JPH09259091A - Host load decentralization system for transaction processing - Google Patents

Abstract

PROBLEM TO BE SOLVED: To decentralize a load among hosts with a small overhead by an on-line system equipped with plural host computers which perform transaction processes in parallel. SOLUTION: Transactions (t11-t16) sent from plural terminals (TERM5-TERM 11) are received by a single communication control processor (FEP4) through a line and the FEP4 transmits respective hosts according to a previously set distribution rate. Consequently, the number of transactions need not be adjusted among the hosts and load decentralization matching the performance is actualized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a load balancing method for transaction processing, and more particularly to an online system including a host computer that shares a database and performs transaction processing in parallel so as to balance the load between the host computers. Host load balancing method for transaction processing

[0002]

2. Description of the Related Art A plurality of processing systems (hosts) that share a database and distribute transaction processing
In an online system provided with the above, as a conventional method for distributing the load of transaction processing in which the loads of a plurality of processing systems are evenly distributed, for example, Japanese Patent Laid-Open No. 63-318662 discloses the following. A scheme has been proposed. That is, a shared memory device shared by all hosts is provided, an input transaction processing queue shared by all hosts and an output message queue corresponding to each host are provided in the shared memory device, and when each host accepts a transaction, all input transaction processing is performed. Register in the queue. Each host retrieves the next transaction from the input transaction processing queue and executes the transaction each time one transaction processing is completed. Then, the output message output after the transaction processing is executed is stored in the output message queue corresponding to the host that should output the message. Each host extracts a message from the output message queue corresponding to the host and sends it to the terminal. According to this method, a particular host is not idle and the load is not concentrated, so that load distribution is realized.

[0003]

The conventional load balancing method described above has the following problems.

The first problem is that the throughput of the entire system is not improved even if the transaction becomes a light process.

The second problem is that the throughput of the entire system cannot be improved even if the number of hosts is increased.

The reason for the first problem is that, in the conventional load balancing method, when one transaction is executed, reception processing, queuing processing (queue in, queue out), transaction execution Processing, queuing processing (queue-in, queue-out), message transmission processing becomes a procedure. When the transaction becomes light, the weight of the queuing processing, which is not related to the lightness of the transaction, becomes large. Is.

The reason for the second problem is that the queuing process cannot be executed by a plurality of hosts at the same time because the shared memory must be locked (exclusive control). This is because the lock waiting time increases due to the collision of.

Therefore, the present invention has been made in view of the above problems, and an object thereof is to provide an inter-host load balancing system capable of achieving throughput depending on the weight of transactions and the number of hosts. Especially.

[0009]

In order to achieve the above object, the present invention provides an online system provided with a plurality of host computers that share a database and perform transaction processing in parallel. A host load balancing method for transaction processing is provided, which comprises communication control processing means for receiving and distributing the received transaction to each host computer, and for balancing the load between the host computers.

In the present invention, a plurality of transactions are distributed to each computer via the communication control processing means at a ratio according to the processing performance ratio of each host computer.
Each host computer receives the number of transactions suitable for its processing performance, and this makes it possible to absorb the transaction bias between multiple terminals, and if the transaction load is constant, the transaction with another host computer It is not necessary to adjust the number of transactions between them.

Further, according to the present invention, the communication control processing means roughly pre-distributes the load even if the transaction load varies. Therefore, the bias of the load is reduced, and the shared memory is used only for the biased portion to coordinate with other host computers, whereby the load can be distributed with a small amount of queuing processing.

[0012]

Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a system configuration diagram for explaining the first embodiment of the present invention. This online system includes a plurality of host computers CPU1 to CPU3,
Communication control processing device (Front End Processor) F for controlling communication of all host computers CPU1 to CPU3
EP4 and a plurality of terminals TERM5 to TERM7 connected to the communication control processing device FEP4 via lines. Multiple host computers CPU1 to CPU3
Have the same processing performance, that is, (CPU1: CPU2: C
PU3) = (1: 1: 1).

The communication control processing unit FEP4 is provided with a host selecting means 8 for allocating a transaction to each of the host computers CPU1 to CPU3.
It is preset to distribute transactions to the host computer at a distribution ratio of (CPU2: CPU3) = (1: 1: 1). Multiple terminals TERM5-TERM
Transactions t11 to t1 having the same load from 7
6 is transmitted (from terminal TERM5 to t11 to t1
3, terminals TERM6 to t14, terminals TERM7 to t
15, t16), these arrive at the communication control processing unit FEP4 via the line, but in accordance with the arrival order of these transactions, the transactions t11 to t16.
Are referred to as T21 to T26.

In the host selecting means 8, the distribution ratio is (CP
U1: CPU2: CPU3) = (1: 1: 1), for example, by notifying the transactions in the order of CPU1 → CPU2 → CPU3, the CPU1 receives T2.
1, T24 is the CPU2, T22, T25 is the CPU
3 is notified of T23 and T26, the number of transactions input to each host computer becomes suitable for the host performance, and the load distribution among the host computers can be achieved.

As a second embodiment of the present invention, a case where the processing performance of the host computer varies will be described below with reference to FIG. Host computer CPU1
Regarding the processing performance of the CPU 3, the ratio is (CPU1: CPU
2: CPU3) = (2: 1: 1) (that is, CPU1 has twice the processing performance of the other CPUs 2 and 3). In this case, the communication control processor FE
In P4, (CPU1: CPU2: CPU3) = (2:
It is set in advance so that transactions are distributed to each CPU at a distribution ratio of 1: 1).

From a plurality of terminals TERM5-TERM7,
Transaction t1 with the same load on the CPU
1 to t18 are transmitted, and these are transmitted to the communication control processing unit FEP.
In the order of reception at 4, the contents are set to T21 to T28 again. In the communication control processing device FEP4, for example, the CPU 1 → CPU 2 → CPU 1 → CPU 3 are detected in this order so that the ratio becomes the above ratio.
1, T23, T25, T27, CPU2 is notified of transactions T22 and T26, and CPU3 is notified of transactions T24 and T28. The transaction ratio of CPU1 to CPU3 is 2: 1: 1, and the load is commensurate with the host performance. Dispersion can be achieved.

As a third embodiment of the present invention, a case where the load of each transaction varies will be described below with reference to FIG. This online system uses host computers CPU1 to CP having the same processing performance.
U3, communication control processing unit FEP4 for controlling communication of all the host computers, shared memory device 9 shared by all host computers, communication control processing unit FEP
Terminals TERM5 to T connected to the No. 4 via lines.
It is composed of ERM7.

The communication control processing unit FEP4 is provided with a host selecting means 8 for allocating a transaction arriving from a terminal to each host (CPU1: CPU2: CP).
U3) = (1: 1: 1) is preset to distribute transactions to the host computer.

When transactions t11 to t16 whose loads are not constant are transmitted from a plurality of terminals TERM5 to TERM7, these transactions reach the communication control processing unit FEP4 via the line, and the transactions are re-established in accordance with this arrival order. Let T21 to T26.

In the host selecting means 8, the transaction distribution ratio is (CPU1: CPU2: CPU3) =
(1: 1: 1), for example, CPU1 → CPU2
→ By detecting transactions in the order of CPU3, transactions T21 and T24 are transmitted to CPU1,
Transactions T22 and T25 are sent to the CPU2 as CP
U3 is notified of transactions T23 and T26,
When receiving a transaction, each CPU processes the transaction if the host can process the transaction, and stores the transaction in the shared memory device 9 if the host cannot process the transaction.

With respect to the transaction stored in the shared memory device 9, the CPU which has become ready to process the transaction retrieves the transaction from the shared memory device 9 and processes it.

Since T22 is a transaction with a high load and the transaction T25 is received by the CPU2, the transaction T22 is being processed.
If cannot be processed, the CPU 2 stores the transaction T25 in the shared memory device 9. The load T25 stored in the shared memory device 9 is taken out and processed by the CPU 1 which has finished processing T24 and is in an idle state, for example, whereby load distribution is realized with a small amount of queuing processing.

[0024]

As described above, according to the load balancing method of the present invention, even if there is a bias in the number of transactions among a plurality of terminals, these transactions can be processed by a single communication control processing device. The reception has the advantage of being able to absorb the transaction bias between the terminals.

Further, according to the load balancing system of the present invention, each host computer receives the number of transactions suitable for its processing performance. There is no need to adjust the number of transactions between.

Further, according to the present invention, even if the transaction load varies, the communication control processing device can:
In order to roughly distribute the load, uneven load is reduced, and the shared memory is used only for the uneven part to coordinate with other host computers. Can be realized.

As a result, each host can allocate most of the CPU time to the transaction processing, and if the transaction is light, the throughput can be increased and the lock operation processing can be reduced. It can be expected to improve the performance according to the number.

[Brief description of drawings]

FIG. 1 is a diagram for explaining a first embodiment of the present invention.

FIG. 2 is a diagram for explaining a second embodiment of the present invention.

FIG. 3 is a diagram for explaining a third embodiment of the present invention.

[Explanation of symbols]

 1 to 3 host computer (CPU) 4 communication control processing unit (FEP) 5 to 7 terminal 8 host selecting means 9 shared memory device

Claims (3)

[Claims] 1. An online system provided with a plurality of host computers that share a database and perform transaction processing in parallel. Communication control processing for receiving transactions from a plurality of terminals and distributing the received transactions to each host computer. A host load balancing system for transaction processing, comprising means for balancing the load between the host computers. 2. The communication control processing means controls the plurality of transactions to be distributed to the respective host computers at a ratio according to the processing performance ratio of the respective host computers, and each host computer is suitable for its processing performance. 2. The host load balancing method for transaction processing according to claim 1, wherein a certain number of transactions are received. 3. The communication control processing means distributes a plurality of transactions to the respective host computers in consideration of the load ratio when the loads of the transactions on the host computers vary. A host load balancing system for transaction processing according to claim 1 or 2.