A Survey on Mobile Agent

Communication Protocols

Salman Ahmed, Aamer Nadeem

Center for Software Dependability

Mohammad Ali Jinnah University

Islamabad, Pakistan

salman_zaibi15@yahoo.com, anadeem@jinnah.edu.pk

Abstract An agent is a self decision making software entity

which acts on behalf of the user. A mobile agent roams the

internet in order to access different services. To attain large

goals different agents work and communicate together in

order to achieve goal efficiently. Reliable communication

between multi mobile agents is still a challenge. For reliable

communication, sender must current

location in order to deliver the message. Mobile

location management consists of two phases. First one is

tracking phase and other is message delivery phase, both of

these phases have their own overheads. Agent

communication failure can occur due to the triangle problem

or message chase problem. The message chase problem

arises in highly mobile agents. Due to the message chase

problem, a message may never be delivered. To deliver the

message to agent in a timely and reliable manner is one of

the challenging issues in multi mobile agent environments.

In this paper, we discuss and analyze existing location

management schemes. The existing approaches lever up the

overhead in first phase but trim down the overhead in

second phase or vice versa. The overhead could be in the

form of number of location update messages, memory or

delay.

Keywords-Distributed System; Mobile agent; Agent Location

Management; Agent Mobility Management;

I. INTRODUCTION

Mobile agents refer to self-contained and identifiable computer programs, bundled with their code, data and execution state, that can move within a heterogeneous network of computer systems. They can suspend their execution at an arbitrary point and transport themselves to another computer system. During this migration, the agent is transmitted completely, that is, as a set of code, data and execution state. At the destination computer system, an

was suspended before [1].

Agents are applicable to many computing areas like distributed systems because of their autonomy and mobility factors. The different agent oriented applications are air traffic control system [6], business process management [2], industrial systems management [7], information gathering [8] etc. E-commerce and

information search on the web are most prominent agent oriented applications which are evolving and being deployed in agent environments [14], [15].

Mobile agents work in a team to complete a given task assigned by the user. Mobile agents must coordinate and cooperate with each other for decision making and information sharing. To accomplish their task, the agents roam over the network to acquire different services. To

essential part of agent communication. The existing mobile agent location management schemes have many issues like triangle problem, and message chase problem etc., which affect the efficiency of identifying agent current location and delivering a message. In agent oriented environment there are parent agents (called master agents) and child agents (called slave agents). Master agent divides a heavyweight task into smaller subtasks and assigns them to slave agents. Master and slave agents need to communicate with each other in a reliable and timely manner to complete their assigned task. To deliver the message efficiently and reliably, mobile agent communication schemes generally have two phases. The first phase is agent tracking phase which deals with locating the agent current location. Second phase deals with the delivery of message to the agent at its current location. Several techniques have been proposed like central server, forward pointers, broadcast or multicast, home agency, hierarchical approach, intelligent update,

-based update scheme, message efficiently forwarding scheme etc for agent communication management systems.

The existing agent communication management systems suffer from some communication problems like triangle and message chase problems which may result in communication failure or delay. One of the reasons of communication failure is the autonomous nature of mobile agents. The mobile agents can migrate to other agencies anywhere and anytime to acquire different services. Due to this nature of mobile agents, they may migrate to new agency without receiving a delivered message. Another problem is the triangle problem that arises in agent

978-1-4673-4451-7/12/$31.00 ©2012 IEEE

communication. This problem arises generally in central server approach when two agents situated at adjacent agencies, want to communicate with each other. One of the agents wants to lolocation; it sends a location inquiry to a possibly distant central server. The message chase problem is one of the important communication failures which can occur in highly mobile agents. In highly mobile agent environment, an agent migrates fast from one host to another. A message will follow the chain created by the mobile agent to get delivered. In highly mobile agent environment it may be the case that whenever the message is near to be delivered

host without receiving the message. This situation may lead to a race condition and message may never get delivered.

In this paper, we discuss and analyze different location management approaches which address the problem of message chase for reliable and efficient delivery of messages in high mobility environment. The rest of paper is organized as follows: In section 2, we discuss the background of the mobile agent communication approaches; section 3 discusses the existing approaches related to agent location management systems; section 4 compares the approaches discussed in section 3. Finally, section 5 concludes the survey on mobile agent location management schemes.

II. BACKGROUND

According to the migration behavior of mobile agents, location management schemes are divided into two types as defined by Baumann [9]. The first one is Preordained Path and the second is Autonomous Migration. In the Preordained Path scheme, a mobile agent migrates between hosts only based on a predefined route. A mobile agent in such a scheme can be easily located by using binary search [10] or some other approach. The recent research is focused on Autonomous Migration in which the route of a mobile agent can be determined dynamically which supports mobility and autonomous nature of mobile agents. Broadcast/Multicast, Central Server, Home agency, Email scheme, Hierarchical schemes, Forwarding pointers etc are some initial level schemes used to handle autonomous migration of mobile agents.

In broadcast approach, sender broadcasts the location inquiry query to the all hosts in the network. The recipient agent then replies back its current location to the sender agent. The sender agent directly sends message to the recipient current location. There are some other techniques related to broadcast approach. These approaches work efficiently in local network particularly in bus-based multiprocessor systems but infeasible in large-scale networks like the internet because of large communication overhead. In broadcast based approaches, the messages are sent to all hosts in the network but in multicast the messages are sent to a group or groups in the network.

A central location server is used to keep the current address of all mobile agents in the network. The sender agent first retrieves the current location of receiving agent from the central server (CS) and then sends the message directly to that retrieved location. Although Central Server (CS) approach is easy to implement, but the central location server is really a potential bottleneck of performance and a single point of failure.

Home agency approach is similar to CS scheme but the only difference is, in CS all agents in the network notify their migration to central server but in home agency approach each agent notifies its own home agency on each migration. Each agent in the network has its home agency.

In forward pointers, each host on the migration path of an agent has a forwarding pointer pointing to the next host on the migration path so that messages can be forwarded to the recipient along the path. When an agent migrates to any agency in the network before starting its execution the agent first sends a location update message to its last visited agency. The FP has less reliance on a location server and incurs no location registration overhead. But chain break due to any node crash in the network will affect the message delivery. The agent may not be able to update its location due to broken link between the current and previously visited agency. If an agent migrates rapidly from one agency to another, the message may never get delivered due the race condition problem. In race condition problem agent migrates faster than the message, thus it is possible that whenever message reaches the agent current location, agent has already left its current location and migrated to another agency. There could be delay in message delivery due to the long forward pointing chain.

Another approach [12] by Cao et al. that also uses forwarding pointers but does not use forwarding chains, splits a mobile agent from its mailbox. The mailbox can migrate on request of its agent if the agent thinks that it is worthwhile to have the mailbox close to it but only along

migration path. Thus, all agencies that a mailbox visits are a subset of all agencies that its agent visits. The mailbox migrates with lower frequency than the agent. On each agency that the mailbox visits, there will be a forwarding pointer to the current location of the mailbox. The agent always knows the location of its mailbox and communicates with its mailbox either by occasionally fetching messages or by being informed actively by the mailbox about new messages.

The scheme of forward pointing can be used with home agency to prevent the drawbacks of forward pointing scheme. In integrated home based and forward pointing approaches, when an agent wants to send a message to any other agent in the network, first it communicates with name server to retrieve the home agency location of

agents home agency list. After getting the location of

current location is the responsibility of home agency. In

integrated home agency and forwarding pointing location management scheme a forward chain is maintained from home agency to agent current location. The received messages at home agency follows the forwarding path chain to get delivered.

III. LOCATION MANAGEMENT SCHEMES

Many of the approaches have already been proposed to design the agent location management schemes. Some of approaches focus on reducing overheads in agent tracking and message delivery phase such as I-Update, Ratio based update scheme, MEFS etc are discuss below.

A. I-Update Algorithm

The purpose of I-Update algorithm is to optimize the performance of a mobile agent location management scheme by shortening the length of a channel for the minimum transmission cost every time when a mobile agent arrives at a new location. However, instead of exploring the network exhaustively for an absolutely shortest channel length without cost-effectiveness, the operation of I-Update merely allows a mobile agent to determine a relative shortest channel from among the hosts on the candidate channels [5]. In I-Update, each and every node maintains the address of every other node in the network with their transmission cost. Every host in the network stores this information in membership table. examine the figure 1 in which hosts are drawn with circle like Hi, Hi+1, Hi+2, Hc. According to I-Update scheme host H

such as Hi+1, Hi+2, Hc. Similarly Hc contain transmission cost of other hosts such as LMS, Hi, Hi+1, Hi+2. Same is the case with LMS, Hi+1, Hi+2. The transmission cost from a LMS to each host on a channel is called the channel cost. For example, suppose that a mobile agent MA is at host B now and its channel is (LMS, A, B) in this case, the channel cost of host B is the sum of the cost from LMS to A and the cost from A to B. The channel cost of each host on a channel is stored in a CCT (Channel Cost Table) table. Each mobile agent carries a CCT table as it migrates between hosts.

In I-update whenever agent migrates to a new location it updates one of its previously visited hosts. For example in figure 1 agent migrate from host Hi+2 to Hc. After reaching Hc, before executing agent first calculate the shortest transmission cost path to previous hosts. The last host that it visited was Hi+2 and the channel was LMS Hi Hi+1 Hi+2 at that time. MA stores the records of Hi, Hi+1, and Hi+2, and their corresponding channel costs, 3, 5, and 8 respectively, currently form a relative shortest channel. The transmission cost from LMS to itself is 0. The membership costs from Hc to the four hosts are 15, 14, 8, and 7 respectively. Secondly, MA adds the channel cost of the four hosts in the CCT table with their corresponding membership costs. For example, the membership cost from Hc to Hi+2 is 7, the channel cost of Hi+2 is 8, and thus the sum of the two costs are 15. After the computations, the sums are 15, 17, 13, and 15

respectively. It is obvious that the best choice is Hi+1 with its sum 13, and then MA updates its location information for this host. The new channel created by updating Hi+1 is LMS Hi Hi+1 Hc. Now, if a new message receives at LMS for agent it will be deliver to agent by following new channel.

This approach efficiently manages shortest channel which in result fast delivery of interaction message to agent in timely manner. I-Update uses mailbox which travel near to agent on the channel for reliable communication. The major drawback of this approach is extra memory overhead of storing and maintaining each and every host address with their transmission cost in membership table which is very unstable for the application on internet.

Figure 1. An example to illustrate I-Update algorithm. (Reproduced

from [5]).

B. Ratio-based Update Scheme

In the ratio-based update scheme which is based on the home based LMS architecture, a mobile agent send out an update message at the best update timing to minimize the overall transmission cost of produced messages in a mobile agent system. To be more precise, a mobile agent updates its location information as the ratio between its current update cost and message delivery cost is below a threshold value [11].

Figure 2. Illustration of I-Update algorithm. (Reproduced from [11]).

In ratio based update scheme ratio r is calculated by dividing the update cost by channel cost. The ratio R is a fixed threshold value determined according to application requirement. If r < R a location update message is sent to home agency from agent current location. The new channel is now initiated from the current host. For example in figure 2 agent arrives at host Hi+k , it calculate

its home agency. It will leave the forward pointer of its next host Hi+k+1 to its current host Hi+k just before

departure. cond last node of current channel. When

agent arrives at location Hi+k+1 it will ask for the update cost and current channel cost from Hi to Hi+k+1. calculated and compared with R, which is small. The agent updates its current location to home agency. The new channel is started from the host Hi+k+1. A new channel is initiated whenever r < R. Now if a new message is received at home agency it follows the new channel started from the last location update by agent. The message follows the forward pointer chain host by host to get delivered at agent current location.

This scheme shortens the message delivery channel by keep calculating ratio r host to host. The message travelled through optimized channel to get delivered to agent. This approach can also be integrated with Mobile Agent platforms by little modification to existing schemes, like Mole [17] which use home agency approach or aglets [18] which adopts forward pointers approach.

This approach is not much efficient in highly mobile environment. The problem of message chase problem still exists in this approach. The new channel is minimized for new messages received at home agency. The message at old channel may be chasing the agent all time and may

ulated by location update cost to home agency and current channel

location update cost creates a network overhead. Similarly the host to host channel cost calculation also creates a network overhead. Before executing, agents have to wait for the calculation of

affects the overall performance.

C. Message Efficiently Forwarding Scheme

In Message Efficiently Forwarding Scheme approach MEFS [13] mobile agents are allowed to communicate seamlessly regardless of their location, each agent is assigned a global unique name when it initializes from which its address can be easily resolved. When migrating, agent must unregister, leaving a forwarding pointer, when it leaves a node and register when arrives at a new one. To solve the message-chasing problem, they use synchronous communication when

as little as possible. In MEFS Message Efficiently Forwarding Schema each agent must calculate the velocity of itself once reaching a node. The velocity can be the average speed in a given period of time. If the velocity exceeds a given maximal V0, the agent should establish a connection with its home place, getting and then deleting the chasing message number list for it. Agent will be blocked at current node until all chasing messages are received. This is so-called Over-speed Agent Blocking [13].

The MEFS approach manages the fast mobile agents by stopping the agent to move further and ensures the reliability of messages by maintaining a list of delivered

messages. By following this approach chance of communication failure in high mobile agent is reduced.

The drawback of this approach is the blockage of agent at time of over fast mobility which affects the execution performances of agent. This approach suffers from really inefficient phase when the mobility and the number of messages for an agent increases. Although this approach ensures the reliable delivery of messages but possibly suffer from delay delivery in case of long forward chain. The issues regarding forward pointer scheme also affects the performance of MEFS like chain break due to host or link crash, garbage pointers disposal, long forwarding chains and cycles.

D.

[3], dU (d Updates) means that a mobile agent updates its location once every d movements. SSM (Sequential Searching Method) means that the

one by one from the latest record. This scheme formulates the cost functions of location update and searching. It demonstrates that there must be an optimal threshold d that makes the total cost minimized. In other words, d is the balancing point between updating and searching. More updating operations produce more updating cost and at the same time, need less searching operations and hence less searching cost.

E. Time-based Scheme

Songsiri [4] devised a time-based scheme, in which a mobile agent updates its location information to its LMS

approach considers a free roaming mobile agent scenario where the owner dispatches its mobile agent to carry out some tasks during its lifetime. Each mobile agent has a unique name defined using the naming function, and lifetime. Once its lifetime has expired the mobile agent migrates back to the owner. Besides performing some tasks on the visited host, the mobile agent must periodically update its current location to a name server. The visited hosts must maintain pointers to the host to be visited next by the mobile agent until the consecutive update is done. Once a new location update has been done, the mobile agent sends messages informing each visited host to erase their forwarding pointers.

In the dU scheme, a mobile agent updates its location information after d migrations between hosts, while in the time-a mobile agent updates its location information to its LMS. Both of the two schemes use different criteria to decide the adequate update timing, and their evaluation also proved the overall performance can be largely improved by this way. Both the dU scheme and the time-based scheme have been proved, at the right timing, to improve the performance of mobile agent location management schemes. Because the higher the transmission cost of update messages is, the lower the transmission cost of interaction messages will be, and vice versa, therefore, there should exist a specific ratio between the transmission

costs of update messages and interaction messages as the best update timing mechanism.

The above mentioned approaches manage the problem of location management. Each approach discussed above has its own overheads while managing agent tracking and message delivery phase. I-Update approach efficiently manages the message delivery but creates extra memory and network overheads by storing transmission cost of every host in the network to each host in the network. The schemes like dU-SSM and optimal periodical update are better than FP and home agency in terms of reducing location management costs. MEFS has higher reliability in message delivery phase of highly mobile able agents. The drawback of MEFS approach is blockage of agent itself when exceeded its speed limit. In next section we compare these approaches by mapping a table based on some identified parameters.

IV. COMPARISON

From the existing schemes, we identify some parameters to compare the overall performance. The parameters are Network overhead (Network OH), Memory overhead (Memory OH), Reliability and Efficient Message Delivery (Efficient MD). These parameters are defined below.

A. Network OH

The network overhead is calculated on the basis of two types of messages which have to be exchanged within the network. One is the number of exchange of location updates messages and the other is number of exchange of messages to calculate the transmission cost between two hosts. The more the message exchanges the network over head is high. The less the number of message exchanges the network over head is low.

For example, suppose number of update message is denoted by U and the number of agent migration is denoted by K. If U = K it means on each migration agent sends one location update message. The network overhead in case of U = K is medium. If U > K it means on each migration agent sends more than one location update message. The network overhead in case of U > K is high. In other words, if number of update messages (U) after each migration is greater than 1, the network overhead is high. If U < K it means on K migration agent sends U location update message. The network overhead in case of U < K is low.

If any scheme calculates transmission cost, it has to be added with the location update cost. First calculation of transmission cost is dependent on the specific scheme that how many messages it transmits between the hosts. Secondly, how frequently agent needs to calculate the transmission cost while migrating from one host to another. Let us assume agent sends only two messages from one host to another for calculation of message transmission cost. Let A and B are two hosts and agent current location is A. We want to calculate transmission

cost between A and B. A sends a cost calculate message to B. The time taken to transmit that message from A to B is transmission cost between A and B. One message is sent from A to B and second message from B to A for notification to A about cost calculated.

For example, suppose transmission cost messages are denoted by TC and the number of agent migrations is denoted by K. If TC = 2K it means on each migration agent calculate transmission cost. The network overhead in case of TC = 2K is medium. If TC > 2K it means on each migration agent calculate transmission cost more than 1 host. The network overhead in case of TC > 2K is high. In other words if number of TC messages after each migration is greater than 2, the network overhead is high. If TC < 2K it means on K migration agent calculate TC only once. The network overhead in case of TC < 2K is low. In other words if number of TC messages are only two after K migrations than the network overhead is considered to be low.

While visualizing the network overhead we consider both Location Update and TC calculation overhead.

B. Memory OH

The memory overhead is visualized by the storage of host address and the transmission cost between two hosts. The more transmission cost and host address is stored the memory over head is High. The less transmission cost and host address is stored the memory overhead is Low.

For example, suppose number of location of host is denoted by L and number of migrations by agent is denoted by K. If L = K it means whenever agent migrate to new location it must be stored on any host in the network. The memory overhead in case of L = K is medium. If L > K it means after each migration new location of agent is stored on more than one host in the network. The memory overhead in case of L > K is high. If L < K it means after K migrations of agent the location stored L is less than K. The memory overhead in case of L < K is low.

Same is the case with number of transmission cost entries on a host T. If K = T the network overhead is medium. If K > T the network overhead is low. If K < T the network overhead is high. While visualizing the memory overhead we consider both overheads of storing of host address and storing the transmission cost between two hosts.

C. Message Chase

The factor of message chase is regarding the reliability of messages delivery. The reliability is probability that a message that is sent to the agent current location is delivered or not. If the message is always delivered to agent on its current location then the scheme do not have message chase problem. We mention NO for message chase problem if scheme is reliable. We mention YES for message chase problem if scheme is not reliable. Different

existing schemes reduce the probability of message chase problem to make the communication reliable.

D. Efficient Message Delivery (Efficient MD)

The Efficient MD is the factor that determines how much a message takes time to get deliver to agent current location. If the messages are delivered to agent current location in minimum time than Efficient MD is said to be high. If messages are delivered to agent on its current location with some delay than expected time the Efficient MD is said to be low. The efficiency of message delivery is based on the path, which is selected to deliver the message. If path has less transmission cost than it will be delivered in efficient time and if path has greater transmission cost then it will take time to deliver message to agent current location.

Table 5 shows the comparison between discussed approaches on the bases of identified parameters. The network over head is high in I-Update approach because mailbox approach is also integrated with this approach. Some communication messages are also exchange between home agency and mailbox, agent and its mailbox which increase the network overhead. Although I-Update message delivery time is vey less because of its appropriate channel creation during migration but the memory over head and network overhead is very high which is unstable in large networks.

TABLE 5. COMPARISON TABLE.

Network

OH

Memory

OH

Message

Chase

Efficient

MD

Home Medium Low Yes Low

I-Update Medium High NO High

Ratio based Medium Low NO Medium

dU Low Low NO Medium

Time based Low Low NO Medium

MEFS Medium Medium NO Medium

V. CONCLUSION

This paper analyzes and evaluates the existing agent communication schemes for mobile agent systems. We compared these approaches by defining some parameters like network overhead, memory overhead, message chase, and efficient message delivery. According to our survey the scheme I-Update [5] is considered to be efficient in terms of fast and reliable message delivery since it uses the minimum cost path. However, due to high memory overhead, this scheme is not scalable for large networks like internet. Further research needs to be done in future to overcome the drawbacks of existing approaches. In particular, there is a need for an efficient agent communication scheme that keeps the network and memory overheads low.

VI. REFERENCES

[1] Peter Braun, Mobile Agents Basic Concepts, Mobility Models, andthe Tracy Toolkit, 2004.

[2] N. R. Jennings, P. Faratin, M. J. Johnson, T. J. Norman, P. and M. e. Wiegand. Agent based business process

management. International Journal of Cooperative Information Systems, 5(2&3):105 130, 1996.

[3] T.-Y. Li and K.- l location update and searchingProceedings of the first international joint conference on Autonomous agents and multiagent systems, 2002, pp. 639-646.

[4] itecture and optimal periodicalCIMCA 05: Proceedings of the International Conference on Computational Intelligence for Modeling, Control and Automation and International Conference on Intelligent Agents, Web Technologies and Internet Commerce, Vol. 1, 2005, pp. 778-785.

[5] Tzong I Wang, Ting Yuan Yeh, Development of Low Cost MessageDelivery Path for Mobile Agent Communication, Journal of Convergence Information Technology Volume 5, Number 2, April 2010.

[6] M. Ljunberg and A. Lucas. The OASIS air traffic management system. In Proceedings of the Second Pacific Rim International Conference on AI (PRICAI-92), Seoul, Korea, 1992.

[7] N. R. Jennings, J. M. Corera, and I. Laresgoiti. Developing industrial multi-agent systems. In Proceedings of the First International Conference on Multi-Agent Systems (ICMAS-95), pages 423-430, San Francisco, CA, June 1995.

[8] P. Maes. Agents that reduce work and information overload. Communications of the ACM, 37(7):31 40, July 1994.

[9]

Faculty of Computer Science, 1999.

[10]

pp.703-707, 1996.

[11] Ting Yuan Yeh, Tzong I Wang, Minimizing Message Delivery Cost in Mobile Agent Location Management, International Journal of Advancements in Computing Technology Volume 3, Number 5, June 2011.

[12] Jiannong Cao, Xinyu Feng, Jian Lu, and Sajal K. Das. Mailbox-based scheme for designing mobile agent communication protocols. Computer, 35(9):54 60, 2002.

[13] ZHOU Jingyang, JIA Zhiyong and CHEN Daoxu. Designing Reliable Communication Protocols for Mobile Agents. Proceedings of the 23 rd International Conference on Distributed Computing Systems Workshops 2003.

[14] ctronic Commerce using Mobile Agents -107 for

implementing MAgNET (Mobile agent for Networked Electronic Trading), 1998-1999.

[15] P. Dasgupta, N. Narasimhan, L. Moser, and P. Melliar-Smith.

IEEE Transactions on Knowledge Data Engineering 11(4):509- 525 (1999).

[16]

Telecommunications Conference, 2000, vol.1, pp.421-427, 2000.

[17] Addison Wesley, 1998.

[18] J. Baumann, F. Hohl, K. Rothermel, M. Straer, W. Theilmann, -Practice &

ey & Sons, Ltd., vol.32, no.6, pp.575-603, 2002.