JPH06274432A - System and method for managing distributed computer system - Google Patents

Abstract

PURPOSE:To effectively utilize distributed system resources and to improve the fault countermeasure, in the distributed computer environment. CONSTITUTION:In the system management system by a client/server system of plural computers coupled by a network, this system is provided with a program execution managing means provided with a means for receiving a connection request from a client to a server, a distributed system management control means for receiving the connection request through its means, and a distributed system resources information data base. Also, the distributed system management control means is provided with a server selecting means for selecting the computer having an optimal server by using information of the data base, and a means for informing the program execution managing means about a result of its selection, and the program execution managing means is provided with a means for requesting connection preparations including server program actuation to the computer having the server, in this distributed computer system management system.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a technique for effectively utilizing distributed system resources and dealing with failures in a distributed computer environment.

[0002]

2. Description of the Related Art As a conventional technology of computing resource information service (naming service) in a distributed system, Sun Microsystems NIS (Network Info) is known.
(Mation Service). This is a system that centrally manages static management information such as user registration information, computer address information, and network configuration information (UNIX Communicatio).
n Notes 36, UNIX MAGAZINE
(1991.6), p61-71. And UNIXCom
communication Notes 37, UNIX
MAGAZINE (1991.7), p69-8
1. ).

Hewlett Packard's HP Open
View or SunNet from Sun Microsystems
A network monitoring system represented by Manager provides a mechanism for collecting dynamic information such as network traffic information. Hewlett-Packard's location broker is RPC (Remote
This is a name service that manages the location of the server network-wide in a client / server type application using the Procedure Control (the forefront of distributed computing, Nikkei Electronics, no. 502 (1990.6), p137-
138).

As a conventional load balancing technique, there is a task broker of Hewlett-Packard. When a request for processing is made, a method is used in which a requesting client machine inquires of all server machines in the network about the computer operating status (the front line of distributed computing, Nikkei Electronics, no. 502 (19).
90.6), p145-146).

OS as an integrated distributed system management technology
F (Open SoftwareFoundation)
n) DCE (Distributed Comput)
ing Environment (Full picture of distributed computing environment DCE, Computer T
oday, no. 44 (1991.7), p17-4.
5).

[0006]

In order to manage a distributed system, a naming service that manages distributed computing resource information, load distribution that efficiently uses the entire distributed system, and a system that prepares for failure occurrence are provided. Elemental technologies such as multiplexing are required. Further, in order to make a user think of a complicated distributed system configuration as if it were one computer, it is necessary to integrate these elemental technologies.

[0007] The conventional technology of the naming service, Sun Microsystems NIS (Network Inf)
information service) is limited to a system that manages static management information such as user registration information, computer address information, and network configuration information. Considering load balancing and troubleshooting, dynamic information such as the operating status of each computer and network is also necessary, but such information is not managed.

Hewlett Packard HP Open
View or SunNet from Sun Microsystems
A network monitoring system typified by Manager provides a mechanism for collecting such dynamic information, but the collected information is mostly related to the operating status of the network, and the more important operating status of each computer (CPU Load, memory usage, disk usage, process execution status, etc.) is not collected in the standard specifications.

In the Hewlett-Packard location broker, the user (client) does not need to be aware of the location of the server, but does not fully consider the load distribution of the distributed system when selecting the server.

A task broker, which is a conventional technique for load balancing, is intended for load balancing of program execution of batch processing type, and is not suitable for a client / server type system that often involves interactive interaction with a user.
Further, when a processing request is made, it is necessary for the requesting client machine to make an inquiry about the computer operating status to all the server machines in the distributed system.

DC which is an integrated distributed system management technology
E (Distributed Computing E)
nvironment) is a system that integrates basic distributed system management element technologies such as communication means, naming services, and security services.
With this alone, it is not possible to perform management including load balancing and troubleshooting.

[0012]

A first aspect of the present invention is a system management system based on a client-server system of a plurality of computers connected by a network, wherein a client-server system requests a connection request from a client to a server. And a distributed system resource information database; and a distributed system resource information database, the distributed system management comprising: a program execution management unit including a receiving unit; a distributed system management control unit that receives a connection request from the client via the program execution management unit; The control means includes a server selection means for selecting a computer having an optimum server using the information in the distributed system resource information database, and means for notifying the program execution management means of the selection result of the server selection means. Execution management means is the server It is a distributed computer system management method and a means for requesting connection preparation comprising ram start.

According to a second aspect of the present invention, the distributed system management control means comprises a failure management table for registering the connection status of the clients and servers of the entire distributed system, and the program execution management means is a failure related to a computer managed by itself. The distributed computer system management method described above is provided with a management table.

According to a third aspect of the present invention, a plurality of distributed system management control means are provided in a distributed system, and the program execution management means issues a connection request to all distributed system management control means in the distributed system at start-up, and , A means for determining one distributed system management control means in a predetermined order and another distributed system management control means to be a backup in case of failure of the distributed system management control means, the two distributed system management control means Is the distributed computer system management method described above in which each has a copy of the program information managed by the program execution management means.

According to a fourth aspect of the present invention, a plurality of distributed system resource information databases are provided in a distributed system, and one or more remaining distributed system resource information databases are reserved for the master distributed resource information database, and the master A backup copy of the distributed system resource information database is provided, and when there is a data update request, the master distributed system resource information database is updated, and then a request is sent to the backup distributed system resource information database. The database is updated, the distributed system resource information database is given an ID number that is updated every time there is a data update, and when a failure occurs in the master distributed system resource information database, the backup distributed system resource information database Has the latest ID number and The data update request is a distributed computer system management method was decided to continue the service processing of the distributed system resource information database by selecting one in response to the new master of the distributed system resource information database first.

[0016]

OPERATION AND EXAMPLE An example in which the program execution management means, the distributed system management control means, and the distributed system resource information database are implemented using the C language on UNIX will be described. ONC / RPC version 4 from Sun Microsystems was used for communication between the constituent modules. The management control target is a client / server type program including X windows. X Window System is UNIX
It is used as a standard window system in various models, and most of the existing application programs can be used in this system.

FIG. 1 shows a mechanism for managing distributed computing resources by the distributed system management control means. The function of the distributed system management control means is divided into an information collection system (hatched part in the figure) and a system control system (hatched part in the figure). The system control system provides the user with an efficient computer environment by using the network status information collected by the information collection system. Conventionally, a user specifies a computer in which a server exists and connects to it, or switches to a server of another computer when a failure occurs. However, under the control of the distributed system management control means, the user can use the system without being aware of the location of the server or the occurrence of a failure.

Next, the information collecting system will be described. The computational resource information includes static information such as user information and computer placement information that is predetermined when the system is constructed, and dynamic information that changes from moment to moment such as the load on the computer and the usage status of the disk. Most existing computing resource information services handle static information, but here, the computing resource information including dynamic information is centrally managed by the distributed system resource information database, and load balancing and automatic failure are performed. We considered to perform distributed system control such as recovery in an integrated manner.

A standard network management protocol SNMP (Simple Network) is used for collecting dynamic information.
A network monitoring system using a Management Protocol was used. As shown in Fig. 1, the information collected by the agents allocated to each computer is
The network management station has a mechanism for collecting and displaying. Information collected by the agent is a management information base (MIB, Management Information Form).
function base) is defined as standard.
However, considering distributed system control such as load balancing and automatic failure recovery, standard MIB alone is not enough, so information such as computer usage status, disk usage status, and program operating status is extended to MIB. Defined.

The dynamic information of the entire distributed system collected by the network monitoring system in this manner is periodically written in the distributed system resource information database as shown in FIG. The existing computational resource information service whose main purpose is data reference cannot support frequent data updates from the dynamic information gathering system, but here is a client / server distributed system resource information database improved on this point. It was created using the standard database system dbm. The machine where the distributed system resource information database is located has a distributed system resource information server that directly manipulates data, and processes requests such as reference, addition, modification, and deletion of data from clients.

Next, the system control system will be described.
Each computer under the control of the distributed system management control means is provided with a program execution management means for managing the program start-up on each computer and for troubleshooting. The program execution management means is an interface to the distributed system management control means from the viewpoint of the user or the application program. That is, the user or application program can use the computing resources and control functions of the entire distributed system managed by the distributed system management control means as long as they know how to communicate with the program execution management means in their computer.

The operation of the system control system when the user program (client) requests connection to the server will be described with reference to FIG. When the client on the computer A requests the program execution management means to connect to the server (1), the connection request from the client is transmitted to the distributed system management control means interface on the computer B via the program execution management means ( 2). In the distributed system management control means, after the access right authentication means authenticates the access right to the server (3), the server selection means uses the information of the distributed system resource information database at that time to have an optimum server. Select a computer (4). Further, the failure countermeasure means registers in the failure management table described later in preparation for the occurrence of a server failure (5),
The program execution management means on the client side is notified of the server selection result (6). The client-side program execution management means requests the server-side program execution management means in the selected computer C to prepare for connection such as server program startup (7), and after preparation is complete (8).
(9) The client is notified of the connection destination (10). As a result, the client connects to the server and starts processing (11). Below, the main points of the technique which comprises each module in a distributed system management control means are demonstrated.

First, the management of access to computing resources will be described. The access right authentication means is composed of a database for access management to a computing resource such as a server program managed by the distributed system management control means, and an interface for operating the database. The access management database is composed of an access information database that describes a set of computational resources and groups that can access it, and a user information database that describes a set of users and security groups to which the users belong. When a connection request from the user program (client) to the server program arrives, the user information database is first referenced to check the security group to which the user belongs. Next, referring to the access information database, a computer having a server program accessible by the security group can be selected. In this way, after the computing resources available to the user are authenticated, the server is selected from the selected plurality of candidates. The access management database is included in the distributed system resource information database.

The algorithm for server selection will be described. The server selecting means obtains the evaluation value (E) using the empirical formula shown below for each computer having the server to be selected, and selects the computer having the largest evaluation value. The various data used for the calculation of this empirical formula are gathered in the distributed system resource information database by the information gathering system. E = k1 ・ Broad / Amips + k2 ・ Cfmem
+ K3 · Dactive Here, Amips: Typical processing speed of a computer (MIPS
Value) Bload: Load average of cpu in the past 1 minute Cfmem: Amount of free memory in main memory Dactive: Whether server is started (1
or0) k1, k2, k3: coefficients.

Fault countermeasures will be described. In preparation for a failure of a program under management, the distributed system management control means has a distributed system resource information database with a connection status list (failure management table) of clients and servers of the entire distributed system. Further, the program execution management means has a partial failure management table for only the computer managed by itself. The failure countermeasure means manages the connection between the client and the server in the distributed system by operating these management tables.

In troubleshooting, three operations are performed: deleting the corresponding entry from the failure management table, registering it in the failure history list, and scheduling a new server. This operation is shown in FIG. When the failure of the program under management is detected (1), the program execution management means on the server side notifies the distributed system management control means interface of the failure occurrence (2), and deletes the entry from its own failure management table (3). ). In the distributed system management control means,
The failure countermeasure unit deletes the related entry from the failure management table in the distributed system resource information database (4), and registers the entry in the failure history list used when the server is rescheduled (5). On the client side, the failure is detected because the server does not respond (6), this is notified to the program execution management means on the client side (7), the entry in the failure management table is deleted (8), and the error is sent back to the server. Request a connection (9) (10).
Whether to reconnect to a new server when a failure occurs is selected by the client requesting the connection. Subsequent processing is the same as normal server selection.

Next, multiplexing of distributed system management control means and failure countermeasures will be described. The distributed system management control means was multiplexed in order to improve reliability and response by localizing processing. Here, the mechanism will be described. When the processing is concentrated on one distributed system management control means, it becomes a bottleneck and the efficiency of the entire system is reduced.
In order to solve this, a configuration is adopted in which a plurality of distributed system management control means share and process requests from the program execution management means of each computer.

This was realized as shown in FIG. The program execution management means issues a connection request to all distributed system management control means in the distributed system at the time of startup.
In order of quick response, the distributed system management control means (hereinafter referred to as "binder"), which serves as a caretaker of itself, and the distributed system management control means (hereinafter referred to as "safety"), which serves as a backup system in the event of a binder failure, are determined. Since this determination method is completely arbitrary, a certain distributed system management control means may serve both as a binder and a safety for different program execution management means. Since the binder and the safety have a copy of the program information managed by the program execution management means in charge, the request from the program execution management means can be processed consistently by switching to the safety even if the binder fails.

Finally, the multiplexing of the distributed system resource information database and failure countermeasures will be described. Each distributed system resource information database that is multiplexed to ensure fault tolerance performs two types of operations, master and slave, depending on the situation. Only one distributed system resource information database serving as a master exists in a distributed system under management, and only the master distributed system resource information database can be updated. There is at least one distributed system resource information database that is a slave, and the distributed system resource information database of a slave can be referenced but cannot be updated by anyone other than the master. When there is a data update request, the master updates its own database and then sends a request to all slaves to update the replica databases of the slaves.

In order to maintain the consistency of the distributed system resource information database, which is a problem in multiplexing, an identification number (hereinafter referred to as I
(Referred to as D). When updating the data, the master checks the ID held by the slave database and performs the update process only when the ID matches the master ID.
If they do not match, copy the entire master's database to the slaves so that the data between the replicas is consistent.

FIG. 5 shows the operation of automatic failure recovery when a failure occurs in the master. First, the master is operating on the computer A shown in FIG. The failure of the master is detected by the client that issues a request to the master distributed system resource information database. If there is no response from the master, the client that issued the request determines that a failure has occurred (1), and performs broadcast communication to search for another slave (2). Among the responding slaves, the slave of computer B having the latest ID number and responding to the broadcast communication from the client promptly is selected as a new master (3), and a data update request is issued to this (3). 4). Finally, the new master on the computer B copies the updated data to the remaining slaves (5) and continues the service processing of the distributed system resource information database.

[0032]

According to the present invention, the position of the server can be controlled by the distributed system management control means having the program execution management means and the server selection means for selecting the computer having the optimum server by using the information of the distributed system resource information database. You can run the program unknowingly,
It became possible to consider the load distribution of the system when selecting the server.

Further, the distributed system management control means has a failure management table for registering the connection status of the clients and servers of the entire distributed system, and at the same time, the program execution management means creates a failure management table for the computer managed by itself. With the provision, the failure can be automatically detected and a different server can be selected and switched, so that the user can proceed with the work without being aware of the failure.

The program execution management means is provided with one distributed system management control means and means for determining another distributed system management control means to be a backup when the distributed system management control means fails. Even when the distributed system management control means fails, it is possible to continue the processing for each program execution without delay.

The distributed system resource information database also includes one distributed system resource information database and one or more other distributed system resource information databases that serve as a backup in case of failure of the distributed system resource information database. Even if the distributed system resource information database fails, the distributed system resource information database can be switched over to automatically continue the distributed system resource information service, and the processing can continue without delay for each program execution. It has become possible.

[Brief description of drawings]

FIG. 1 shows the overall configuration of this system.

FIG. 2 shows an operation of processing a connection request from a client to a server under the control of distributed system management control means.

FIG. 3 shows a failure countermeasure operation for automatically switching servers when a server failure occurs under the control of distributed system management control means.

FIG. 4 shows a method of multiplexing distributed system management control means.

FIG. 5 shows an automatic failure recovery method for a distributed system resource information database.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Hiroaki Okahara 1-2-1, Marunouchi, Chiyoda-ku, Tokyo Nihon Kokan Co., Ltd. (72) Issei Nakamura 1-2-1, Marunouchi, Chiyoda-ku, Tokyo Nippon Steel Tube Co., Ltd.

Claims (4)

[Claims] 1. A system execution system by a client-server system of a plurality of computers connected by a network, the program execution management means comprising means installed in the computer and receiving a connection request from a client to a server. And a distributed system management control unit that receives a connection request from the client via the program execution management unit, and a distributed system resource information database, and the distributed system management control unit is a distributed system resource information database. The information processing apparatus includes a server selecting unit that uses information to select a computer having an optimum server and a unit that notifies the program execution managing unit of the selection result of the server selecting unit, and the program execution managing unit is a computer having the server. Request connection preparation including server program startup A distributed computer system management method having means for performing. 2. The distributed system management control means comprises a failure management table for registering the connection status of clients and servers in the entire distributed system, and the program execution management means comprises a failure management table for a computer managed by itself. The distributed computer system management method according to claim 1, wherein 3. A plurality of distributed system management control means are provided in a distributed system, and the program execution management means issues a connection request to all distributed system management control means in the distributed system at the time of startup and has a predetermined order. And a means for deciding one distributed system management control means and another distributed system management control means to be a backup when the distributed system management control means fails, and the two distributed system management control means execute program execution. 2. The distributed computer system management system according to claim 1, wherein each of the distributed computer system management systems has a copy of the program information managed by the management means. 4. A distributed system resource information database of a master, wherein a plurality of the distributed system resource information databases are provided in a distributed system, and one or more remaining distributed system resource information databases are reserved for a master distributed resource information database. When a data update request is made, the master's distributed system resource information database is updated, and then a request is sent to the spare distributed system resource information database to update the spare distributed system resource information database. IDs updated every time data is updated in the distributed system resource information database
If a number is assigned and a failure occurs in the master distributed system resource information database, the one that has the latest ID number in the spare distributed system resource information database and that first responded to the data update request is newly updated. A method for managing a distributed computer system, characterized in that the service processing of the distributed system resource information database is continued by selecting the master system as the distributed system resource information database.