JP2003044310A - Clustering system and method for restoring data when fault occurs in clustering system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a clustering system with improved processing efficiency by eliminating, transfer of management data for fail over among the respective server machines. SOLUTION: The respective server machines 11 to 14 are provided with management data storage parts 60 to 61 to store the management data of processings to be executed in the server machines, batteries 70 to 73 to supply power to interface parts 50 to 53 to electrically and mechanically connect at least the management data storage parts 60 to 61 and the server machines with a communication line 110 when a fault occurs in the server machines, the interface parts 50 to 53 to receive power supply by a battery in the first server machine transfer the management data read from the management data storage parts 60 to 61 to a second server machine and continues service to a client based on the management data received by the second server machine when the fault occurs in a first server machine among the respective server machines.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a clustering system in which a plurality of server machines are connected to a changeover switch and operated in a coordinated manner, so that even if one server machine fails, service to clients can be continued. Regarding the system.

[0002]

2. Description of the Related Art Conventionally, a plurality of server machines are connected to a changeover switch by a communication line and operate cooperatively so that even if one server machine fails, clustering can continue service to clients. The system is known. In this clustering system, a plurality of server machines appear to the client as if they were one server. The clustering function is supported by an OS such as UNIX (registered trademark) or Windows NT (registered trademark).

Further, the clustering system is a kind of fault-tolerant system because it can provide a service without stopping the function of the system even when the number of clients is increased and the number of server machines is increased or the maintenance is performed. Conceivable.

A fault-tolerant system generally has redundancy such as duplication of various components, and normally uses only one component except for backup of data. Let the other component rest,
When a failure occurs in one of the components, the other component is immediately used so that the data can be automatically recovered and the processing can be continued when the failure occurs.

Generally, a clustering system has two
More than one server machine is connected to the changeover switch by a communication line, and while each server machine performs the service requested by multiple clients, for failover of journal data etc. required for each server machine (data recovery Management data) is transferred from each server machine to a spare server machine allocated to each server machine during the service.

FIG. 5 is a block diagram showing the configuration of a conventional clustering system.

The clustering system 100 of FIG.
Is a plurality of server machines (first server 101, second server 102, third server 103, ..., Nth server 10).
4), service requests from clients, various instructions and requests between server machines, or management data for switching the destination of failover management data such as journal data indicating the history of processing results in each server machine. Changeover switch 105, a main data changeover switch 106 for switching a transmission / reception destination of generally large-capacity main data such as voice data and image data processed by each server machine, and main data is multiplexed and stored. Main data storage unit 107, and a power supply unit 108 that supplies electric power obtained from a commercial power supply to each of the above units.
And each server machine and management data changeover switch 105
And a communication line 110, which is, for example, an Ethernet bus.
And each server machine and main data changeover switch 10
6 and a high-speed communication line 120 that is a high-capacity and high-speed connection, for example, an optical fiber bus, and a power supply line 130 that supplies power from the power supply unit 108 to each unit.

The types of data handled between each server machine and the management data changeover switch 105 include, for example, not only management data (journal information etc.) for failover at the time of system down, but also load balancing. Data is also handled. Therefore, the amount of data is not so large as the main data handling image data and audio data, and, for example, as the system becomes larger, a higher communication speed is required. An Ethernet connection is generally used between each server machine and the management data changeover switch 105.

On the other hand, the type of data handled between each server machine and the main data changeover switch 106 is, for example, voice data or image data. Therefore,
Since the data amount is overwhelmingly large as compared with the management data and the like, the connection is made by an optical fiber bus or the like which is faster than the connection by the ordinary Ethernet or the like used in the communication line 110.

Further, in each server machine, the management data in each server is converted or adjusted into a data format suitable for communication by Ethernet, and conversely, the communication data of Ethernet is suitable as management data in each server. Management data interface units 150 to 153 for converting or adjusting to different data formats, and management data storage units 160 to 1 for storing management data such as journal data indicating processing contents or processing results requested for each server machine.
Each has a management data interface unit 150 to 153 and a management data storage unit 1.
Between 60 and 163, the internal bus 14 of each server machine
They are connected by 0-143.

In addition, among the power supply lines 130, a power supply line 130a for supplying electric power to the first server 101 is particularly used.
The power supply line for supplying power to the second server 102 is 130b, the power supply line for supplying power to the third server 103 is 130c, and the power supply line for supplying power to the nth server 104 is 130d.

Next, the operation of backing up (duplicating and saving) the management data during the normal operation of the clustering system 100 shown in FIG. 5 will be described.

FIG. 6 is a flow chart of the operation of redundantly storing management data in the conventional clustering system 100.

Here, for example, it is assumed that the first server 101 performs processing in response to a request from a client.

First, the first server 101 confirms whether or not a processing request from a client is generated (S1). If there is no processing request (S1: NO),
The request from the client is constantly monitored by repeating the confirmation in step S1. When there is a processing request (S1: YES), the first server 101 confirms whether or not the requested content is management data (for example, journal data) storage (S2). If the management data is not stored (S2: NO), normal processing is performed according to the requested content (S5). If the management data is to be stored (S2: YES), the first server 101 performs an interrupt process for storing the management data (S2).
3).

The interrupt processing for storing the management data in step S3 is based on the processing in the CPU (not shown) of the first server 101.
When the process (job) currently being executed by the server 101 is finished, the start of the next job is temporarily suspended, and the address position of the main data stored as the job result in the main data storage unit 107 The management data is generated and stored in the management data storage unit 160 provided in a part of the main memory or the like.

Further, in the conventional interrupt processing, since the management data is duplicated and stored (saved), the management data such as the journal data saved in the management data storage unit 160 is transferred from the management interface unit 150 to the first Two
It is sent (transfer T1) to the management interface unit 151 in the server 102.

Therefore, in the transfer process to another server shown in FIG.
A management data transfer request is sent to the server 102 (S
6). Next, it is confirmed whether or not the second server 102 of the transfer destination is busy (S7). If the second server 102 is busy (S7: YES), the management data cannot be transferred. The server 101 waits until the second server 102 is no longer busy. When the second server 102 is no longer busy (S7: NO), the management data stored in the management data storage unit 160 of the first server 101 is transferred to the second server 102 (S8). After that, after the result that the management data is stored in the management data storage unit 161 in the second server is confirmed, the next process (job) is performed.
Will be restarted.

Next, it is confirmed whether or not the interrupt processing in step S3 is completed (S4), and if it is completed (S4).
4: YES), the process returns to step S1, and if not completed (S4: NO), the process returns to step S3 to continue the interrupt process.

During normal operation, for example, the first server 1
A management data storage request is generated every time one job in 01 is completed, the updated management data contents are stored in the management data storage unit 160, and the management data storage is duplicated. 2 is also transferred to the server 102. That is, during the processing in the first server 101, the above interrupt processing frequently occurs, and the transfer T1 of the management data from the first server 101 to the second server 102 is executed.

While the normal processing is being performed as described above, for example, the power supply line 130 to the first server 101
The case where a failure occurs in a will be described below. In practice, a failure in a server machine may occur at various locations, but it is considered that the failure is the power supply line. The line was designated as the fault location.

FIG. 7 is a flowchart of the operation of the conventional clustering system 100 when a failure is detected.

First, for example, whether or not the second server 102, which has received the management data from the failed first server 101 in the clustering system 100, has failed in the first server 101 that is executing the process. It is confirmed (S11).

The confirmation of the occurrence of the failure in step S11 is performed by, for example, when the failure occurs in the power supply line 130a (S11: YES), the power supply to the first server 101 is cut off. On the server 102,
The management data frequently transferred (T1) from the first server 101 will not be received suddenly without going through a normal termination procedure. Therefore, the second server 102 can confirm whether or not a failure has occurred in the first server 101 by confirming whether or not the received data has suddenly disappeared. In addition, in the second server 102, when no failure occurs (S11: NO), the occurrence of the failure can be constantly monitored by repeating step S11 again.

In the second server 102 which has detected the occurrence of a failure in the first server 101, the supply source of the management data to the management data changeover switch 105 is, for example, from the first server 101 to the second server. Then, the receiving side that is the destination of the management data is switched from the second server 102 to the third server 103, for example (S12). That is, since the second server 102, which was a spare server machine that only stores backup management data in the initial setting, becomes the main processing server machine that is the source of management data when a failure occurs, the management data The connection switch 105 for switching the connection with the second server 102 is switched to the output side of the management data, and the third server 103 (spare server machine) is newly added as the reception side of the management data.
A new connection is set up so that the can be connected.

Further, the second server 102, which has detected that the first server 101 has failed, also switches the main data changeover switch 106 so that the main data storage unit 1
The main data is sent and received between 07 and the client via the second server 102 (S12).

When a failure occurs, the second server 102 confirms the contents of the management data stored (S13), and starts processing from the job next to the most recently stored contents (final storage contents). The second server 102 continues to perform the processing that the first server 101 has performed halfway (S14).

In the conventional clustering system, by using the above configuration and method, even when a failure occurs in one server machine, the processing requested by the client is continued by another server machine. Since it can be carried out, the requested processing could be automatically restored.

[0029]

However, as described above, the transfer T1 of the management data from the first server 101 to the second server 102 is frequently performed, for example, every time one job is completed. Therefore, the processing efficiency of the CPU in the first server 101 is deteriorated. Furthermore, since the communication speed of the communication line 110, which is an Ethernet bus or the like, is slower than that of the internal bus 140 or the like in the first server 101, the processing efficiency of the first server 101 as a whole is also reduced.

In addition to the above-mentioned management data for failover such as journal data, the communication line 110 is
For example, since the management data of the first server 101 is frequently transferred T1 to the second server 102 using the communication line 110 because it is also used for load balancing and the like, clustering
The processing efficiency of the system 100 as a whole is also reduced.

The present invention has been made to solve the above-described conventional problems, and eliminates the inter-server transfer of failover management data between server machines during normal operation, thereby improving processing efficiency. To provide an improved clustering system.

[0032]

In order to achieve the above object, the clustering system of the present invention according to claim 1 is such that a plurality of server machines are connected to a changeover switch by communication lines and are operated cooperatively. This is a clustering system capable of continuing service to clients even if one server machine fails, and each server machine stores management data of processing executed in the server machine. And a battery that supplies power to at least the management data storage unit and the interface unit that electrically and mechanically connects the server machine and the communication line when a failure occurs in the server unit and the server machine. When a failure occurs in the first server machine in each server machine, the battery in the first server machine is used to supply power. The interface unit that receives the power supply transfers the management data read from the management data storage unit to the second server machine and continues the service to the client based on the management data received by the second server machine. Characterize.

The invention according to claim 2 is the clustering system according to claim 1, in which the interface unit of each server machine detects a fault occurring during normal operation and switches the power supply to a battery. After that, a failure operation control circuit for controlling the transfer processing of the management data to the second server machine is provided.

According to a third aspect of the present invention, in the clustering system according to the first aspect, the interface unit of each server machine generates a clock signal having a frequency lower than the frequency of the clock signal during normal operation. When a failure occurs in the low frequency clock generation circuit and the first server machine, the clock signal used in the interface unit is changed from the relatively high frequency clock signal used in normal operation to the low frequency clock signal. A clock switching circuit for switching is provided.

Further, the present invention of claim 4 is based on claims 1 to 3.
In the clustering system according to any one of
The interface section of each server machine is a board format or a card format that can be connected through a standardized predetermined slot, and is characterized by being integrated by incorporating a battery and a management data storage section.

Further, in the data recovery method at the time of failure in the clustering system of the present invention as set forth in claim 5, the plurality of server machines are connected to the changeover switch by the communication lines and are operated cooperatively, A method for recovering data when a failure occurs in a clustering system capable of continuing service to a client even if one server machine fails, wherein the first server machine controlling the service requested by the client is ,
When the management data is stored inside the first server machine and it is detected that a failure has occurred in the first server machine, the first server machine determines the inside of the first server machine. Of at least the storage unit for storing the management data and the interface unit for connecting to the communication line, the power supply switched to the built-in battery, and the first server machine transmits the management data from the interface unit to the second unit. And the second server machine uses the received management data to continue the service requested by the client.

The present invention of claim 6 is the method of recovering data when a failure occurs in the clustering system according to claim 5, wherein the first server machine switches the power source to an internal battery, It is characterized in that the clock signal is switched to a low frequency one, and the processing of transferring the management data to the second server machine when a failure occurs is carried out at a lower speed than in the normal operation.

[0038]

BEST MODE FOR CARRYING OUT THE INVENTION The present invention will be described below based on the illustrated embodiments.

FIG. 1 is a block diagram showing the configuration of a clustering system according to the first embodiment of the present invention.

In FIG. 1, parts having the same functions as those of the conventional clustering system shown in FIG.
The same reference numerals are given and duplicate description is omitted.

The clustering system 1 of FIG. 1 is similar to the conventional one in that it has a plurality of server machines, but each server machine has a first server 11 and a second server 1.
2, the third server 13, ..., The nth server 14 are different in internal configuration from the conventional one. The management data changeover switch 105, the main data changeover switch 106, the main data storage unit 107, the power supply unit 108, the communication line 110, the high-speed communication line 120, and the power supply line 130 which are other configurations are shown in FIG. It is similar to the conventional configuration.

In each server machine of this embodiment, the management data interface sections 50 to 53 inside each server are provided.
However, it has the function of converting or adjusting management data into a data format suitable for Ethernet communication, as in the past, or conversely converting or adjusting the Ethernet communication data into a data format suitable as management data inside each server. Management data storage units 60 to 63 that store management data such as journal data indicating the processing content or processing result requested for each server machine, and further to those management data storage units 60 to 63. Batteries 70 to 7 for supplying power source power when a failure occurs in power source power supply of
3 inside.

The batteries 70 to 73 are the management data storage unit 60.
In addition to the power capacities required for memory backup of Nos. 63 to 63, the power capacities required to transfer the storage contents of the management data storage units 60 to 63 to other server machines are also provided.

Further, the conventional management data storage units 160-1
63 generally uses a part of the main memory, which is the main work area of each server machine,
As shown in FIG. 5, the management data interface unit 1
The configuration is different from that of 50 to 153, and is connected by the internal bus 140 of each server machine. However, in this embodiment, the management data storage units 60 to 63 are arranged inside the management data interface units 50 to 53.
Further, as the management data storage units 60 to 63, for example,
64Mbit (1Mword x 16) which is a volatile memory
4 bits (4 bank) SDRAM (32 Mbytes in total) are used.

FIG. 2 is a diagram showing the internal configuration of the first server 11 in FIG.

Only the internal configuration of the first server 11 will be described below, and the second server 12 to the nth server 1 in FIG.
Since No. 4 has the same internal configuration as the first server 11, its description is omitted.

In the first server 11 of FIG. 2, in addition to the management data interface unit 50, a server main control unit 150 for executing processing relating to main data desired by the client, and the main data storage of FIG. Part 107
It has a main data interface section 160 for storing and reading main data.
The input / output unit between the server and the client, or other configuration as a server, is not described because it has little relation to the main configuration of the present invention and little difference from the conventional configuration. Further, the power source unit 108 is provided in each of the above units.
From the power supply line 130.

In the management data interface unit 50, in addition to the management data storage unit 60 and the battery 70, the management data changeover switch 1 via the communication line 110 is provided.
A communication control circuit 64 that controls communication with the server 05 and a server main control unit 150 sends and receives a signal to detect a failure in the first server 11, and if a failure is detected, the management data storage unit 60 is stored. A failure operation control circuit 65 for controlling the transfer of the data stored in the server to another server, and a clock switching circuit 66 for switching a clock signal to be used at the time of normal operation and at the time of failure under the control of the failure operation control circuit 65. A communication line connector 67 including a connector for connecting the communication line 110 to the first server 11; a low-frequency clock generation circuit 68 for supplying a relatively low-frequency clock signal for failure; By the control of the server internal connection unit 74 including a connector for connecting the data interface unit 50 to the internal bus 140, and the failure operation control circuit 65. Power supply unit 108 during normal operation and failure
A power supply switching circuit 75 that switches between external power supply power supplied from the battery 70 and internal power supply power supplied from the battery 70, and a high frequency clock generation circuit 78 that supplies a relatively high frequency clock signal for normal operation. is doing.

Further, in the fault battery operation range 56 to which power is supplied from the battery 70 at the time of fault, the management data storage unit 60, the communication control circuit 64, the fault operation control circuit 65,
A clock switching circuit 66, a communication line connecting section 67, and
A low frequency clock generator circuit 68 is included.

Since the management data interface section 50 is generally designed to support a standardized bus format such as Ethernet (registered trademark),
The configuration is also a board type or a card type incorporating an Ethernet (registered trademark) interface circuit, and by inserting the board type or card type interface into a connectable predetermined slot, electrical connection and mechanical The connection is made. Therefore, the battery 70 and the management data storage unit 60 are arranged on the board-type or card-type interface, that is, incorporated in and integrated with the interface.

A difference of about 10 times between the operating frequency by the high frequency clock generating circuit 78 which supplies the clock signal for the normal operation and the operating frequency by the low frequency clock generating circuit 68 which supplies the clock signal for the fault. There is. That is, the high frequency clock generation circuit 78
The operating frequency becomes 1/10 by switching from the low frequency clock generating circuit 68 to the low frequency clock generating circuit 68. Taking a specific communication speed as an example, the communication speed from 1 Gbps to 100 Mbps in full-duplex communication during normal operation.

The capacity of the battery 70 is required to be about 3000 mAh because the power consumption is required to be about 10 W in the normal operation communication system. When converted to a general dry battery (nickel-hydrogen battery), this corresponds to the size of a single size dry battery, which is the largest size commercially available, for three batteries. Mounting a battery of this size on a board or card is impractical. However, in the method using the low frequency of the present embodiment described above, the power consumption is 600 mW.
Power consumption is 200m as it can be suppressed to a certain degree
It will be about Ah. This is a dry battery (nickel hydrogen battery)
When converted to, the size of the AA size battery, which is the smallest size on the market (in dry batteries: excluding button batteries), is 1
It will be the number.

In the present embodiment, since the battery having a small size as described above can be used, for example, the interface can be in the form of a card such as a PCI adapter card and the battery can be built therein.

When the conventional clustering system is changed to the system of the present embodiment by utilizing the above board-type or card-type interface, it is only necessary to change the interface card. It is not necessary to purchase the system, and the conventional clustering system can be changed to the clustering system of the present embodiment with a small installation cost and a small change man-hour.

Next, the operation of saving the management data during the normal operation of the clustering system 1 shown in FIGS. 1 and 2 will be described. In this embodiment,
In normal operation, the management data is only saved in the management data storage section backed up by the battery, and in the same way as before, the management data is also saved in other server machines during dual operation. No processing is performed.

FIG. 3 is a flowchart of the management data saving operation in the clustering system 1 of this embodiment.

Here, for example, it is assumed that the first server 11 executes the process in response to a request from the client.

First, the first server 11 confirms whether or not a processing request from the client is issued (S21). If there is no processing request (S21: NO), the confirmation in step S21 is performed. By repeating the above, the request from the client is constantly monitored.

When there is a processing request (S21: YES), the first server 11 confirms whether or not the request content is the storage of management data (for example, journal data) (S22). When the management data is not saved (S22:
If NO, the normal process is executed according to the request contents (S25). When the management data is stored (S22:
If the answer is YES, the first server 11 executes the interrupt process for saving the management data (S23).

In the interrupt process for saving the management data in step S23, for example, when the process (job) currently being executed by the server main control unit 150 of the first server 11 is completed, the next job is started. Management data for the address position of the main data stored as the job result in the main data storage unit 107 is temporarily held, and the management data interface unit 50 is generated.
It is stored in the management data storage unit 60 provided therein. In the present embodiment, during normal operation, only the internal processing of the server machine that executes the processing for the main data is performed, and the transfer for duplicating the management data to other server machines is not executed. Therefore, the timing for restarting the next process (job) becomes earlier.

Next, it is confirmed whether or not the interrupt processing of step S23 is completed (S24). If it is completed (S24: YES), the procedure returns to step S21, and if it is not completed (S24: NO). Is step S23
Then, the interruption processing is continued.

During normal operation, for example, the first server 1
A management data storage request is generated each time one job in 1 is completed, and the contents of the updated management data are stored in the management data storage unit 60. Since this is processing only by the internal bus 140 or the like, which is relatively faster than the communication line 110 or the like, the holding time during the execution of the job, that is, the time at which the interrupt processing is executed is reduced as compared with the conventional one. To do.

While the normal processing is being performed as described above, for example, the power line 130a to the first server 11
The case where a failure occurs in the above will be described below.

FIG. 4 is a flow chart of the operation when a failure is detected in the clustering system 1 of this embodiment.

First, for example, in the first server 11 in which a failure occurs in the clustering system 1, the failure operation control circuit 65 constantly monitors the server main control unit 150 and the like. 65 confirms whether or not a failure has occurred in the first server 11 that is executing the process (S
31).

The confirmation of the occurrence of the failure in step S31 can be made, for example, when the failure occurs in the power supply line 130a (S31: YES).
Since the power supply to the power source is cut off, the failure operation control circuit 65 detects an abnormal drop in the voltage level. Therefore, the faulty operation detection circuit 65 confirms a sudden voltage level abnormality different from that during normal operation, thereby
It is possible to confirm whether or not a failure has occurred in 1. Further, in the failure operation control circuit 65, when the failure does not occur (S31: NO), the occurrence of the failure can be constantly monitored by repeating step S31 again.

The failure operation control circuit 65, which has detected the occurrence of a failure in the first server 11, controls the power supply switching circuit 75 from the power supply supplied from the power supply unit 108 during normal operation to the battery at the time of failure. Control is performed so as to switch to the power supply power supplied from 70 (S32). Further, the failure operation control circuit 65 disconnects the internal circuit of the server internal connection section 74 or the server main control section 150 and the like in the future from the circuit inside in order to avoid a malfunction when a failure occurs. The signal from is not received (S33).

Further, the failure operation control circuit 65 controls the clock switching circuit 66 from the high frequency clock generating circuit 78 during normal operation to the low frequency clock generating circuit 6 during failure.
The control is performed so as to switch to the 8 side.

By the above control, at least the battery operating range 56 at the time of failure in the management data interface section 50
Each unit inside of is operated by a clock signal having a frequency lower than that in a normal operation when the power supply power from the battery 70 is supplied when a failure occurs. Therefore, the battery 70 does not need the high power required for high-speed operation, and the battery capacity can be reduced.

Since the battery capacity can be reduced and the external dimensions of the battery mounted on the interface card can be reduced, the design of the interface card which needs to be designed within a predetermined dimension can be facilitated.

Thereafter, the failure operation control circuit 65 controls the communication control circuit 64 to read the management data stored in the management data storage unit 60 under the operating environment of the low frequency clock signal, and the communication line 110. And transfer to the second server 12 via the management data changeover switch 105 T2
Is carried out (S35).

At this time, the communication control circuit 64 causes the second server 1 to operate at the communication speed based on the low-frequency clock signal.
The management data stored in the management data storage unit 60 is burst-transferred in a single bank and then transferred to the second server 12.

By switching to the communication speed based on this low-frequency clock signal, the transfer time is increased by about 8 times as compared with the transfer time in the normal operation.
It takes about 3 seconds to transfer the bytes. However, since failover processing by a general clustering system requires at least about one minute, the decrease in communication speed in this embodiment is small in the delay of the entire failover processing. I don't think it matters.

Second receiving the management data transferred T2
In the server 12, the output source of the management data when the failure occurs is the first server 1 with respect to the management data changeover switch 105.
The first server is switched to the second server 12, and the transmission destination of the management data when the failure occurs is switched from the second server 12 to the third server 13 (S36). That is, in the initial setting, the second server 12, which is a standby server machine that receives management data when a failure occurs and continues the processing requested by the client, is the main processing server that internally stores the management data when a failure occurs. Since it becomes a machine, the second server 12 in the management data changeover switch 105
The connection part to and is switched so as to be the output side of the management data at the time of the occurrence of a failure, and the third server 13 (new standby server) is newly provided as the reception side of the management data when the failure occurs in the second server 12. The connection unit with the third server 13 is set so that the side server machine) is connected.

The second server 1 which has received the management data
2, the main data changeover switch 106 is also switched, and the main data storage unit 107 and the client are set so that transmission / reception of main data passes through the second server 12 (S36).

When the second server 12 receives the management data of the first server 11 due to the failure of the first server 11, the second server 12 confirms the content of the received management data (S3).
7) By starting the process from the job next to the last received content (final received content), the second server 12 can continue to perform the processing that the first server 11 has performed halfway. (S38).

Clustering system 1 of this embodiment
Now, by using the above configuration and method, even when one server machine fails, the processing requested by the client can be continuously executed by other server machines. It is possible to automatically recover the data of the performed processing.

As described above, the first server 11 of this embodiment
Then, the transfer T2 of the management data to the second server 12 is not performed during the normal operation, and is only when the failure occurs in the first server 11. Therefore, during normal operation, the first server 11 does not cause a processing delay due to the transfer of the management data to another server, so that the deterioration of the CPU processing efficiency is reduced.

Further, in the first server 11 of this embodiment,
Since the management data is backed up by the battery and stored in the management data storage unit 60 inside the first server 11, the communication speed of the communication line 110, which is an Ethernet bus or the like, is slower than that of the internal bus 140 or the like in the first server 11. Therefore, the processing efficiency of the server, which is lowered, can be improved.

Further, since communication data for failover such as journal data described above is not transmitted / received to / from the communication line 110 except when a failure occurs, for example, when the communication line 110 is used for load balancing, etc. The processing efficiency of the system can be improved.

In the above embodiment, the management data is transferred to the second server 12 when a failure occurs in the first server 11, but the present invention is not limited to this. Instead, for example, when a failure occurs in any of the n servers, the present invention can be applied to the case of designating an arbitrary server from the remaining servers and transferring management data.

Further, in the above-described embodiment, the management data interface section is the standardized Ethernet (registered trademark) and has the interface card format. Therefore, it is not necessary to change the entire clustering system, and the installation cost and the change can be changed. Although the number of steps has been reduced, the present invention is not limited to this, and the present invention can be implemented by applying other standardized bus formats or interface card formats, for example, to facilitate the design of the interface card. can do.

In the above embodiment, each server machine and the management data changeover switch 105 are connected by Ethernet bus, and each server machine and the main data changeover switch 106 are connected by optical fiber bus. However, for example, the system may be configured such that the connection by the Ethernet bus is omitted and all the connections are performed by the optical fiber bus.

The method of transferring the management data by reducing the power consumption by lowering the communication speed of this embodiment is not limited to the above-mentioned normal Ethernet (registered trademark) connection, but other network connection methods. For example, it can be applied to Gigabit Ethernet (registered trademark).

Further, in the above-described embodiment, when a failure occurs, the clock signal for transferring the management data is switched to a low speed (low frequency) and used. However, when there is enough battery capacity or originally, Since the frequency of the clock signal is low, the clock signal may be switched without switching when the power required to transfer the management data is small.

Further, in the above-mentioned embodiment, the case where the fault occurs in the power supply line has been described, but the present invention is not limited to this, and all kinds of faults occurring in each server, for example, the signal line. It can be applied to a failure of the internal processing circuit or the like.

The battery operating range at the time of occurrence of a failure in the above-described embodiment may be configured to include other circuits not shown.

[0088]

As described above, the clustering of the present invention
The system and the data recovery method at the time of occurrence of the failure are such that the transfer of the management data from the first server machine which is executing the main data processing to the spare second server machine is not executed during the normal operation. Since it is performed only when a failure occurs in the server machine of No. 2, there is no delay due to the transfer processing of the management data in the first server machine during normal operation, and the deterioration of processing efficiency can be improved.

Further, in the clustering system of the present invention and the method of recovering data when a failure occurs, the management data is backed up by the battery and is stored only inside the server machine. It is also possible to improve the decrease in internal processing efficiency of the server machine caused by the processing time difference from the bus.

Further, since the management data for failover such as journal data is not transmitted / received to / from the communication line of the clustering system of the present invention except when a failure occurs, the communication line can be frequently used for load balancing or the like. Therefore, the processing efficiency of the clustering system can be improved.

If the management data interface unit is of a standardized card format, it is not necessary to change the entire system when changing from the conventional clustering system to the clustering system of the present invention. The cost and man-hours for change can be reduced, and the size of the battery can be reduced, so that the interface card can be easily designed.

[Brief description of drawings]

FIG. 1 is a schematic diagram illustrating a clustering according to a first embodiment of the present invention.
It is a block diagram which shows the structure of a system.

FIG. 2 is a diagram showing an internal configuration of a first server in FIG.

FIG. 3 is a flowchart of a management data saving operation in the clustering system of the present embodiment.

FIG. 4 is a flowchart of an operation at the time of detecting a failure in the clustering system of this embodiment.

FIG. 5 is a block diagram showing a configuration of a conventional clustering system.

6 (a) and 6 (b) are flowcharts of the operation of redundantly storing management data in the conventional clustering system.

FIG. 7 is a flowchart of an operation when a failure is detected in the conventional clustering system.

[Explanation of symbols]

1,100 Clustering system, 11,10
1 1st server, 12, 102 2nd server, 1
3, 103 Third server, 14, 104 nth server, 50 to 53, 150 to 153 Management data interface section, 56 Battery failure operating range, 60 to
63, 160 to 163 Management data storage unit, 64 Communication control circuit, 65 Failure operation control circuit, 66 Clock switching circuit, 67 Communication line connection unit, 68 Low frequency clock generation circuit, 70 to 73 Battery, 74
Internal server connection, 75 power supply switching circuit, 78
High-frequency clock generation circuit, 105 management data changeover switch, 106 main data changeover switch, 107 main data storage section, 108 power supply section, 110 communication line, 120 high-speed communication line, 13
0 power supply line, 140-143 internal bus, 15
0 Main control unit for server, 160 Main data interface unit.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Tsunehiro Kajita             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office (72) Inventor Akihiro Ogura             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office (72) Inventor Mitsuhiro Nishida             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office (72) Inventor Atsuya Takeuchi             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office (72) Inventor Nobuyoshi Tanaka             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office (72) Inventor Hiroshi Ito             1623 1423 Shimotsuruma, Yamato-shi, Kanagawa Japan             BM Co., Ltd. Daiwa Office F term (reference) 5B011 DA02 DB21 EA01 EB03 FF01                       FF04 GG16 JA02 JA05 JB02                 5B034 AA04 CC01 DD05                 5B045 BB15 GG01 JJ44                 5B089 GA02 JA35 JB17 KA12 KB04                       KC28 MD09 ME09

Claims (6)

[Claims] 1. A clustering system in which a plurality of server machines are connected to a changeover switch by a communication line and are operated in a coordinated manner so that even if one server machine fails, the service to clients can be continued. Each server machine stores a management data storage unit for storing management data of processing executed in the server machine, and at least the management data storage unit and the server when a failure occurs in the server machine. A battery is provided that supplies power to an interface unit that electrically and mechanically connects the machine and the communication line, and if a failure occurs in the first server machine among the server machines, The interface unit that receives power supply from the battery in the server machine of No. 1 reads the management data read from the management data storage unit. A clustering system for transferring data to a second server machine and continuing service to the client based on the management data received by the second server machine. 2. The interface unit of each server machine is configured to transfer management data to the second server machine after detecting a failure occurring during normal operation and switching the power source to the battery. The clustering system according to claim 1, further comprising a failure operation control circuit for controlling. 3. A low-frequency clock generation circuit for generating a clock signal having a frequency lower than the frequency of the clock signal at the time of normal operation in the interface section of each server machine, and a failure occurs in the first server machine. In this case, a clock switching circuit that switches the clock signal used in the interface section from the relatively high frequency clock signal used during normal operation to the low frequency clock signal is provided. The described clustering system. 4. The interface unit of each server machine is a board type or a card type that can be connected in a standardized predetermined slot, and is integrated by incorporating the battery and the management data storage unit. The clustering system according to any one of claims 1 to 3, characterized in that: 5. A failure in a clustering system in which a plurality of server machines are connected to a changeover switch by a communication line and operate cooperatively so that even if one server machine fails, service to clients can be continued. In the data recovery method at the time of occurrence, the first server machine that controls the service requested by the client stores the management data inside the first server machine during normal operation, and the first server machine stores the management data. When it is detected that a failure has occurred in the server machine, the first server machine has an interface section for connecting at least a storage section for storing the management data and a communication line inside the first server machine. The power supply power supplied to the power supply unit to the power supply power supplied from the built-in battery. In the clustering system, the management data is transferred from the interface unit to the second server machine, and the second server machine continues the service requested by the client using the received management data. Data recovery method when a failure occurs. 6. The first server machine switches the power source to a built-in battery, and then switches a clock signal to a low-frequency one so that the management data at the time of the occurrence of the failure is stored in the second server machine. 6. The method for recovering data when a failure occurs in the clustering system according to claim 5, wherein the process of transferring the data to the cluster is performed at a speed lower than that during normal operation.