JPH10320223A - System, method for processing check point communication and storage medium recording the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To execute an application with communication processing at high speed by transmitting a communication packet as speedily as possible while dynamically changing the interval of check point sampling corresponding to the frequency and time interval of communication. SOLUTION: A communication interval observing means 104 is located between an input/output request holding means 103 and a transport 106 and each time a packet is transmitted/received, it is recorded in a packet transmission/ reception time table 107 for each connection when the latest packet is transmitted/received. This time table 107 holds the transmission time of the latest communication packet transmitted by an FT computer 101 and the time of the latest communication packet received by a communication means 105 and dispatched to the transport 106 for each connection. Next, a check point roll-back means 102 samples the next check point by calculating check point intervals while using check point sampling time based on the packet transmission/reception time table 107 and packet information.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention proceeds with processing while collecting checkpoints.
It relates to a suitable checkpoint communication processing system applied to a computer system that realizes recovery from failure by rolling back to the last collected checkpoint and re-executing the process, especially when to collect the next checkpoint. The present invention relates to a checkpoint communication processing system and a checkpoint communication processing method that can achieve a significant speedup of communication processing in a computer system by optimizing.

[0002]

2. Description of the Related Art Conventionally, processing is performed while collecting checkpoints, and when a failure occurs, it is possible to recover from the failure by rolling back to the last collected checkpoint and re-executing the processing. In such a computer system, it is necessary to store the issuance of communication output processing until the next checkpoint processing is started, and to issue the communication output processing collectively after the checkpoint.

[0003] This is because, when a failure occurs, the computer system returns to the immediately preceding checkpoint. Therefore, unless all the communication processes generated before the failure occurs are canceled, an application program or the like is issued at the time of recovery from the failure. This is because the consistency of the communication data obtained cannot be guaranteed.

Conventionally, in order to improve the deterioration due to the delay as much as possible, a certain time has been exceeded after the previous checkpoint collection.

[0005] The accumulated communication processing request exceeds a certain amount.

When one of the conditions is satisfied, a checkpoint is collected.

However, these methods have not been able to achieve a sufficient speed improvement.

[0008]

A computer system employing a conventional checkpoint-rollback system, for example, a fault-tolerant computer capable of coping with a failure or a failure (fault-tolerant computer, hereinafter referred to as an FT computer) In particular, in the case of an application that advances processing while exchanging information with another computer using communication, there is a problem that processing time becomes longer.

In view of the above, the present invention provides a checkpoint / rollback type computer system that transmits a communication packet as soon as possible by dynamically changing the checkpoint collection interval depending on the communication frequency and time interval. It is an object of the present invention to provide a checkpoint communication system and a checkpoint communication processing method that enable high-speed execution of an application involving communication processing.

[0010]

[Means for Solving the Problems]

(1) According to the present invention, when communicating with another computer on a network, a communication packet to be transmitted to the other computer is held until a checkpoint is collected, and the held communication packets are collectively collected after the checkpoint is collected. A checkpoint communication processing system for a computer that, by transmitting, rolls back to the last checkpoint taken when a failure occurs and transmits the held communication packet to re-execute processing and realize recovery from the failure A packet storage means for storing packet information, which is information of communication packets transmitted to and received from the other computer, and a reception time of the packet information of the communication packet to be transmitted which is received before the latest checkpoint is collected. Time changing means for changing to the collection time, and The time interval between the transmission time and the arrival time of the packet to be transmitted to the other computer and the packet received from the other computer is determined based on the updated packet information, and until the next checkpoint is collected based on this time interval. Check point interval calculating means for calculating the time of the check point, and the next check point is collected based on the time until the check point calculated by the check point interval calculating means is collected.

[0011] According to this configuration, the checkpoint interval can be taken with good timing by automatically changing the checkpoint interval, which was conventionally a fixed interval, according to the interval at which communication packets are transmitted and received. The packet can be transmitted smoothly.

(2) In the checkpoint communication processing system according to (1), the checkpoint interval calculating means searches for a received packet having the same address and port number as the transmitted packet of the packet information,
Calculate the difference between the reception time of the transmission packet and the reception time of the reception packet and use this as the transmission processing time, search for a transmission packet of the same address and port number as the reception packet of the packet information, and search the transmission packet of the reception packet and the transmission packet. Calculate the difference between the reception times and use this as the reception processing time, add the transmission processing time and the reception processing time until the next checkpoint collection, divide this by the number of calculated differences, and then calculate the checkpoint It is characterized by the time taken until the sample is collected.

According to this configuration, by automatically changing the interval between checkpoints according to the interval at which communication packets are transmitted / received, checkpoints can be taken with good timing and packets can be transmitted smoothly.

Further, by calculating the average of the transmission processing time and the reception processing time of the communication packet and using the average as the check point interval, it is possible to determine the check point interval more suitable for the actual processing.

(3) In the checkpoint communication processing system according to (1) or (2), the checkpoint interval calculating means may include a maximum value and a minimum time until a preset checkpoint is collected. When the time required to collect the calculated checkpoint based on the value is equal to or more than the maximum value, the value is set to the maximum value, and when it is equal to or less than the minimum value, the value is set to the minimum value.

According to this configuration, by automatically changing the interval between checkpoints according to the interval at which communication packets are transmitted and received, checkpoints can be taken with good timing and packets can be transmitted smoothly.

Also, the maximum value and the minimum value of the checkpoint interval are set in advance, and when the calculated checkpoint exceeds the maximum value, the maximum value is set, and when the calculated checkpoint falls below the minimum value, the minimum value is set. By doing so, I / O requests other than communication processing caused by too long a checkpoint interval are prevented from delaying their response time, and checkpoints caused by a too short checkpoint interval are collected. The checkpoint interval can be adjusted to an appropriate value by preventing the time required for the system from being increased and losing the processing time of the system.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION

(First Embodiment) An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a system configuration diagram of the first embodiment. The FT computer 101 assures that, in addition to the elements constituting a normal computer, it can be returned and re-executed when a failure occurs in order to realize checkpoint rollback as described in the related art. Proceed with processing while collecting checkpoints as appropriate, and when a failure is detected, return the processor or memory to the state before the failure that was previously collected, remove the failure, and re-execute Checkpoint / rollback means 102 and requests to all input / output devices (input / output requests)
Is held until the checkpoint is collected and the status is determined.

The checkpoint / rollback means 102 records a checkpoint collection time indicating when the latest checkpoint was collected. The conventional method of determining when to collect checkpoints is to collect checkpoints at predetermined time intervals, or to determine whether the input / output request holding unit 103 has accumulated a certain number or more of input / output requests. There is. In the present embodiment, as will be described later, a new method for determining the timing at which checkpoints are collected is proposed.

Further, the FT computer 101 includes a transport 106 used for communication with a normal computer, and a communication means 105.

The communication means 105 is means for operating an interface for connecting to an Ethernet (Ethernet) or the like often used in a LAN.
Is a means for transmitting and receiving communication packets to and from another computer. Here, for example, TCP / IP (Trans
Mission Control Protocol /
Internet Protocol).

The communication interval observing means 104, as shown in FIG.
06, each time a packet is transmitted / received, the time at which the latest packet was transmitted / received is recorded in the packet transmission / reception time table 107 for each connection.

The packet transmission / reception time table 107 has F
The transmission time of the latest communication packet transmitted by the T computer and the time of the latest communication packet received by the communication means 105 and passed to the transport 106 are held for each connection.

The network 109 is a LAN (Local
1 Area Network, etc., and connects the FT computer 101 and a client computer or the like that receives data from the FT computer 101 and the like.

FIG. 2 shows a packet transmission / reception time table 10.
7 is an example of data storage. In TCP / IP, the address (IP address) of a computer at the other end of communication is stored, and a virtual connection is determined by a port number indicating an application in the computer. The packet transmission / reception time table 107 stores the address and the port number of the communication packet in the address 201 and the port number 202. Further, a flag indicating whether a communication packet is transmitted or received is stored in the transmission / reception flag 203. The time 204 stores the time at which the communication packet was received.

FIG. 3 is a flow chart showing the processing of the communication interval observation means 104. Here, each time a packet is transmitted or received, the packet transmission / reception time table 107 indicates when the latest packet has been transmitted / received for each connection.
To record.

To describe each step, first, it is determined whether the transmitted / received communication packet is a communication packet for disconnecting the connection with the communication partner (A1).

If the packet is a communication packet for disconnecting the connection (YES in A1), the address and port number are extracted from the communication packet, and an entry having the same address and port number is searched from the packet transmission / reception time table 107 and deleted. (A2) The processing ends.

If the communication packet is not a communication packet for disconnecting the connection (NO in A1), it is determined whether the communication packet is a transmission packet transmitted from the FT computer (A).
3).

If the packet is a transmission packet (YE of A3)
S), and extracts the destination address and port number from the transmission packet (A4). Next, the extracted address, port number,
The transmission flag and the current time are recorded (A5), and the process ends.

If the packet is not a transmission packet (NO in A3),
Since the received packet is received by the FT computer 101,
The source address and port number are extracted from the received packet (A6). Next, the extracted address, port number, reception flag, and current time are recorded in the packet transmission / reception time table 107 (A7), and the process ends.

FIG. 4 is a flowchart showing the processing of the checkpoint / rollback means 102. The checkpoint collection time indicates the time at which the latest checkpoint was collected, and the checkpoint interval indicates the time until the next checkpoint / roll pack unit 102 takes a checkpoint.

To describe each step, the checkpoint collection time is set to the current time, and the checkpoint is initialized to a predetermined value (B1). Then, the operation is stopped for a predetermined checkpoint interval (B2).

Next, the packet transmission / reception time table 107
It is determined whether all the entries have been scanned (B3).

When all entries are scanned (YE of B3)
S), a checkpoint is collected (B4). And
The checkpoint interval is recalculated using the packet transmission / reception time table 107 and the checkpoint collection time (B5). The details of this recalculation method will be described later. Next, the checkpoint collection time is set as the current time (B6).

When all the entries have not been scanned (B3
NO), the next entry of the packet transmission / reception time table 107 is taken out (B7). Next, it is determined whether the extracted packet is a transmission packet (B8).

In the case of a transmission packet (YES in B8), it is determined whether this packet was registered before the checkpoint collection time (B9). If it has not been registered before the checkpoint collection time (NO in B9), an entry is returned to the packet transmission / reception time table 107 (B1).
0). If the packet was registered before the checkpoint collection time (YES in B9), that is, in this case, the transmission packet was received before the previous checkpoint. Therefore, the reception time of the packet that has already passed the checkpoint and is ready to be transmitted is changed to the checkpoint time (that is, the time at which the previous checkpoint was collected). Since the checkpoint time is the current time, the time of the entry is set to the current time (B11),
Return the entry to the packet transmission / reception time table (B1
0).

If the packet is not a transmission packet (NO in B8),
Return to B3.

By the above processing, the processing time of the client computer (the time from when the transmission packet is received from the FT computer to when the packet is transmitted to the FT computer) described below can be accurately obtained. This is because the actual transmission of the above packet is delayed up to the checkpoint, and if the time accepted by the communication interval observing unit 104 is used as it is, the checkpoint delay time is added.

Next, how the checkpoint / rollback means 102 calculates the checkpoint interval using the packet transmission / reception time table 107 and the checkpoint collection time described above (B5 in FIG. 4). )
Will be described in detail.

FIG. 5 is a conceptual diagram showing the timing of transmitting and receiving packets exchanged between an application running on the FT computer and an application running on a certain computer on the network.

The application on the FT computer 101 transmits the packet 511, and then receives the packet 5 from the application on the client computer of the communication partner.
Receive 12. The transmission packet 513 is further transmitted based on the contents.

The transmission packet 511 is first sent to the packet transmission / reception time table 10 by the communication interval observing means 104.
7 is registered. The reception time is time 501. Next, transmission is delayed until check point 521. At the time of this checkpoint 521, the process of B11 in FIG. 4 is performed, so that the reception time of the transmission packet 511 is
It is changed to 2. At the time of the check point 522, the following calculation is performed.

Received packet 512 and transmitted packet 511
The time difference 507 of the packet transmission / reception time table 1
Calculate from 07. Packet transmission / reception time table 107
Records the reception time 504 of the reception packet 512 and the reception time 502 of the transmission packet 511.
The difference 507 can be obtained by using those values. The difference 507 can be considered as the processing time until the next packet of the other party application is transmitted.

Next, a time difference 508 between the transmission packet 513 and the reception packet 512 is obtained. This can also be calculated from the entry of the packet transmission / reception time table 107. This difference 508 can be considered as the processing time of the application on the FT computer 101.

Processing time 507 of partner application
Then, the time obtained by adding the processing time 508 of the application on the FT computer 101 becomes the time suitable for checkpoint collection. Because at this time interval, FT calculator 1
01 sends an outgoing packet,
Because it can be predicted.

Checkpoint / rollback means 10
In step 2, immediately before the checkpoint is collected, the following calculation is performed to determine the next checkpoint interval.

Checkpoint / rollback means 10
2 is shown in the flow charts of FIGS. 6, 7 and 8. 6 and 7 show the times 507 and 5 described above.
08 is calculated and its average is calculated.

In FIG. 8, the checkpoint interval is calculated based on the value.

The process of FIG. 6 for calculating the average time until transmission will be described.

It is determined whether the entire packet transmission / reception time table 107 has been scanned (C1).

If all scans have been performed (YES in C1), the total processing time divided by the number of packets is used as the average time until transmission (C2).

If all scans have not been performed (N of C1)
O), it is determined whether the scanned transmission packet is a transmission packet received after the checkpoint collection time (C3).

When the scanned transmission packet is received after the check point (YES in C3),
It is determined whether there is a received packet having the same address and port number as the transmitted packet (C4). When there is a received packet with the same address and port number (Y in C4)
ES), the difference between the reception time of the transmission packet and the reception time of the reception packet is calculated, the number of packets is increased by one in addition to the total processing time (C5), and the process returns to C1. If there is no received packet with the same address and port number (NO in C4), C
Return to 1. Here, the total processing time is the sum of all the calculated differences up to the next checkpoint collection,
The average time until transmission is obtained by dividing this by the number of differences (here, the number of packets).

If the scanned transmission packet is not one received after the check point (NO in C3),
Return to 1.

Next, a description will be given of how to calculate the average time until reception in FIG.

It is determined whether the entire packet transmission / reception time table 107 has been scanned (D1).

If all scans have been performed (YES in D1), the total processing time divided by the number of packets is used as the average time until reception (D2).

If all scans are not performed (N of D1)
O), it is determined whether the scanned received packet is a received packet received after the checkpoint collection time (D3).

If the scanned received packet is received after the checkpoint (YES in D3),
It is determined whether there is a transmission packet having the same address and port number as the reception packet (D4). When there is a received packet with the same address and port number (Y in D4)
ES), the difference between the reception time of the transmission packet and the reception time of the reception packet is calculated, the number of packets is increased by one in addition to the total processing time (D5), and the process returns to D1. If there is no received packet with the same address and port number (NO in D4), D
Return to 1. Here, the total processing time is the sum of all the calculated differences up to the next checkpoint collection,
The value obtained by dividing this by the number of differences (here, the number of packets) is the average time until reception.

If the scanned received packet is not the one received after the check point (NO in D3), D
Return to 1.

Next, a process of calculating the checkpoint interval based on the average time until transmission and the average time until reception obtained in FIGS. 6 and 7 will be described with reference to the flowchart in FIG.

The average time until transmission in the processing of FIGS.
The average time until reception is calculated (E1). Next, the checkpoint interval is set to “average time until transmission + average time until reception”.

Since the checkpoint interval cannot be changed indefinitely, a maximum value and a minimum value are set in advance so as not to exceed the range. If the interval between checkpoints is too long, processing of input / output requests other than communication will be delayed for a long time, which has the effect of lengthening the response time. Also, if the checkpoint interval is short, the time for collecting checkpoints increases, and the processing time of the system is lost.

It is determined whether the check point interval does not exceed a preset maximum value (E3).

If the checkpoint interval exceeds the maximum value (NO in E3), the checkpoint interval is set to the maximum value (E5), and the process is terminated.

If the checkpoint interval does not exceed the maximum value (YES in E3), it is determined whether the checkpoint interval is less than a preset minimum value (E3).
4).

When the value is not below the minimum value (YE of E4)
S), the process ends. If below the minimum value (E
(NO in 4), the checkpoint interval is set to the minimum value (E6), and the process ends.

As described above, in the present embodiment, the checkpoint interval is dynamically changed according to the values of the packet transmission interval and the reception interval, so that the checkpoint is taken immediately after the packet transmission occurs. Then, the packet to be transmitted can be transmitted immediately.

[0071]

In an FT computer using a normal checkpoint rollback method, checkpoints are often taken at fixed time intervals. According to the present invention, by automatically changing the interval of the checkpoint according to the interval at which the packet is exchanged, the checkpoint can be taken with good timing, and the packet can be transmitted smoothly.

Further, according to the present invention, the average of the transmission processing time and the reception processing time of the communication packet is calculated, and this is used as the check point interval, so that the check point interval more suitable for the actual processing is determined. can do.

According to the present invention, the maximum value and the minimum value of the checkpoint interval are set in advance, and when the calculated checkpoint exceeds the maximum value, the maximum value is set, and the checkpoint interval is set to the minimum value. If the checkpoint interval becomes too short, the checkpoint interval is too long to prevent I / O requests other than communication processing from being delayed, thereby preventing their response time from becoming long, and the checkpoint interval becoming too short. As a result, it is possible to adjust the checkpoint interval to an appropriate value by preventing the time for collecting checkpoints from being increased and losing the processing time of the system.

[Brief description of the drawings]

FIG. 1 shows a schematic configuration of a checkpoint communication processing system according to a first embodiment of the present invention.

FIG. 2 is an exemplary diagram showing an example of data storage in a packet transmission / reception table according to the first embodiment according to the embodiment;

FIG. 3 is a flowchart showing processing of a communication interval observing unit according to the first embodiment, according to the embodiment;

FIG. 4 is a flowchart showing a processing operation of a checkpoint / rollback unit according to the first embodiment according to the embodiment;

FIG. 5 is a conceptual diagram of packet transmission / reception and a checkpoint according to the first embodiment according to the first embodiment;

FIG. 6 is a flowchart illustrating a process of calculating an average time until transmission of a checkpoint / rollback unit according to the first embodiment;

FIG. 7 is a flowchart showing a process for calculating an average time until reception of a checkpoint / rollback unit according to the first embodiment according to the embodiment;

FIG. 8 is a flowchart showing a process of calculating a checkpoint interval of the checkpoint / rollback unit according to the first embodiment according to the embodiment;

[Explanation of symbols]

 Reference Signs List 101 FT computer 102 Checkpoint / rollback means 103 Input / output request holding means 104 Communication interval observing means 105 Communication means 106 Transport 107 Packet transmission / reception time table 108 Application 109 Network

Claims (9)

[Claims] 1. A communication packet to be transmitted to another computer when communicating with another computer on a network is held until a checkpoint is taken, and after the checkpoint is taken, the held communication packets are transmitted collectively. By doing so, when a failure occurs, the checkpoint communication processing system of the computer that rolls back to the last collected checkpoint and transmits the held communication packet to re-execute the processing and realize recovery from the failure, A packet storage unit for storing packet information that is information of communication packets transmitted to and received from the other computer, and a reception time of packet information of communication packets to be transmitted received before a current checkpoint is collected, and a latest checkpoint collection time. Time changing means for changing to The time interval between the transmission time and the arrival time of the packet to be transmitted to the other computer and the packet received from the other computer is determined based on the updated packet information, and until the next checkpoint is collected based on this time interval. A checkpoint interval calculating means for calculating the time of the checkpoint, and collecting the next checkpoint based on the time until the checkpoint calculated by the checkpoint interval calculating means is collected. . 2. The checkpoint communication processing system according to claim 1, wherein the packet information is an address and a port number of the other computer, a distinction between transmission and reception of the communication packet, and a time when the communication packet is received. . 3. The checkpoint interval calculation means,
A search is made for a reception packet having the same address and port number as the transmission packet of the packet information, and a difference between the reception time of the transmission packet and the reception time of the reception packet is calculated as a transmission processing time, which is the same as the reception packet of the packet information. Search the transmission packet of the address and port number of the received packet, calculate the difference between the reception time of the reception packet and the reception time of the transmission packet, and use this as the reception processing time. The transmission processing time and the reception processing time until the next checkpoint collection 3. The checkpoint communication processing system according to claim 1, wherein a value obtained by dividing the calculated value by the calculated number of differences is used as a time until a next checkpoint is collected. 4. The checkpoint interval calculation means,
If the time before collecting the calculated checkpoint based on the maximum value and the minimum value of the time before collecting the preset checkpoint is greater than or equal to the maximum value, the value is less than or equal to the maximum value and less than or equal to the minimum value 4. The checkpoint communication processing system according to claim 1, wherein the minimum value is set to the minimum value in the case of. 5. A communication packet to be transmitted to another computer when communicating with another computer on a network is held until a checkpoint is collected, and the collected communication packets are transmitted collectively after the checkpoint is collected. By doing so, when a failure occurs, the checkpoint communication processing system of the computer that rolls back to the last collected checkpoint and transmits the held communication packet to re-execute the processing and realize recovery from the failure, Stores packet information, which is information on communication packets transmitted to and received from the other computers, and changes the reception time of the packet information of communication packets to be transmitted that was received before the current checkpoint was collected to the latest checkpoint collection time Based on the changed packet information The time interval between the transmission time and the arrival time of the packet to be transmitted to the other computer and the packet received from the other computer is obtained, and the time until the next checkpoint is obtained is calculated based on the time interval. A checkpoint communication processing method, wherein the next checkpoint is collected based on the calculated time until the checkpoint is collected. 6. The checkpoint communication processing method according to claim 5, wherein the packet information is an address and a port number of the other computer, a distinction between transmission and reception of the communication packet, and a time when the communication packet is received. . 7. When calculating the time until the checkpoint is collected, a reception packet having the same address and port number as the transmission packet of the packet information is searched, and a reception time of the transmission packet and the reception packet is received. Is calculated, and this is set as a transmission processing time. A transmission packet having the same address and port number as the reception packet of the packet information is searched, and a difference between the reception time of the reception packet and the reception time of the transmission packet is calculated and received. The processing time is defined as the time until the next checkpoint is obtained by adding the above-mentioned transmission processing time and the above-mentioned reception processing time until the next checkpoint is obtained, and dividing this by the number of differences calculated above. 7. The checkpoint communication processing method according to claim 5, wherein: 8. When calculating the time until the checkpoint is collected, a time until the calculated checkpoint is collected based on the maximum value and the minimum value of the preset time before the checkpoint is collected. 8. The checkpoint communication processing method according to claim 5, wherein the time is set to the maximum value when the time is equal to or more than the maximum value, and is set to the minimum value when the time is equal to or less than the minimum value. 9. A storage medium storing a program for the checkpoint communication processing method according to claim 5.