JPH07295843A - Load scattered platform and its load scattering processing method - Google Patents

Abstract

PURPOSE:To provide a load scattered platform and its load scattering processing method which try to attain the high performance of a system, i.e., to shorten the waiting time of each task by making the scattered unit platforms scatter the load in cooperation with each other. CONSTITUTION:A data base DB1, the task modules M and the application programs AP are connected to a unit PF 1 by a DB interface 18, an M interface 17 and an AP interface 13. An accesser 16 manages the access requests given from the programs AP, and each unit PF recognizes its own load state by a data collector 14. At the same time, the PFs notify their load states to each other by a negotiator 15 so that the load states are equalized among the units PF. Furthermore each data base, each task module and each application program are connected to all or plural unit platforms.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to load balancing for use in connection of databases and application programs and connection between various systems in the management and operation of networks used in the fields of communication and office work. Computerized platform and its load balancing processing method.

[0002]

2. Description of the Related Art Conventionally, as shown in FIG. 2, an application program (hereinafter, referred to as AP) that performs a series of work using a database and a database (hereinafter, referred to as DB) are directly connected to each other. It was necessary to change the AP when distribution was difficult or when the DB was changed. That is, AP (1) cannot perform a series of work using the data group 1 and the data group 2, and it is necessary to newly recreate the AP or the system.

Also, in the task module that performs a specific task as shown in FIG. 3, the AP and the task module are directly connected. Therefore, when the task module is changed, it is necessary to change the AP. Further, the change in the case of an AP using a plurality of task modules and a plurality of DBs was very complicated.

As a means for solving this, as shown in FIG. 4, AP and DB, AP and task module are not directly connected but separated, and a platform (hereinafter abbreviated as PF) which is a connecting portion thereof is interposed. By configuring it, you can distribute the data and change the DB and task module.
It is possible to reduce the influence on P. Here, the platform is a plurality of DBs, task modules and APs.
Is software for connecting to, for example, the Institute of Electronics, Information and Communication Engineers VOL.E76-B, No.4, April 1993, pp.391-401 N.
Kashima et al. "Optical cable network operation in
"Subscriber loops". Each AP, DB,
The task module is connected to the platform.

[0005]

However, as shown in FIG. 4, since a large number of DBs, task modules, and APs are connected via a single platform, it becomes necessary to perform distributed processing. Sometimes. For example, there is a need for decentralization because of the processing capability of the device, the installation location, and the reliability. 5 to 7 show examples in which the basic configuration of FIG. 4 is dispersed. Here, the unit PF is
All the functions of the platform (PF) in FIG. 4 are distributed as a unit platform (unit PF).

Even with the PF dispersed in this way, it is necessary to exhibit the best possible characteristics as the entire system. For example, if nothing is controlled, the load concentrates only on one unit PF, and when an AP connected to that unit PF is in a waiting state for a very long time, another unit P
F may be operated under a light load.

On the other hand, the idea of distributed cooperation has been proposed as a means for improving the performance of a distributed system. For example, 1984 REFilman and
It is disclosed in the book of Coordinated Computing by DPFriedman and in the article "The agent-oriented" capable of autonomous operation emerged in Nikkei Intelligent Systems in 1992. That is, the execution unit that operates autonomously. The idea is that the agents proceed with processing in cooperation with each other.

In view of the above problems, it is an object of the present invention to improve the system performance, that is, to reduce the waiting time of a task, by coordinating the distributed unit platforms to distribute the load. A distributed platform and a load distribution processing method thereof are provided.

[0009]

In order to achieve the above object, the present invention provides, in claim 1, a database group, a task module group, and an application program group for executing a predetermined task using the two groups. , A distributed processing system composed of a unit platform group which is a coupling part between the three groups, each unit platform having interface means for interfacing with a database, a task module and an application program, and an application program. Access control means for managing access requests from each unit platform, processing status collection means for each unit platform to recognize its own load status, and each unit platform to notify each other of the above load status, and load between unit platforms Condition The is composed of a distributed cooperative processing means for controlling to equalize, each database, each task module, each application program proposes a load distribution platform that formed by connecting a plurality of units platforms.

Further, in claim 2, a database group,
Platform load balancing processing in a distributed processing system including a task module group, an application program group that executes a predetermined task using the two groups, and a unit platform group that is a coupling unit between the three groups A method, in which each unit platform periodically activates and executes a test application program that is common to the unit platforms and that accesses a test task module, and obtains its processing time T ₁ to recognize its own processing status. The first process,
Unit platform # 1 above process time T ₁ is exceeds the set time T ₀ is, a second step of notifying the processing time T ₁ to the other units platform that is connected to his, the unit platform # 1 other than each unit platform compares the magnitude of their test application program processing time T ₂ and the processing time T _1, only when the T ₂ <T _1, third to respond to the unit platform # 1 In the process, the unit platform # 1 connects and switches some of the application programs connected to itself to the unit platform that has returned the minimum value of the processing time received from other unit platforms, By the fourth process of distributing a part of the load and performing cooperative processing, The load state of the bets platform regularly review, we propose a load balancing method of processing platform to level the load state for the entire unit platform.

[0011]

According to the first aspect of the present invention, in each unit platform, the interface means provides an interface between the database, the task module, and the application program, and the unit platform is connected. To be done. Further, the access management unit manages the access request from the application program, the processing status collection unit recognizes its own load state, and the distributed cooperative processing units notify each unit platform of the load state. , It is controlled so as to level the load condition between unit platforms. Furthermore, each database, each task module, and each application program are connected to a plurality of unit platforms. As a result, as shown in FIGS. 8 and 9, the processing status collecting means of each unit platform recognizes the load status of the unit platform, and the distributed cooperative processing means of each unit platform loads the other unit platforms. A request for distribution, a response processing, and a processing request for load distribution are performed.Each unit platform issues a request for load distribution to another unit platform when an overload occurs, and All unit platforms will receive this request and will respond back as needed, ultimately deciding which unit platform to ask for overload sharing. Therefore, an overloaded unit platform can request a part of the load from another minimum loaded unit platform.

According to a second aspect of the present invention, each unit platform periodically activates and executes a test application program that is common to the unit platforms and that accesses a test task module.
The processing time T ₁ is obtained, the processing status of each unit platform is recognized, and the other unit connected to itself by the unit platform # 1 whose processing time T ₁ exceeds the set time T _0. platform the processing time T ₁ is notified to the by the unit platform # 1 except for the units platform, are compared the magnitude relation between the processing time T ₂ of the own test application program and the processing time T _1, T ₂ Only when <T _1, the unit platform # 1 is returned. Based on the result of this reply, the unit platform # 1 returns a minimum value among the processing times received by some of the application programs connected to the unit platform # 1 from other unit platforms. The connection is switched to the platform, a part of the load of the unit platform # 1 is distributed and coordinated, the load status of each unit platform is periodically reviewed, and the load status of the entire unit platform is leveled.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a configuration diagram showing a unit PF according to an embodiment of the present invention. In the figure, 1 is a unit PF, in which the test application program AP (T
EST) 11, test task module M (TEST)
12, application program AP (1), AP
AP interface 1 that interfaces with (2)
3, a data collector 14 having a processing status collecting function, a negotiator 15 having a distributed cooperative processing function, an accessor 16, operation task modules M (1), M
An M interface 17 that interfaces with (2),
And a DB that interfaces with a database (DB)
The interface 18 is built in.

The accessor 16 is the AP interface 1
Multiple application programs AP through 3
(1), having an access management function for managing access requests from the AP (2). For example, a priority order is determined in advance in an application program, and priority order access control (priority access control) is performed based on this priority order. To do.

Next, the operation of this embodiment having the above-mentioned structure will be described with reference to FIGS. Unit PF
1 is a test application program AP (TES) by the data collector 14 and the negotiator 15.
T) 11 and test task module M (TEST) 12
Is started and executed periodically, and the test application AP
The processing time T1 of the (TEST) 11 is measured (S1).
The test application AP (TEST) 11 is shown in FIG.
Test task module M (TEST) 1 as shown in
Access to 2 shall be requested.

As a result, the load status of the unit PF1 is determined by the start start time of the test application AP (TEST) 11 and the test task module M (TEST) 1.
The difference T ₁ from the response completion time from 2 can be measured by executing the test application AP (TEST) 11.

That is, the measurement time T ₁ reflects the load state of the unit PF1. For example, when the priority order of access control (priority access control) is set to five levels, if the priority order of the test application AP (TEST) 11 is set to X = 3, the processing of the application AP having a medium priority order is performed. It will measure time.
When the number of accesses to the unit PF1 increases, the application AP enters the waiting state under the control based on the priority order, and the processing time increases.

The processing time T ₁ is reported to the data collector 14 by the test application AP (TEST), and the data collector 14 determines whether the processing time T ₁ is within a preset reference time T ₀ or not. (S2).

As a result of this determination, when the processing time T ₁ exceeds the preset time T ₀ , the negotiator 15
The processing time T ₁ is notified to all the other units PF1 connected thereto via (S3). This notification is also received by the negotiator 15 of each unit PF.

After that, each notified unit PF1 compares the processing time T ₂ of its own test application AP (TEST) with the notified processing time T ₁ (S4).
S5), as a result of this comparison, the processing time T ₂ is the processing time T ₁
Only when it is shorter, the unit PF1 that has notified the processing time T ₁ (the unit PF that has declared overload)
The processing time T ₂ is returned to 1) (S6, S7).

As a result, the unit PF of the processing time T ₁ is
1 (unit PF1 which has declared overload) requests a part of load (processing) to the unit PF1 which has returned the minimum processing time (S8), and the corresponding application program AP to the unit PF1 which has returned the minimum processing time. Reconnect (S9).

As described above, the unit P in the overloaded state
It becomes possible for F1 to request a part of the processing to an appropriate other unit PF1. As a result, the load can be averaged over the entire system, and high performance can be achieved. Also,
Since the distributed cooperation is performed only when the load exceeds a certain threshold, the overhead for communication is reduced and high-speed processing becomes possible.

Further, each unit PF is of the same rating, and only when there is a unit PF whose load exceeds a certain threshold and which is desired to make a request for reducing the load, this unit P.
Hierarchy occurs due to the cooperative processing of the entire system with F as the main body. When this collaborative process ends, the equivalence relation is established again. Due to this homogeneity, even if one unit PF fails, it has little influence on the entire system and has excellent system expandability.

Next, an operation example in the case where the number of unit PFs n = 3 will be described with reference to FIG. Unit PF # 1
# 3, application program AP (1), AP
(2) The operation task modules M1 to M3 and the database DB1 are L represented by Ethernet.
DCN (Data Commu) such as AN (Local Area Network)
nication Network: a data transfer network) ((a) in FIG. 12). Logically
As shown in (b), DB1, Mi (i = 1, 2, 3) are unit PF # 1, unit PF # 2, unit PF #.
Connected to 3.

Initially, it is assumed that the application program AP is connected only to the unit PF # 1 as shown by the solid line. Although not shown in FIG. 12B, it is assumed that the other application program AP group is also connected to the unit PF # 2 and the unit PF # 3. All of these application programs AP are the other units P
It is assumed to have the function of connecting to F.

When the unit PF # 1 measures an overload in the process shown in the flowchart of FIG. 11 (T ₁ > T
₀ ), the unit PF # 1 issues a processing request to the unit PF # 2 or the unit PF # 3. Figure 11
The unit PF of the processing request destination is determined according to the algorithm (1) (for example, the unit PF # 2).

The unit PF # 1 selects a part of the application program APs from the group of application programs AP (for example, half of the APs in the waiting state by counting after the queue) and connects to the application program AP. Notify you of the previous change. The application program AP that has received the notification changes the request destination from the unit PF # 1 to the unit PF # 2. This makes it possible to reduce the processing time as a whole.

It should be noted, relates the processing time T _2, stores the test application program processing time measured just prior to comparing the magnitude relation between the processing time T ₁ and the processing time T _2, it is used this value good.

[0029]

As described above, according to the load distribution platform according to claim 1 of the present invention, since the processing status collection means of the platform itself and the distributed cooperation processing means are provided in the unit platform, the load is reduced. When a certain threshold is exceeded, other unit platforms are notified, and the load that exceeds the threshold is distributed to other unit platforms for processing, so a distributed non-cooperative algorithm is implemented in the distributed cooperative processing means. This makes it possible to avoid processing delays that are often associated with distributed processing, reduce overhead for communication, and speed up processing. Further, in the configuration of the present invention, since the unit platforms have the same rating, even if one unit platform fails, it has little influence on the entire system and has excellent system expandability.

According to the platform load distribution processing method of the second aspect, it becomes possible for an overloaded unit platform to request a part of the processing to an appropriate other unit platform, and load the entire system. Can be averaged and high performance can be achieved.
Further, since distributed cooperation is performed only when the load exceeds a certain threshold, the overhead for communication is reduced and high-speed processing becomes possible. Furthermore, each unit platform is of the same rating, and only when there is a unit platform for which the load exceeds a certain threshold and is requested to reduce the load, this unit platform takes the lead in coordinating the entire system. Therefore, even if one unit platform fails, there is little effect on the entire system and the system is highly expandable.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of the present invention.

FIG. 2 is a diagram illustrating a conventional technique.

FIG. 3 is a diagram illustrating a conventional technique.

FIG. 4 is a diagram illustrating a conventional technique.

FIG. 5 is a diagram illustrating a conventional technique.

FIG. 6 is a diagram illustrating a conventional technique.

FIG. 7 is a diagram illustrating a conventional technique.

FIG. 8 is a configuration diagram for explaining the outline of the present invention.

FIG. 9 is a configuration diagram for explaining the outline of the present invention.

FIG. 10 is a diagram for explaining the operation of the embodiment of the present invention.

FIG. 11 is a flowchart illustrating the operation of the embodiment of the present invention.

FIG. 12 is a diagram illustrating an embodiment of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Unit platform, 11 ... Test application program AP (TEST), 12 ... Test task module M (TEST), 13 ... AP interfacer, 14 ... Data collector (processing status collection means), 1
5 ... Negotiator (distributed cooperative processing means), 16 ... Accessor, 17 ... M interface, 18 ... DB interface.

Claims (2)

[Claims] 1. A distributed system including a database group, a task module group, an application program group that executes a predetermined task using the two groups, and a unit platform group that is a coupling unit between the three groups. In a type processing system, each unit platform has an interface means for interfacing with a database, a task module and an application program, an access management means for managing an access request from the application program, and each unit platform has its own load. The processing status collecting means for recognizing the state, and each unit platform is composed of a distributed cooperative processing means for notifying each other of the load state and controlling so as to level the load state among the unit platforms, Database, each task module, load distribution platform that each application program is characterized by comprising connected to a plurality of units platforms. 2. A distributed system comprising a database group, a task module group, an application program group for executing a predetermined task by using the two groups, and a unit platform group which is a coupling unit between the three groups. A method of load balancing processing of platforms in a type processing system, wherein each unit platform periodically activates and executes a test application program that is common to the unit platforms and that accesses a test task module, and obtains its processing time T ₁ . Then, the first process of recognizing one's own processing status, and the unit platform # 1 whose processing time T ₁ exceeds the set time T ₀ , transfers the processing time T ₁ to another unit platform connected to itself. Second step of notification and the unit platform # Each unit platforms except that compares the magnitude relation between the processing time T ₂ of the own test application program and T ₁ the processing time, only when the T ₂ <T _1, and replies to the unit platform # 1 In the third step, the unit platform # 1 connects and switches some of the application programs connected to itself to the unit platform that has returned the minimum value of the processing time received from another unit platform, The load distribution of the platform characterized by periodically reviewing the load status of each unit platform and leveling the load status of the entire unit platform by the fourth process of distributing and coordinating a part of own load. Processing method.