CN101308463B - Self-adapting method of grid environment facing to proxy programming calculation task - Google Patents

Abstract

The self-adaptive method for agent programming and computing tasks in the grid environment is based on the existing open grid architecture. By deploying agent management service and agent message service, two grid services that meet the FIPA specification and WSRF specification, and deploying the GA container , to support the management and movement of agents. By extending the existing grid job type, that is, adding a new job type "AGENT", the creation process of the task agent is fully compatible with the creation process of the existing grid task; at the same time, by defining the task of agent-oriented programming, and the agent The moving and positioning process enables the agent task to adapt to changes in the grid environment and make intelligent moving decisions.

Description

An Adaptive Method for Agent-Oriented Programming Computing Tasks in Grid Environment

technical field

The invention relates to a method for solving calculation task adaptive grid environment through a mobile agent, and belongs to the intersection field of the grid and the mobile agent.

Background technique

Grid and agent solve the problems of coordination, negotiation and collaboration in distributed systems from different angles. The grid provides a safe and reliable resource sharing infrastructure for dynamic geographically distributed virtual organizations through a series of tools, so the grid can be considered as the "skeleton" of the distributed computing environment. An agent is an autonomous entity that can adapt to a dynamic environment and achieves its design goals through autonomous behavior. Among them, the mobile agent not only has the autonomy of the agent but also has mobility, so the agent can be considered as the "brain" of the distributed computing environment. ". The grid lacks intelligence but is safe and reliable, while the agent is the opposite. Therefore, combining the respective advantages of the grid and the agent can better solve large-scale distributed computing problems.

At present, there are solutions to apply agent technology to the grid environment to solve several problems, such as using agents to schedule grid tasks, using agents to predict the attributes of grid resources and tasks, using mobile agents to obtain grid resource information, etc. However, in In these solutions, the grid and the proxy do not make up for each other's shortcomings. The security issues of the proxy, resource monitoring, and data transmission solutions are still self-contained, and the grid platform and the proxy platform operate completely independently. Therefore, it is necessary to propose a new and integrated platform combining the advantages of both. Existing platforms such as DEINF/UFMA's MAG project propose to develop a middleware MAG on the basis of grid middleware InteGrade and agent middleware Jade to support task error tolerance. However, the task agent itself does not have intelligence in this solution. Performance, detection of task errors, and decision-making of mobile behavior are all done by other agents.

In addition, grid scheduling has always been a hot spot in grid research. At present, most of the task scheduling algorithms in the grid come from clusters. However, the grid environment is very different from the cluster environment. For example, the grid resources are highly dynamic and unpredictable, while the cluster resources are less variable and predictable. The grid environment needs to consider the network environment rather than the cluster environment. Clusters can be ignored etc. Therefore, in addition to the scheduling algorithm, the behavior that enables tasks to adapt to changes in the grid environment is the key to ensuring stable and fast execution of tasks.

Contents of the invention

Technical problem: The purpose of the present invention is to provide an adaptive method for agent-oriented programming computing tasks in a grid environment, to solve the problem of efficient integration of grid environment and mobile agents, and to solve the task agent's self-adaptation in the grid environment question.

Technical solution: The self-adaptive method for agent-oriented programming computing tasks in the grid environment proposed by the present invention is characterized in that the method adds a new job type "AGENT", so that the creation process of task agents is the same as that of existing grid tasks Fully compatible; by deploying the G-A container, the container provides agent management services and agent message services that meet FIPA specifications and WSRF specifications, supports the movement and positioning process of task agents, and enables task agents to adapt to changes in the grid environment and move intelligently The specific process of decision-making is as follows:

Step 1: Computing program design for agent programming:

11. Write the calculation behavior, which contains the code of the calculation task,

12. Write the mobile policy behavior, which is executed in a thread independent of the agent, senses the change of the environment and selects the target of the movement,

13. Write the serialized intermediate data behavior, define the data that needs to be serialized in the program, and schedule the behavior to execute periodically or before moving,

Step 2: Create a task agent:

21. Write a job description file,

22. Task scheduling,

23. Task submission:

i. The grid client program submits the job to the resource's job management factory service;

ii. The job management factory service verifies the user agent certificate and authorizes the user's actions,

If the verification and authorization are successful, go to the next step, otherwise the client security verification fails;

iii. The job management factory service parses the job description file, and if data needs to be prepared, call the grid data transfer service to transfer the required data to the specified directory;

iv. The job management factory service finds that the job type is "AGENT", and transfers control to the agent management service;

v. The proxy management service loads the code of the proxy. If the loading is successful, it will return the global proxy ID, otherwise it will return the failure to create the proxy;

vi. The agent management service updates the agent information table and synchronizes the information with the main agent management service;

vii. The task agent is created;

Step 3: Mobile Task Agent:

31. The agent's move policy behavior triggers the move,

32. The agent interrupts the current execution and executes the behavior of serializing intermediate data,

33. The agent informs the local agent that the management service needs to move,

34. The local agent management service informs the remote agent management service that an agent will be moved to him through the agent message transmission service, and at the same time calls the grid data transmission service to transmit the data related to the agent to the remote resource,

35. After the local agent service receives the message of successful transmission, it sends the ACL command message to start the agent to the remote agent management service,

36. The remote agent management service rebuilds the agent according to the transmitted agent information, and returns a success message if successful, and updates the agent information table; otherwise, returns failure,

37. When the local agent management service receives a successful message, the information of the local agent is deleted, and the information table of the local agent is updated at the same time; otherwise, the resource is reselected and go to step 34,

Step 4: Positioning of the task agent:

41. According to the task proxy ID, the user first queries the proxy location from the proxy information table cache of the host where the user is located, and if the query is successful, then go to step 44; otherwise, go to the next step;

42. The user requests the location of the task agent from the agent management service in the main G-A container according to the task agent identifier;

43. The main agent management service queries the agent information table, and if the query is successful, returns the location of the task agent to the user, otherwise returns a failure message;

44. The user sends a request for service information to the task agent according to the obtained position of the task agent, and if the request is successful, the cache of the agent information table of the host computer where the user is located is updated. Otherwise go to step 42.

The implementation and deployment steps of the agent management service and agent message service of the FIPA specification and the WSRF specification are as follows:

①Proxy management service: The characteristic of this service is that its function meets the FIPA specification, and its implementation meets the WSRF specification. The functions to be implemented are as follows:

A. Support proxy creation, including:

A1. Support the invocation of job management middleware, and create task agents according to task attributes;

A2. Support agent recovery from backup intermediate data;

A3. Reconstruction of mobile task agents;

B. Support proxy deletion;

B1. Deletion of failed proxy data;

B2. After the agent moves successfully, delete the local agent data;

C. Support automatic maintenance agent information table;

C1. Update of agent information after agent creation,

C2. Update of agent information after agent deletion,

D. Support agent location query;

② Deploy the main G-A container, which serves as the host container for the main resource monitoring and discovery service and the main agent management service;

③ Deploy child G-A containers, they must point out the host container of the main resource monitoring and discovery service and the main agent management service.

Beneficial effects: the adaptive method for agent-oriented programming computing tasks in the grid environment proposed by the present invention has the following advantages compared with the existing method for solving computing tasks adaptive grid environment through mobile agents:

1. The G-A container proposed by the present invention combines the grid and agent technology more closely, which means that the agent will directly use the existing grid security infrastructure, grid data transmission service, grid resource monitoring and discovery service, and provide Agents provide direct support for security, data transfer, scheduling, and decision-making.

2. By implementing agent management service and agent message transmission service that meet both FIPA specification and WSRF specification, compared with the current combination of the two through multi-layer middleware, the complexity of grid-to-agent technical support is simplified.

3. Design grid tasks through agent-oriented programming, allowing programmers to determine data storage behavior and move decision-making behavior, so that tasks can not only intelligently decide when to move, but also restore to the previous one that is closer after moving to the target container Execution status, to ensure the stability and effective execution of tasks.

4. Through the newly defined task type "AGENT", it is determined that the submitted task will be a mobile agent. The definition of the task description document and the submission process of the task are fully compatible with the submission process of other types of tasks such as "FORK", "MPI" and "MULT" in the grid, so as to keep a small number of programs on the existing client and server Modifications to support the management of new agent tasks.

Description of drawings

Figure 1 is a diagram of the G-A container architecture;

Figure 2 is a flow chart of creating a task agent;

Fig. 3 is a flow chart of mobile task agent;

Fig. 4 is a flow chart of positioning task agent.

Detailed ways

In order to better explain the reasons for the above-mentioned technical solution and its implementation details, each step is described in detail as follows:

Step 1: Implementation and deployment of G-A container (Grid-Agent Container):

Grid containers such as Globus Toolkit4 provide various resource sharing functions by implementing WSRF-based WEB services, including grid security infrastructure (GSI), grid data transmission, grid resource monitoring and discovery, grid job management, and others Features are still being improved. Based on this open grid service framework, it is feasible to add services that support mobile agents into the framework and directly support the management and interaction of agents. Considering that only multiple mobile agents are concerned here to solve the adaptive problem of large-scale distributed computing, it is necessary to reduce the functions of the multi-agent system. The developed G-A container is a grid container based on an open grid service framework, which can accommodate mobile agents. The specific development and deployment steps are as follows:

④Develop agent management service. The function of this service meets the FIPA specification, and its implementation method meets the WSRF specification. Specifically, the following functions need to be implemented:

A. Support proxy creation, including:

a) Support the invocation of job management middleware, and create task agents according to task attributes;

b) support recovery of agents from backup intermediate data;

c) Reconstruction of task agents that support mobility;

B. Support proxy deletion;

a) Deletion of failed proxy data;

b) After the agent moves successfully, delete the local agent data;

C. Support automatic maintenance agent information table;

a) Update of agent information after agent creation

b) Update of proxy information after proxy deletion

D. Support agent location query;

⑤Develop the agent message transmission service, the function of this service meets the FIPA specification, its implementation method meets the WSRF specification, and realizes the process of supporting ACL messages;

⑥ Deploy the main G-A container. It serves as the host container for the main resource monitoring and discovery service and the main agent management service;

⑦ Deploy the child G-A container. They must point to the host container for the main resource monitoring and discovery service and the main agent management service.

Step 2: Computing program design for agent programming

One of the disadvantages of object-oriented method-based programs executing in a grid environment is that they lack effective means of migration. Usually, the process or thread migration is expensive. For example, in JAVA, the Brakers framework can be used to support thread migration. Since the developer of the program usually knows what data is necessary for the program to resume execution, it is possible to allow the programmer to specify which data needs to be serialized. Based on these data, the instantaneous state before the agent interrupted execution cannot be completely restored, but it can be restored to " Previous" consistent state.

Drawing on the concept of agent behavior in Jade, Jade is an open source JAVA-based agent development framework. The steps to write a task agent are as follows:

⑧ Write computational behavior. It contains code for computing tasks.

⑨ Write mobile strategy behavior. It executes in a thread independent of the agent, sensing changes in the environment and selecting targets for movement.

⑩ Write serialized intermediate data behavior. Define the data that needs to be serialized in the program, and schedule the behavior to execute periodically or before moving.

Step 3: Create a task agent

In order to maintain compatibility with the existing grid job management middleware, the task agent creation process remains the same as the job submission process. In addition to "FORK", "MULTI" and "MPI", the job type also needs to support "AGENT ". When the job management factory service finds that the submitted job type is "AGENT", it will hand over the instantiation process of the job to the agent management service. The specific steps are as follows:

Write a job description file. The file needs to specify two types of job information:

A. Job attributes: job type, program language, execution program, program execution parameters, upload data file address, result output address, error output address, proxy certificate, etc.;

B. Job execution requirements: architecture, CPU processing speed, number of CPUs, available memory size, available hard disk space, etc.;

task scheduling. The scheduler obtains the information of all resources in the grid environment from the grid resource monitoring and discovery service, and selects the resource for execution according to the best matching algorithm (that is, the resource with the highest matching performance) according to the job execution requirements.

任务提交。

Task submission.

A. The grid client program submits the job to the resource's job management factory service;

B. The job management factory service verifies the user agent certificate and authorizes the user's behavior. If the verification and authorization are successful, go to the next step, otherwise the client security verification fails;

C. The job management factory service parses the job description file, and if data needs to be prepared, call the grid data transfer service to transfer the required data to the specified directory;

D. The job management factory service finds that the job type is "AGENT", and transfers control to the agent management service;

E. The proxy management service loads the code of the proxy. If the loading is successful, it will return the global proxy ID (AID), otherwise it will return the failure to create the proxy;

F. The agent management service updates the agent information table and synchronizes information with the main agent management service;

G. The task agent is created.

Step 4: Mobile Task Agent

The agent's movement policy triggers the agent's movement. The agent management is included in the grid container so that the agent can conveniently access the local resource information collected by the local resource monitoring and discovery service, and provide support for the decision-making of agent movement. The steps for agent movement are as follows:

代理的移动策略行为触发移动

The agent's move policy action triggers the move

代理中断当前执行，并执行序列化中间数据行为

The agent interrupts the current execution and executes the serialization intermediate data behavior

代理通知本地代理管理服务需要移动

The agent notifies the local agent management service that it needs to move

本地代理管理服务通过代理消息传输服务告知远程代理管理服务将要向他移动一个代理，同时调用网格数据传输服务将与代理相关数据传输到远程资源。

The local agent management service informs the remote agent management service that an agent will be moved to him through the agent message transfer service, and at the same time calls the grid data transfer service to transfer the data related to the agent to the remote resource.

当本地代理服务收到传输成功的消息后，发送启动代理的ACL命令消息给远程代理管理服务。

After the local agent service receives the message of successful transmission, it sends the ACL command message for starting the agent to the remote agent management service.

远程代理管理服务根据传输过来的代理信息重建代理，如果成功则返回成功消息，并更新代理信息表；否则，返回失败。

The remote agent management service rebuilds the agent according to the transmitted agent information, and returns a success message if successful, and updates the agent information table; otherwise, returns failure.

本地代理管理服务收到成功的消息，则将本地代理的信息删除，同时更新本地代理信息表；否则重新选择资源并转到④。

When the local agent management service receives a successful message, it deletes the information of the local agent and updates the information table of the local agent; otherwise, select the resource again and go to ④.

Step 5: Locating the Task Agent

Since the task agent will move with the change of the environment, tracking the position of the task agent is the key to ensure that the execution status and execution result of the task agent can be obtained. The process of locating the task agent is similar to the process of finding the IP address of the domain name server. The specific steps are as follows:

21. According to the task agent identification (AID), the user first inquires the agent position from the agent information table cache of the host computer where the user is located, and if the query is successful, then go to ④; otherwise, go to the next step;

22. The user requests the location of the task agent from the agent management service in the main G-A container according to the task agent identifier (AID);

23. The main agent management service queries the agent information table, and if the query is successful, returns the position of the task agent to the user, otherwise returns a failure message;

24. The user sends a request for service information to the task agent according to the acquired position of the task agent. If the request is successful, the agent information table cache of the host computer where the user is located is updated. Otherwise go to ②.

Step 6: Acquisition of task results

The task agent will finally transfer the result in the form of a file to the result output address defined by the job attribute, and the user can directly download the data according to the address.

Claims (2)

1. An adaptive method for agent-oriented programming computing tasks in a grid environment, characterized in that the method adds a new job type "AGENT", so that the creation process of task agents is fully compatible with the creation process of existing grid tasks; By deploying the grid agent container, the container provides agent management services and agent message services that meet the FIPA specification and WSRF specification, supports the movement and positioning process of the task agent, and enables the task agent to adapt to changes in the grid environment and intelligently move decisions , the specific process is as follows: Step 1: Computing program design for agent programming: 11. Write the calculation behavior, which contains the code of the calculation task, 12. Write the mobile policy behavior, which is executed in a thread independent of the agent, senses the change of the environment and selects the target of the movement, 13. Write the serialized intermediate data behavior, define the data that needs to be serialized in the program, and schedule the behavior to execute periodically or before moving, Step 2: Create a task agent: 21. Write a job description file, 22. Task scheduling, 23. Task submission: i. The grid client program submits the job to the resource's job management factory service; ii. The job management factory service verifies the user agent certificate and authorizes the user's behavior. If the verification and authorization are successful, go to the next step, otherwise the client security verification fails; iii. The job management factory service parses the job description file, and if data needs to be prepared, call the grid data transfer service to transfer the required data to the specified directory; iv. The job management factory service finds that the job type is "AGENT", and transfers control to the agent management service; v. The proxy management service loads the code of the proxy. If the loading is successful, it will return the global proxy ID, otherwise it will return the failure to create the proxy; vi. The agent management service updates the agent information table and synchronizes the information with the main agent management service; vii. The task agent is created; Step 3: Mobile Task Agent: 31. The agent's move policy behavior triggers the move, 32. The agent interrupts the current execution and executes the behavior of serializing intermediate data, 33. The agent informs the local agent that the management service needs to move, 34. The local agent management service informs the remote agent management service that an agent will be moved to him through the agent message transmission service, and at the same time calls the grid data transmission service to transmit the data related to the agent to the remote resource, 35. After the local agent service receives the message of successful transmission, it sends the ACL command message to start the agent to the remote agent management service, 36. The remote agent management service rebuilds the agent according to the transmitted agent information, and returns a success message if successful, and updates the agent information table; otherwise, returns failure, 37. When the local agent management service receives a successful message, the information of the local agent is deleted, and the information table of the local agent is updated at the same time; otherwise, the resource is reselected and the process goes to step 34, Step 4: Positioning of the task agent: 41. According to the task proxy ID, the user first queries the proxy location from the proxy information table cache of the host where the user is located, and if the query is successful, then go to step 44; otherwise, go to the next step; 42. The user requests the location of the task agent from the agent management service in the main grid agent container according to the task agent identifier; 43. The main agent management service queries the agent information table, and if the query is successful, returns the location of the task agent to the user, otherwise returns a failure message; 44. The user sends a request for service information to the task agent according to the acquired position of the task agent, and if the request is successful, the agent information table cache of the host computer where the user is located is updated. Otherwise go to step 42. 2. in a kind of grid environment according to claim 1, it is characterized in that the agent management service of FIPA specification and WSRF specification and the step of realization and deployment of agent message service are as follows: Agent management service: The feature of this service is that its function meets the FIPA specification, and its implementation meets the WSRF specification. The functions to be implemented are as follows: A. Support proxy creation, including: A1. Support the invocation of job management middleware, and create task agents according to task attributes; A2. Support agent recovery from backup intermediate data; A3. Reconstruction of mobile task agents; B. Support proxy deletion; B1. Deletion of failed proxy data; B2. After the agent moves successfully, delete the local agent data; C. Support automatic maintenance agent information table; C1. Update of agent information after agent creation, C2. Update of agent information after agent deletion, D. Support agent location query; Deploy the main grid agent container, which serves as the host container for the main resource monitoring and discovery service and the main agent management service; Deploy child grid agent containers, which must point to the host container for the main resource monitoring and discovery service and the main agent management service.

Images