Technology. [CaltechCSTR:1990.cs-tr-90-15]
Chalermek Intanagonwiwat and Ramesh Govindan and Deborah Estrin,” Directed diffusion: a scalable and robust communication paradigm for sensor networks. Mobile Computing and Networking. Pages 56- 67.
Ibid, et.al.
Chalermek Intanagonwiwat and Ramesh Govindan and Deborah Estrin,.”Ibid.
