NET-COMPUTER: Internet Computer Architecture Based on Intelligent Agents

Transcription

1 NET-COMPUTER: Internet Computer Architecture Based on Intelligent s Odhiambo Marcel O Department of Electrical and Mining Engineering, University of South Africa (UNISA), P.O. Box 392, UNISA , South Africa. Abstract - Research in Intelligent s has been ongoing for sometimes now yielding interesting results, some of which have been translated into commercial ventures. Intelligent s are executable software components that represent the user, performs tasks on behalf of the user and when the task terminate, the s send the result to the user. s are best suited for the Internet: a collection of computers connected together in a world-wide computer network. A number of systems have been implemented, the simulation test results do show the feasibility of Intelligent s and systems. s technology gives rise to a new computer architecture: the NET-COMPUTER based on HYDRA computer architecture in which the computing power resides in the Internet. In the NETCOMPUTER, the Internet computers form the hardware and software resources, and the user is provided with a simple input/output interface through which he/she can access the Internet to run user tasks. The simulation test result do show the feasibility of the NET-COMPUTER based on HYDRA computer architecture and Intelligent s. Keywords: Intelligent s, Internet, HYDRA computer architecture, NET-COMPUTER 1. Introduction s technology developed from related works on distributed Artificial Intelligence (AI), where the researchers aimed at spreading intelligence behavior through distributed computer systems to tackle inherently distributed problems. They developed intelligent software programs (or s), where the most advanced can communicate, collaborate and even learn from each other as they use their knowledge for problem solving. The Internet has made it possible for computers at both ends of the world connected to one worldwide network: the Internet to talk to each other. From the comfort of their office or home, the user can log onto the Internet, and browse through documents on a remote machine several thousands of miles away. Combining information superhighway with the Internet opens the door for all sorts of communication and information processing possibilities. Automatic banking or Automatic Teller Machines (ATM), Imaging, Remote Teaching and Telemarketing are among the many services that can be provided [1]. A growing segment of the Internet is Electronic Commerce. Consumers are looking for suppliers selling products and services on the Internet. Meanwhile, the suppliers are looking for buyers to increase their market share. The vast amount of information on the Internet causes a great deal of problems for both the consumer and the seller. Searching the vast information on the Internet, a task executed online is not only time consuming but boring as well. Intelligent s are best suited for this type of task. Intelligent s will surf through the clutter on the Internet, resulting in the selection of specific information of interest to the user. The s speed up the process of locating items on the Internet and leave the users more free time to do more productive or enjoyable tasks. 2. Intelligent s The definitions of an fall everywhere along a continuum, from simple macros in which the user enters a few parameters to truly intelligent s which demonstrate learning abilities and artificial intelligence. An [2],[3] is a software entity with some degree of autonomy, carries out operations on behalf of the user or another program, and thus, represents or has knowledge of the user's goals or wishes. s are the user's personal representative or assistant on the Internet carrying the user's identity, access rights (permissions) and responsibilities. The s acts on behalf of the user at the user's request or using some agreed user protocol. s communicate with their peers by exchanging messages on behalf of the user [4]. While s can be as simple as subroutines, typically they are larger entities with some sort of persistent control (e.g. distinct control threads within a single address space, distinct processes on a single machine or separate processes on different machines). 2.1 s on the Internet (Distributed network) s on behalf of the user freely roam the Internet or information superhighway searching for relevant information or services which are of interest to the user. These s do not take actions on behalf of the user, they provide the user with the information, and the user is free to act otherwise. However, a higher level of

2 sophistication involves service performing s which execute the specific tasks on behalf of the user (e.g. find me the cheapest flight to Paris or arrange a meeting with the managing director someday next week). Finally there are predictive s that volunteer information or services to the user without explicitly being asked [3]. For example, the may monitor news groups on the Internet and return discussions that it believes to be of interest to the user. s execution on the Internet raises a host of issues which needs to be addressed if s processing is to be accepted by the computing community. For example, how ambitious should the representing a user be? And, how much should the user trust the especially when one considers that the s are delegated important decisions such as making financial commitments on behalf of their user. How about the security of such a transaction? In traversing the network searching for information, the s will need to have access permission at every site they visit in addition to the site authenticating the identity of the user that the purports to represent [4]. This is to prevent s masquerading as some other user's representative. The need to guard against wayward programs that install viruses, compromise host, or pilfer the database. The resources used by the at each site needs to be charged to the user. These and a host of many others are the issues which must be addressed. Finally, this model gives cleaner and simpler failure semantics. In a regular distributed system, the state between the client and the server is normally distributed. A failure half-way makes it tricky to ensure that both parties roll back their state to maintain consistency. Alternative architecture model such as server-server models shown in Figure 2 are also in existence. In the server model, the server co-ordinate activities of all local s arriving from other hosts. A number of -based systems have been in implemented, tested and the simulation results obtained point to the feasibility of s processing. The systems are implemented either as interpreted program code WAVE [7], [8], TACOMA [9], Telescript [10], [11], [12] and Tcl [13] or compiled program code [14], [15]. In the interpreted program code version, the execution engine runs on each host, receiving, interpreting user program (string) and executing the user command. In the compiled program code version, the compiled program code (compiled code) is loaded on to each host. The user commands (compiled program code) is loaded and executed by the execution engine. 2.2 s Motivation Four main motivations exist for the paradigm. The first, performance improvement is an important motivation. In Figure l (a), we have the common situation where a client and server are located on two different nodes. The sever will typically manage some data. To carry out computation, several messages will often have to be transferred back and forth between the two parties. There are performance advantages in moving the client to the server side of this connection. The amount of network interactions can be reduced as a result of co-locating service requesters and service providers heavily engaged in communication. This scenario is illustrated in Figure 1 (b). A second motivation for the model is that it is intuitive. s acting more or less independently on behalf of somebody are a well known concept. Figure l (c), represents a similar situation as in Figure l (b), but now the client itself does not move. It sends a representative to the server side to do the actual computation. The will typically reduce the amount of data that has to be transferred over the network. In this example, the result is finally sent back the client. A third motivation for the paradigm is that communication is reduced to a site local issue. All that is needed is to move, or co-locate, s that wish to communicate. Locality of s is not hidden, but communication channel are. 3. HYDRA Computer Architecture HYDRA: Parallel and Distributed Swarm computer architecture for execution [15] is based on the Swarm computer architecture [14]. The HYDRA computer depicted in Figure 3, consists of four modules: An Access Point (AP) through which the user can access the HYDRA computer to issue tasks and collects the results after task termination, Routing Module (RM) responsible for routing s to local destinations (PM), Processing Module (PM) is the execution engine and also the destination of an executing, and Network Router (NR) that connects hosts in the network, routes s between the hosts and ensures message delivery to correct destinations. s navigate a network of hosts using hosts addresses (PM and node), executing a task,

3 and when the task terminates, the s collects the result and send to the user. The HYDRA computer is a software implementation coded in the C programming language running in the UNIX/Linux environment using the TCP/IP transport protocol. PM2 host2.. PM n host n. The NR then starts the RM running at each host, creates a connection point for Access Point (user interface) at each host and then, enters a continuous loop monitoring network hosts, RM and AP connections for user connection. The HYDRA computing environment consists of several hosts connected together by an interconnection scheme to provide a distributed computer system in which s execution can be investigated. Traffic Monitor Server Server Traffic Monitor Dock Master Dock Master Navigation Navigation Figure 2: Server Server model Communication Router (CR) Access Point (AP) Network Router (NR) Network Router (NR) Access Point (AP) Routing Module (RM) Routing Module (RM) Transport Protocol Processing Module (PM 1) Processing Module (PM n) The Communication Router (CR) in the HYDRA computer connects hosts in the network and reliably delivers messages to the hosts. During the start up, the CR connects the hosts, starts the routing module at each host, establishes a listening end-point at each host and monitors message arrival at the end-points. The CR identifies both local and remote destinations to reliably deliver messages to these destinations. The CR uses an interconnection scheme [15] to connect hosts in the network, assign host numbers to host names and establish communication channels among the hosts. The hosts in the network are grouped into domains, and the domains interconnected together to form a distributed network. Hosts in a domain are directly connected to every other host in that domain and within each domain, a single host acts as gateway handling out-of domain s i.e. s crossing domain boundaries. After the hosts in the domain have been connected, the CR at each hosts opens the network file containing names of all hosts in the network, reads the hostnames and maps hostnames to PM i.e. PM1 host1, Figure 3: The HYDRA Computer Architecture 4. The NET-COMPUTER Computer System The NET-COMPUTER implementation is based on HYRDRA computer architecture developed specifically for executing Intelligent s. The motivation for developing the NET-COMPUTER is to shift the computing resources to the Internet and provide users with a simple input/output device through which users can access the Internet computing resources. Current estimates indicate that: Nearly 90% of the users run document processing applications. In an 8-hour office working day, effective computer usage is about 40% (the time when the computer is actively in use). In the home environment, this is less than 20%. The ordinary user uses less than 10% of the computing

4 resources e.g. processor speed etc (except for scientific applications and data processing applications e.g. payroll processing). From the above statistics, one can rightly argue that there is no point in providing the user with a powerful machine which spends most of its time idle and thus, forms part of office/home decoration. Thus, why not concentrate the computing resources on the Internet and provide the user with a simple input/output terminal through which to access the computing resources. The user input/output terminal can be as simple as a pager, ipad, tablet, personal organizer, etc. Network access node Internet Servers HYDRA computer architecture User interface execution, a node file containing node distribution (node reference number and links to other nodes in the network) is prepared for each PM. At each node, is stored a unique number which is accessed by s visiting the node. The aim here is demonstrate the ability of the s to access a stored data item. The unique number is stored in the persistent workspace 3 (PW3) which is accessible and can be modified by s belonging to the same task. The unique number stored at each visited destination is retrieved by the s and used in a simple addition routine. The node file is then uploaded to the respective PMs. Internet computing resources (a) (b) host 2: ainur Figure 3: Configuration of the NET-COMPUTER host 1: yavanna The NET-COMPUTER is a collection of Internet computers (servers) running user programs on demand. Both the hardware and software resources reside on the Internet. The user need not be aware of the exact location of the resources. The HYDRA computer architecture attached to the Internet servers shown in Figure 3 (a) receives and executes the s. The Internet servers are connected together to form the Internet computing resources shown in Figure 3 (b). The Internet or cloud provides the computing resources. The user interface to the NET-COMPUTER is a simple device with less complexity in hardware and software since the computing resources reside on the Internet. 5. Simulation Test The NET-COMPUTER was simulated in a network of three hosts: ainur, kira and yavanna connected as shown in Figure 4. The hosts: are connected to networks A, B and C shown in Figure 5 and, run UNIX operating system. The objective of the simulation test to test the following s characteristics: 1. Mobility: the ability of s to navigate the network and reach destination address (host). 2. Demonstrate the ability of the s to carry out useful work. 3. Test the ability of the gateway hosts to route s to host in the other domains. In addition, the user is at liberty to configure and select the hosts on which to run HYDRA computer system as explained in section 3.0. This is done through selection of hosts in each domain. At the beginning of Access Point Figure 4: Simulation test network In the test routine, the s load a value of Ahex to the s workspace 2 (AW2), the s then navigate the network to node 1 in all PMs. On reaching its destination, the s retrieve the number stored at PW3 and add it five times to the number carried by the s and store the result in PW3 as shown in table 1. The s then navigate the network to node 1 in PM3 where each add contents of its AW2 to the contents of PW3 in PM3 and store the result in AW3. The final result is stored in PW4 in PM3 shown in table 2. The test program is described as follows: host 3: kira AW2, AW3, AW4 - variables in the workspace. PW3, PW4 - persistent workspace (nodal variables) 01 Assign a value to AW2. 02 Insert destinations PM (PM1-3) address in the destination list (DL). 03 Insert destination node address (node 1) in the destination list (DL). 04 Spawn to destination address in the destination list. 05 Load a value from persistent workspace PW3 to AW3. 06 Add contents of AW3 to contents AW2 and store the result. 07 Add one (1) to contents of AW4. 08 Test if contents of AW4 equal If contents of AW4 are less than 5, go back to 06.

5 Figure 5: The NET-COMPUTER execution network 10 Store the contents of AW2 in persistent workspace PW3. 11 Store the value 0 in persistent workspace PW4. 12 Insert destination PM (PM3) address in the destination list. 13 Insert destination node (node 1) address in the destination list. 14 Spawn to the address in the destination list. 15 Load a value from persistent workspace PW3 to AW3. 16 Add contents of AW3 to contents AW2 and store the result in AW3. 17 Load a value from persistent workspace PW4 to AW4. 18 Add contents of AW3 to contents AW4 and store the result in AW3. 19 Store the contents of AW3 to persistent workspace PW4 20 Terminate. 6. Simulation Test Results During the simulation test, s traversed the network using destination addresses carrying a data item. At each visited PM, the s retrieved data item stored at the site and added the data to the data item carried by the, executing the addition loop five times and stored the final result at the visited site. At the end of the addition, the retrieved final result stored and stored it in s workspace. The then, traversed the network to PM3. At the destination PM3, the retrieved the stored data, added it to the data carried in the s workspace and stored the result at PM3. The results are presented below in tables 1 and Conclusion The simulation tests were carried out to test s ability to traverse the network using destination (PM) addresses, access a data item stored at a site, manipulate the data item and store the final result at the site or communicate the data item (result) to a different location. In the NET-COMPUTER, the computing resources are distributed over a wide geographical area. Thus, s navigate the network to reach their execution environment and the user at the end of the task execution to deliver the results. The hosts used in the simulation test are connected to physically separate networks. We can therefore conclude that the s traversed the network to reach their destinations. Thus, confirming s navigation ability in the HYDRA computer architecture. The simple addition exercise demonstrated that s can be used to carry out some useful work such as manipulating user program. The ability of the s to access stored information (data), manipulate the data and carry the data to different locations within the network, demonstrates the s ability to collect/access, modify data and communicate the same. The simulation test results have demonstrated the feasibility of the NET- COMPUTER based on HYDRA computer architecture and Intelligent s. However, more research work is needed to adapt the HYDRA architecture for deployment on the Internet. Table 1: s add contents of AW2 to the contents of PW3 PM PM1 PM2 PM3 Number Value in Loop count PW3 Value in A A A AW2 Sub total C D E 1 Sub total E Sub total Sub total A A AA 4 Sub total C C CE 5 Total in PW C C CE Moreover, the delays experienced when sending s and during execution of user program and, the fact that the participating hosts run other programs in addition to HYDRA computer architecture has not been taken into account. These are matters and others will be the subject of further investigation in the next phase of the research. Table 2: s add contents of AW2 to the contents of PW3 and store the results in PW4 in PM3 Loop PM1 PM2 PM3 Sub Total in count Total PM C 4 (AW2) CE (PW3) C 9 (AW2) CE (PW CE (PW3) CE (AW2) 2 (AW3) 7 (AW3 C(AW3) 2 (PW4) (PW4) C 5 (PW4) The NET-COMPUTER aims to provide the user with the ability to log to the system, issue tasks to be executed,

6 log off and at a later time log to the system to collect the task results. Thus, providing an execution environment similar to the mail server described in section 4.0. The simulation was run with the user was connected online, the next phase of the research will investigate the feasibility of the user to log to the NET-COMPUTER, issue a task and log off. At a later time, the user can log to the NET-COMPUTER to collect the results of the task execution. The combination of Intelligent s and the NET- COMPUTER might just provide the mechanism for furthering Electronic Commerce on the Internet. However, a lot of work still needs to be done to make s processing on the Internet feasible. Issues such as security, accessibility, fault-tolerance, authentication, charges for consumed resources, etc still need to be adequately addressed. 8. References white paper. General Magic Inc Latham Street, Mountain View, California [12] White James E. Telescript Technology: Mobile s. General Magic white paper. General Magic Inc Latham Street, Mountain View, California [13] Gray Robert S. Tel: A Transportable Systems. Department of Computer Science, Dartmouth College, Hanover, New Hampshire November 17, [14] Luciano De Errico. Based Distributed Parallel Processing. A Dissertation for the degree of Doctorate of Philosophy, Department of Electrical and Electronic Engineering, University of Surrey. January 17, [15] Marcel O Odhiambo. HYDRA: Parallel and Distributed Swarm Computer Architecture. ROVPIA 99 International Conference on Robotics, Vision and Parallel Processing for Automation July 16 th 18 th, Ipoh, Malaysia. [1] Indermaur Kurt. Baby Steps. BYTE, March 1995, pp [2] Russell Stuart J, Norvig Peter (2003). Artificial Intelligence: A Modern Approach ( (2 nd ed.), Upper Saddle River, New Jersey: Prentice Hall, ISBN , ( chpt. 2 [3] Intelligent s. MIT Encyclopedia ( [4] Genesereth Michael R, Ketchpe Steve P. Software s. Comms. of the ACM, July 1994, Vol. 37, No.7, pp [5] Jennings Nick, Wooldridge Michael. Soft-ware s. IEE Review, January 1996, pp [6] Wayner Peter. Free s. BYTE, March 1995, pp [7] Sapaty PS, Borst PM. An Overview of WAVE Language and Systems for Distributed Processing in Open Networks. University of Surrey, Department of Electronic and Electrical Engineering, Technical report, June [8] Sapaty PS, Borst PM.. The WAVE Model and Architecture for Network Processing in Open Networks. University of Surrey, Department of Electronic and Electrical Engineering, Technical report, Undated. [9] Johansen Dag Joha SE, Robert Renesse, Scheneider Fred B. An Introduction to the TACOMA Distributed System Version 1.0. Technical report: University of Tromso (Norway) and Cornell University (USA), June 24, [10] White James E. Telescript Technology: The Foundation for the Electronic Marketplace. General Magic white paper. General Magic Inc Latham Street, Mountain View, California [11] White James E. Telescript Technology: Scene from the Electronic Marketplace. General Magic