JP2913591B2 - How to get an online journal - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a transaction management system of an online transaction system, and more particularly to an online journal acquisition method suitable for high-traffic transaction processing.

[Prior Art] As a journal (system history) information acquisition method in a conventional online transaction system, a system is adopted in which a journal acquisition request is issued to a non-volatile storage medium every time a journal acquisition request is issued. Was.
For example, a magnetic disk device (DI
SK) or magnetic tape unit (MT).

In such a method, Japanese Patent Application Laid-Open No.
As disclosed in No. 6643, "Online Journal Acquisition Method", a method is adopted in which multiple requests are acquired collectively only when a plurality of requests are queued at the time of journal acquisition.

In this method, the efficiency is increased only when the requests happen to be continuous, but there is a disadvantage that the efficiency is not increased at all if the requests are not continuous. In particular, regarding journal acquisition, in that DISK write requests are issued one by one,
It is the same as the above-mentioned prior art.

FIG. 2 shows an outline of the conventional transaction processing.
When the input transaction message is received from the terminal (101), the journal information is acquired in the DISK 107 (10).
2) After updating the database (DB) (103), a response message is transmitted to the inquiry source terminal (104), and the process ends (105).

[Problem to be Solved by the Invention] In the above-mentioned conventional technology, the following two items are premised.

(1) The memory is a volatile storage medium, and it is not known when and when its contents are lost / destructed.

(2) The journal information is the history information of the system and is used only for recovery from a failure, and loss or destruction of any one is not allowed.

However, with the recent development of FTC (Fault Tolerant Computing) technology, memory, power supply,
The CPU etc. are all duplicated, even if it is a volatile storage medium,
The reliability has been significantly improved, and the situation where the memory contents are lost / destroyed due to a so-called single point of hardware failure has hardly been considered. Therefore, it is necessary to improve the journal acquisition efficiency on the premise that the memory is also a sufficiently reliable medium.

On the other hand, due to the recent wave of liberalization and internationalization of the financial and securities industries, frequent system changes / expansion are performed in the online transaction system (OLTP system). At the same time, with the shortage of programmers, there is an urgent need to improve program development efficiency. From this point of view, there has been a growing interest in using journals that have been used only for failure recovery as test data for application programs. In a word, the type of journal /
The purpose is different. (For example, an input / output journal for logging data to / from the transaction monitor, a file journal obtained when updating a file / database, a system journal for logging system history, etc.). There is also a demand for improving the throughput of the entire system by performing fine-grained processing for each of the diversified journal types. Under the above background, (1) an object of the present invention is to reduce system overhead due to journal acquisition, improve response time for individual transactions, and realize high throughput in the entire transaction processing. .

(2) Another object of the present invention is to use a journal acquisition method in accordance with a request for reliability for each type of journal, and to acquire a journal in fine units for those having a high request for reliability. Is to collectively acquire those for which the request is low, thereby achieving high-reliability and high-throughput transaction processing in the entire system.

[Means for Solving the Problems] To achieve the above object, an online journal acquisition method according to the present invention is an online journal acquisition method for acquiring journal information in an online transaction system, and a predetermined condition is satisfied. Journal information is sequentially stored on a volatile storage medium until the condition is satisfied, and when the above condition is satisfied, the journal information stored up to that time is collectively stored in a nonvolatile storage medium. Things.

In this method, for example, the condition is that a specified number of journal information is accumulated on the volatile storage medium.

Alternatively, the condition is that a specified time has elapsed.

Alternatively, the condition may be that specific journal information is generated.

At least two conditions are adopted among a specified number of pieces of journal information stored on the volatile storage medium, a lapse of a specified time, and the occurrence of specific journal information. When any one of the conditions is satisfied, the journal information may be collectively stored in a nonvolatile storage medium.

In addition, conditions are set independently for each journal type,
It is also possible to accumulate on a volatile storage medium until the condition is satisfied for each journal type, and collectively store the journal information for the journal type for which the condition is satisfied in a nonvolatile storage medium.

[Operation] The journal acquisition method according to the present invention accumulates (stacks) journals in a memory and collectively writes them to a non-volatile storage medium (DISK / MT) at a certain time as long as the demand for reliability and the memory capacity allow. Is what you do. As a result, it is not necessary to write a journal to the DISK / MT for each transaction process, so that the response time can be improved. Also, for the entire system, by processing a plurality of journals collectively, it is possible to collectively refer to / update tables in memory, reduce the number of dynamic steps per journal in the processing, and thereby reduce the CPU processing time. Can be.

The most effective way to improve the throughput of the entire system is batch access (acquisition) to DISK / MT in journal acquisition. In DISK access, the seek time is one digit or more longer than the data transfer time.
It is clear that the input / output (I / O) time can be greatly reduced if the seek operation is performed only once for multiple journals, compared to the conventional method in which a seek is performed for each case. The number of transactions that can be processed in time increases, and the throughput improves.

[Embodiment] Fig. 3 shows a system configuration example of a typical online system to which the present invention is applied.

The system processes transactions entered from thousands of terminals scattered throughout the country. At the highest level, a hot standby system is configured with two host computers 301 directly connected to a channel (302) mainly as a database server. Below that, a front-end processor FEP303 is distributed and arranged at each base mainly for the purpose of terminal control. Since the transmission / reception of large amounts of data is mostly between the FEP and the host computer, high-speed digital lines or ISDN 304
Is used. Under each FEP, dozens to hundreds of terminals 201
Are connected by a dedicated line 305. Main transaction processing is performed between the FEP and the terminal, and between the FEP and the host computer, batch processing is the main stream. The online journal acquisition processing of the present invention exists in the FEP which requires high-throughput and high-response transaction processing.

FIG. 11 shows an example of journal acquisition in a bank online system which is a typical OLTP system.

Suppose a person withdraws 1,000 yen from his account. First, the receipt of the pay-out request message from the ATM and the journal are acquired. Next, the requested customer record is searched, and it is checked whether or not the balance of this customer record is 1000 yen or more, and the result is obtained. If it is 1000 yen or more, the balance is reduced by 1000 yen, and the fact that the payment request message has been sent to the ATM is recorded in the journal. Lastly, upon receiving a payout completion notification from the ATM, the transaction is terminated. As described above, by acquiring all the steps in one transaction in the journal, it is possible to reliably prevent the destruction of the customer record even when the system goes down at any time.

FIG. 4 shows a schematic flow of the transaction processing. First, a reception request is issued to receive an input message from the terminal (101). After receiving the input transaction message, the journal information of the transaction is acquired (102), and then the database is updated (103), a response message is returned to the inquiry source terminal (104), and the transaction is terminated (105). Go to
This transaction processing ends.

FIG. 1 shows a detailed flow of the journal information acquisition processing (102) according to the present invention in the above processing.

In the flow of FIG. 1, when a journal acquisition instruction is received from the transaction processing (401), the journal is stacked on the memory (402). Subsequently, it is checked whether or not the number of journals stacked on the memory has reached a specified number (403). If not reached, the present process is terminated. If the specified number has been reached, the journals in memory will be
Is written (404), and this process ends.

In this journal acquisition processing, when the specified number of journals are stacked on the memory 106, the journals are acquired collectively to the DISK 107.

Next, the journal acquisition processing is performed when the timing of the batch acquisition is set at a time when one of two conditions, that is, when a prescribed number of journals is stacked on the memory or when a prescribed time of the system elapses, is satisfied. 5 is shown in FIG.

In the flow of FIG. 5, when a journal acquisition instruction is received (401), the journal information is stacked on the memory (402). Next, it is checked whether the number of journals stacked on this memory has reached the specified number (40).
3). If not reached, the present process is terminated. If the specified number has been reached, the timer routine is reset (501), the journals in the memory are collectively written to the journal disk (404), and this processing ends. On the other hand, when the elapse of the system specified time is detected by the timer (502), the journal batch acquisition process is started, the timer is reset (501), and the journals in the memory are written to the DISK (404). .

FIG. 6 shows the detailed flow of the journal acquisition process when the timing of the batch acquisition is either when a specific journal is generated or when a specified number of journals are stacked on the memory.

In the flow of FIG. 6, when a journal acquisition instruction is received from the transaction processing (401), the journal is stacked on the memory (402). Next, it is checked whether the journal is a journal with a specific id (601). If it is not a specific journal, it is checked whether the number of journals stacked on the memory has reached a specified number (403). If not reached, the present process is terminated (403). In the above two checks (6
01, 402), and if the condition is satisfied (YES), the journals in the memory are collectively written to the journal disk (404).

FIG. 7 shows a journal format in the case of this method. In this method, JID: journal id is set at the head of the journal information to determine whether each journal is a specific one.

Next, an embodiment in which the above three conditions (the number of stacks, the number of stacks and the specified time, and the number of specific journals and stacks) are individually adopted for each journal type will be described.

FIG. 8 shows a table structure for controlling the batch acquisition timing of various journals. In this table, the number of stacks, the stack time, a specific journal id, and a pointer (address) to each stack area are set for each journal type. In the journal acquisition process in this example, the journal acquisition conditions are different for each type of journal. That is, as shown in the figure, in the journal type 1, the number of stacks and the stack time are used together, and in the journal type 2, the number of stacks and the specific journal id are used.
Is also used. In journal type 3, the number of stacks and the stack time are used together. In addition, three combinations of a stack time and a specific journal id, and three combinations of the number of stacks, a stack time and a specific journal id can be considered. In addition, the number of stacks, stack time,
It may be a single condition for each specific journal id. The system refers to the journal management table 801 and separately performs the processing shown in FIGS. 1, 5, and 6 for each journal type. FIG. 9 shows a journal format suitable for this method. In this method, in addition to journal information 701 and a journal id (JID) 702 for identifying a specific journal, a journal type 901
Area is provided. FIG. 10 shows the journal management table 801 of FIG.
1 shows a layout diagram. A separate journal stack area 1001 is provided for each journal type.

As described above, according to the present embodiment, in the background of diversifying journal acquisition purposes and high reliability of memory, among journals of the same type, those that emphasize efficiency, those that value reliability, Stacks other than those that value reliability can be stacked, and an efficient and highly reliable journal system can be realized in the entire system.

Finally, in the actual design, when a plurality of journals are acquired collectively, at least how much memory stack area is required so as not to affect the transaction processing, and specific effects of the present invention The calculation method is shown below. Here, having no effect on the transaction processing means that when the transaction processing issues a journal acquisition instruction, a situation in which the journal stack area is full and the user has to wait for the journal stack area to become empty does not occur. is there.

For simplicity, it is assumed that there is one journal type and the batch acquisition timing is the specified number = 10.

<Conditions> 1. There is one DISK for journals in the system, and its performance is as follows.

Seek time: 23ms Search time: 9ms Transfer speed: 1MB / s 2. Journal acquisition capacity: 5KB / item 3. Peak load: 50 Transaction / s 4. CPU time: 90ms (at the time of batch acquisition of 10 items) 20ms (each item) <Total Calculation> The number of transactions arriving during the above time is 50 (Transaction / sec) x 0.172 = 8.6 (Transaction), so the minimum required journal stack area is 5KB x 10 + 5KB x 8.6 = 93KB. Actually, if the safety factor is about 1.5 and a journal stack area of about 150 KB is prepared, it is considered that the stack area does not become full in this example. It is considered that the required stack area may be too large depending on the system. In this case, the number of batch acquisitions may be reduced. However, since the number of times of disk access and the CPU processing time increase, the throughput decreases.

Also, the stacking time is 10 (items) / 50 (items / second) = 200 [ms] for stacking 10 items, and if the system specified time is set to 200 ms, each journal will be at most 200 ms. During this time, it will be stacked on the memory.

At this time, the CPU load due to journal acquisition is 10 ⁻³ × 90 ms × 5 (times / second) = 45 [%] DISK load is 10 ⁻³ × ((5 KB × 10 records) / 1 MB / s + 23 + 9) × 5 ( Times / sec) = 41 [%].

On the other hand, a quantitative effect is shown below as compared with the case where a journal is acquired for each case.

First, as for the response time, when writing to the DISK one by one, the DISK writing time is simply 5 KB / 1 (MB / s) +23 ms + 9 ms = 37 ms, and the response time becomes longer by this time.

Also, the CPU load is only journal processing, and 20 (ms) x 50 (items / s) x ^10-3 = 100% DISK load also increases to 37 (ms) x 50 x ^10-3 = 185% However, assuming that journals are acquired one by one, in the above calculation conditions, from the viewpoint of CPU load and DISK load,
It is clear that the system does not work.

It will be understood quantitatively that the collective acquisition of journals can significantly reduce not only the response but also the load on the CPU and DISK as compared with the above example.

[Effects of the Invention] The present invention is configured as described above, and has the following effects.

CPU load, DIS
The K load can be reduced, and the overall system can be improved in throughput. Further, by providing the system specified time, the longest memory stack time of each journal can be limited, and quantitative reliability can be obtained. In addition, by collectively acquiring when a specific journal has occurred, the reliability of a specific journal can be significantly increased.

By setting the above batch acquisition conditions for each journal and controlling it, a certain level of reliability is guaranteed for journals that require some reliability, and high-response, high-throughput journals are used for the entire system. The system can be realized.

[Brief description of the drawings]

1 is a flowchart showing a detailed flow of a journal acquisition process according to the present invention, FIG. 2 is a flowchart showing an outline of a conventional transaction process, FIG. 3 is a block diagram showing an example of a system configuration to which the present invention is applied, and FIG. FIG. 5 is a flowchart showing an outline of a transaction process in the present invention. FIG. 5 is a flowchart showing a detailed flow of another journal acquiring process according to the present invention. FIG. 6 shows a detailed flow of another journal acquiring process according to the present invention. FIG. 7 is a format diagram of a journal in the present invention, FIG. 8 is an explanatory diagram of a journal management table, FIG. 9 is another format diagram of the journal in the present invention, and FIG. Fig. 11 is an explanatory diagram showing the relationship between the stack area and the stack area. FIG. 106 …… Journal stack area (memory), 107 ……
DISK, 201: Terminal group, 301: Host computer, 303: F
EP, 801 ... Journal management table.

Continued on the front page (72) Keijiro Hayashi 5-2-1 Omikacho, Hitachi City, Ibaraki Prefecture Inside the Omika Plant, Hitachi, Ltd. (72) Yoshiaki Adachi 5-2-1 Omikamachi, Hitachi City, Ibaraki Stock (72) Inventor Isao Terakado 5-2-1 Omikacho, Hitachi City, Ibaraki Prefecture Within Hitachi Process Computer Engineering Co., Ltd. (72) Inventor Kenji Matsuzaki 5-2-2 Omikacho, Hitachi City, Ibaraki Prefecture No. 1 Hitachi Process Computer Engineering Co., Ltd. (56) References JP-A-63-217441 (JP, A) JP-A-2-83641 (JP, A) JP-A-2-82325 (JP, A) JP-A 1-213738 (JP, A)

Claims (6)

(57) [Claims] An online journal acquisition method for acquiring journal information in an online transaction system, wherein a condition determined according to the journal type is set independently for each journal type, and the corresponding condition is satisfied. Until the condition is satisfied, the journal information of each generated journal is stored only on the volatile storage medium, and the journal information of the journal type satisfying the corresponding condition is stored when the condition is satisfied. Reading an online journal and collectively storing it in a nonvolatile storage medium. 2. The condition set for each journal type is that a specified number of journal information of the type specified for the journal type is stored on the volatile storage medium. 2. The online journal acquisition method according to claim 1, wherein: 3. The condition set for each journal type is that a specified time set for the journal type has elapsed since the last time the journal information of the type was stored in the nonvolatile storage medium. Claim 1 characterized by the following:
Online journal acquisition method described. 4. The method according to claim 1, wherein the condition set for each journal type is that specific journal information defined for the journal type has occurred.
Online journal acquisition method described. 5. At least one of the conditions set for each journal type is that a specified number of journal information defined for the journal type is stored on the volatile storage medium. 2. The online storage device according to claim 1, wherein the other is that a specified time set for the journal type has elapsed since the last time the journal information of the type was stored in the nonvolatile storage medium. Journal acquisition method. 6. At least one of the conditions set for each journal type is that a specified number of journal information defined for the journal type has been stored on the volatile storage medium, 3 that the specified time defined for the type has elapsed since the last time the journal information of the type was stored in the nonvolatile storage medium, and that the specific journal information specified for the journal type has occurred. 2. The online journal acquisition method according to claim 1, wherein a condition is that at least one of at least two of the two events has occurred.