DESIGN AND VALIDATION OF LOCATION MANAGEMENT TECHNIQUES FOR MOBILE AGENTS

Abstract

A Mobile Agent (MA) is a program, like an active and autonomous object, which may be dispatched from a client to a remote server for execution, assisting users and acting on their behalf. MAs have been used for load balancing, network management, E-commerce and information retrieval, etc. An agent supports disconnected operation and offers a potential solution to network problems associated with hand held devices too. A Mobile Agent System (MAS) is the software which provides an environment for MA to create, dispatch, retract and terminate. Some existing MAS are Aglets, Concordia, and Grasshopper etc. Although a lot of research is underway in MA technology, there are still many issues like bandwidth utilization, standardization of MAS, MA naming schemes, efficient location management and message delivery confirmation, that need further exploring. Locating a MA is a major issue, as the user may need to contact the agent to know the status of its allocated task, to stop it, or to give it some additional task. Also, MAs may need to contact each other for current status collaboration through message passing. Two fundamental issues in message passing are tracking the migration ofthe agent and delivering messages to it, tracking requires the location management of MA. An efficient Location Management Technique (LMT) requires a balance between the costs of search and update. This cost includes time taken by each operation, number of messages sent, size of messages and/or the number of nodes the messages need to travel. The existing LMT can be classified into two categories as per message delivery to the agents: Message delivery with location management and without location management. Some existing LMT like Database logging (DL), Path Proxies (PP), Shadow Protocol (SP), Search by Path Chase (SPC), Black board Protocol (BP), Home Proxy (HP), Home Blackboard (HB), Scalable hash based and Movement based Location Management (MLM) have been studied and their limitations identified. It is observed that only SP has given consideration to orphan agents out of these LMT. Message complexity calculation is also not considered important for them. There is no available framework to aid an understanding and development of LMT. Some existing techniques do not ensure The performance of DL, PP, SPC, BSPC and MCMB have been compared by evaluation for metrics like location update availability, interaction availability, scalability, migration overhead, message complexity and interaction overhead, using a theoretical analysis of each. It has been found that both the proposed LMTs give better values for most of these metrics. For example, BSPC offers better location update availability than SPC when there are more intra-regional than inter-regional migrations of an agent within its birth-region. It offers better interaction availability than SPC. In general, BSPC scalability is better than SPC, whereas lower migration overhead than SPC, PP and DL. BSPC also shows lower interaction overhead than DL, SPC and MCMB. It shows higher message complexity than others but does not exceed the highest value of SPC message complexity for intraregional migrations. Similarly, MCMB offers the better location update availability. It's interaction availability is better than DL, PP, SPC and BSPC. It also shows better scalability than PP, SPC and BSPC and also shows lowest migration overhead. MCMB interaction overhead lies between SPC and PP and has lesser message complexity than PP and SPC. Out of the work contained in this Thesis three papers are published in reputed International Journals. One presentation is published online by ICTP Italy, WET' 05. Six papers were published in the proceedings of International Conferences, held in India and USA. One research paper was published in proceedings of National Conference in Himachal Pradesh, India where it received the best paper award. xii confirmed message delivery; some cannot handle concurrent interaction and migration phases, some exhibit high associated update and search cost while some use longer locating times. In order to fill some of the above gaps: a theoretical framework is presented here to aid in the design, development, efficiency measurement and comparison of LMT. An efficient LMT should provide a current location database and operations for its management, Orphan detection and termination, locating an agent and message delivery confirmation. Both static and dynamic algorithms for detection and termination of orphan agents are proposed. Detection and termination of orphan agents is as important to LMT as garbage collection is to Distributed Systems and memory management is to Operating Systems. Mobile agent naming schemes also play an important role in providing a starting point for locating mobile agents. Two novel location update schemes viz., 'Birth region wise' and 'Neighbourhood wise' movement based have been developed. In 'Birth region wise' movement based scheme, a MA does not update its location as long as it roams within its birth region. The 'Neighbourhood wise' movement based scheme takes advantage of the MA being in its Immediate Neighbor Node (INN), by not updating its location, but only registering at the destination node. The proposed framework and location update schemes have been used for the development of two novel LMT, named Broadcasting with Search by Path Chase (BSPC) and MultiCasting and Movement Based location management technique (MCMB). BSPC is based on Search by Path Chase (SPC). It uses broadcasting as a reactive approach to locate the agent, if the locating request comes for an agent roaming within its birth-region; else a chain of proxies is followed. MCMB is based on optimal movement based LMT. Multicasting is used to locate the agent in its INN; else a chain of proxies is followed. BSPC and MCMB models have been built using TINA (Time pertiNet Analyzer) and checked for their validity using reachability analysis tool. Both the models are found bounded, live and deadlock free. The models have been animated for many locating queries, using the stepper simulator tool of TINA. Token reachability to the final state, after passing through different location registers, shows successful completion of the task of locating the agent during animation. xi