JP2008009797A - Uninterruptible memory replication method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method not to interrupt a call processing service even during data transfer processing about an uninterruptible memory replication method. <P>SOLUTION: In a server for performing a call processing service of an active/hot standby configuration, an active system for memory replication divides a call processing process into a thread for call processing service and a thread for memory transfer, respectively allocates the threads to different CPUs, and transfers copy data of a memory asynchronously with committing memory contents of the call processing service to a hot standby system without interrupting all call processing services including a call for committing the memory contents and performing memory transfer, and the hot standby system starts a data reception thread started from a call processing process in a standby state so as to rearrange the copy data of the memory transferred form the active system in an actual data area. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an uninterrupted replication method. Here, replication refers to obtaining a copy of data.

  A server-based call processing server stores session (call) information as data on a memory, and is called when a system is switched (switched over) between Active / Hot-Standby (active / hot standby) servers. In order to remedy this, data transfer between the active / hot standby servers is necessary. This transfer processing has been a factor that hinders call processing performance due to the huge amount of transfer.

  In recent years, with the progress of multi-CPU servers and multi-core CPUs, it has become possible to use multiple CPU cores within a single server. It is becoming possible to take measures.

  FIG. 10 is a diagram showing a system configuration example that is an object of the present invention. In the figure, reference numeral 10 denotes a call processing server, which is divided into an active system and a hot standby system. Assume that # 1 is an active system and # 2 is a hot standby system. In the call processing server 10, 1 is a logical memory space, and 2 is a call processing process. Reference numeral 3 denotes a thread (processing unit) provided in the call processing process 2. In the hot standby server 10, 4 is a CPU and is connected to the thread 3 in the call processing process 2.

  Reference numeral 6 denotes a plurality of CPUs provided in the active call processing server 10. Here, the case where CPU1 to CPU4 are provided as CPU4 is shown. Each CPU 6 is connected to a thread 3 in the call processing process 2, and a set of a plurality of threads is provided for each CPU. Reference numeral 5 denotes a NIC (Network Interface Card) for connecting the call processing server 10 to the outside, and is connected to a LAN (Local Area Network) 15. The call processing server # 1 and the call processing server # 2 perform data communication via the LAN 20.

  The call processing server 10 establishes a session (call) such as SIP (Session Initiation Protocol) used in the field of VoIP (voice processing) and TV (television) telephone, etc. This is a server for establishing real-time communication between End-to-End.

  In the call processing server 10, a call processing process for managing calls is operating, and the call processing process is a multi-thread configured by a plurality of threads allocated to any one of a plurality of CPUs by a thread scheduler such as an OS (operating system). It is a thread configuration process. The same logical memory space can be referenced from threads on the same process.

  The call processing server 10 has a dual configuration of an active system and a hot standby system, and is connected by a LAN 15 via a server NIC. On the hot standby side, the same call processing process operates with the same hardware configuration, and the call processing can be started immediately upon a switchover. The logical memory space 1 that can be referenced from the call processing process has the same size and configuration as the active system.

  FIG. 11 is an explanatory diagram of memory areas distributed on the server. A call is composed of several functions such as exclusive control, call control, protocol control, and service scenario. Each functional object has a data area called call-related data associated with a call. As shown in FIG. 11, there are a method in which the data area is localized for each function as shown in FIG. In the figure, a program area is an area for a program to operate, and a call related data area is an exclusive control call related data area, a call control call related data area, a service scenario call related data area, and a protocol controlled call. Divided into related data areas. That is, call-related data for each function is localized for each function. 17 of the figure shows call related data used by the instance.

  FIG. 12 is a diagram showing a boot flow of the call processing server. First, the operation of the active server will be described. First, when the server power is turned on (S1), the OS is started (S2). Next, the cluster middle is activated (S3). Here, the cluster middle is middleware for switching and controlling the system. Middleware is a bridge between an OS and an application.

  Thereafter, the system is determined (S4). Here, the active system is determined. Next, the application is activated (S5), and a call processing service is started (S6). Here, at the start of the service, the peripheral device is notified of activation of the new active system (request for rewriting of the address table). Thereafter, the link establishment is waited (S7).

  Next, the operation of the hot standby server will be described. First, the server power is turned on (S8). Next, the OS is activated (S9). Next, the cluster middle is activated (S10). Thereafter, the system is determined (S11). Here, it becomes a hot standby system. Next, the application is activated (S12) and waits for link establishment (S13).

  After the above operation, a link is established between the active server and the hot standby server (S14). When the link is established, full copy processing is performed from the active system to the hot standby system memory (S15). The system switching operation ends with the above sequence. Next, a memory copy is performed for each call from the active system to the hot standby system (S16).

  FIG. 13 is a diagram showing switch overflow of the call processing server. The operations of the active server and the hot standby server are shown. First, the memory for each call is being copied from the active system to the hot standby system (S1). Here, an abnormality is detected in the active system (S2). The operation after detecting an abnormality (whether the server is switched immediately, the call rescue is restarted once in the own system, the accepted event is processed, or discarded) is implementation-dependent.

  When an abnormality is detected, the active server issues a switch start notification to the hot standby server (S3). The hot standby system (new active server) changes the state from the hot standby to the active system (S4). Then, the application is activated (S5) and waits for link establishment (S6). In the active system (new hot standby system) in which the abnormality is detected, the OS is restarted (S7). Hereinafter, the process until the link establishment wait is the same as the hot standby system boot process flow shown in FIG. The new hot standby system waits for link establishment (S8).

  When a link is established in this state (S9), a full memory copy operation from the active system to the hot standby system is performed (S10). This completes the switch overflow process. Next, a memory copy operation for each call from the active system to the hot standby system is started (S11).

This type of conventional device manages the signal transmission / reception status between the call control system device and the controlled system device, and when the health check signal or call processing signal cannot be transmitted / received, the failure point is on the call control device side. If the same call control system device controls multiple controlled system devices, the signal transmission / reception with other controlled system devices is performed. A technique is known in which the state is checked to perform the separation (see, for example, Patent Document 1).
JP 2003-78567 A (paragraphs 0012 to 0018, FIGS. 1 to 3)

  The conventional office switch employs a hardware duplication system having an active / hot standby configuration, and is realized by a hardware DMA transfer system as a duplication system between active / hot standby. As a result, even if active / hot standby system switching occurs, the same contents are always maintained in both memory systems, and calls that have progressed halfway can be relieved and processing can be continued.

  However, in order to realize call relief at the time of an active / hot standby switchover in a call processing server such as a SIP server using a general-purpose server, data transfer via an NIC between active / hot standby servers in a cluster system It is necessary to realize memory duplication by.

  If the function to transfer all generated / changed / deleted data via the NIC to the hot standby system is performed on the same thread as the thread that performs the call processing, the processing on the multi-thread / multi-CPU is also used for the call processing. Since this occupies one of the CPUs, the call processing service is interrupted and hindered.

  FIG. 14 is an explanatory diagram of conventional problems. If there is a call processing thread 1 and transfer processing is performed to the hot standby server on the same thread, the call processing thread 1 is eaten up for transfer processing during this time, and call processing that has entered another waiting queue There is a problem that threads cannot work.

  The present invention has been made in view of such problems, and an object of the present invention is to provide an uninterrupted memory replication method that does not require interruption of a call processing service even during data transfer processing.

  (1) According to the first aspect of the present invention, in a server that performs a call processing service in an active / hot standby configuration, a call processing process is performed in an active system for memory replication, and a thread that performs the call processing service and a memory are transferred. Trigger the memory contents of the call processing service to the hot standby system without interrupting all the call processing services including calls that are divided into threads, assigned to different CPUs, commit the memory contents, and perform memory transfers In the hot standby system, the data reception thread started from the call processing process in the hot standby state is activated to actually copy the memory copy data transferred from the active server. Relocation to the data area of And wherein the door.

  (2) According to the second aspect of the present invention, in the server that performs the call processing service in the active / hot standby configuration, in the active system, for the memory replication, the call processing process is performed in the active system, and the memory is copied. It is divided into a thread and a thread that performs memory transfer, and each is assigned to a different CPU, so that the memory contents are committed, and the call processing service except for calls that perform memory transfer is not interrupted. The memory copy is transferred in synchronization with the commit timing of the memory contents of the call processing service. In the hot standby system, the data reception thread started from the call processing process in the hot standby state is started and transferred from the active system. Copied data in the actual data area Rearranging, as is characterized in that the.

  (3) The invention according to claim 3 is the server that performs the call processing service in the active / hot standby configuration, and the hot standby server that has received the data from the active server receives the data from the call processing process in the hot standby state. The reception thread and the copy data relocation thread are activated, and at least the data reception thread is assigned to a CPU different from the relocation thread, so that the copy data of the memory transferred from the active system is temporarily copied to the buffer and queued. It is characterized in that it is rearranged in the actual data area later.

  (4) According to the invention of claim 4, when starting a thread that performs copying of the memory, it is not assigned to a fixed CPU, but is based on any of the CPUs other than the CPU assigned to the data transfer thread. Also, a CPU having a low CPU usage rate is searched and assigned to the CPU.

  (1) According to the first aspect of the present invention, the call processing can be continued without affecting the transfer delay between servers across the network by dividing the call processing and data transfer into two threads.

  (2) According to the invention described in claim 2, the call processing, data copy, and data transfer are divided into three threads, and the call processing is stopped during the data copy. Will be unaffected and call processing will continue. According to the present invention, even if the contents of the memory are rewritten during the continuous call processing, the copy is already taken into another buffer, so that it is not affected.

(3) According to the invention described in claim 3, it is possible to eliminate incompleteness of received data on the hot standby side.
(4) According to the invention described in claim 4, the CPU for allocating data copy processing according to claim 2 is allocated to an empty CPU without being fixed, and one CPU is occupied for copy processing with a processing time shorter than that for data transfer. Without using the call processing idle time, the CPU can be used effectively and the call processing performance can be improved.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The present invention described below operates with the system configuration shown in FIG.
FIG. 1 is a diagram showing a configuration example of the first invention. The same components as those in FIG. 10 are denoted by the same reference numerals. In the figure, 10 is a server, # 1 is an active server that is an active call processing server, and # 2 is a hot standby server that is a hot standby call processing server. Reference numeral 6 denotes a CPU provided in the active server 10. Here, the case where CPU1 and CPU2 are provided is shown. Reference numeral 4 denotes a CPU provided in the hot standby server 10. Although only one is shown in the figure, a plurality of CPUs 4 can be provided as necessary.

  On the active server 10 side, the threads are divided into threads that perform call processing (thread 1 and thread 2) and threads that perform data transfer. The CPU 2 is in charge of the call processing threads 1 and 2, and the CPU 1 is in charge of the data transfer thread. On the hot standby server 10 side, the data reception thread operates for data reception processing, and specifically, the CPU 4 is in charge. In the active server 10, the data transfer thread is connected to the hot standby server 10 via the NIC 5 and the LAN 15 (see FIG. 10). The operation of the system configured as described above will be described as follows.

  In the server 10 that performs the call processing service in the active / hot standby configuration, for memory replication, the call processing process in the active system is divided into a thread that performs the call processing service (call processing thread) and a thread that transfers the memory (data transfer thread). ). Each is assigned to another CPU. Here, it is assumed that an arbitrary thread in one process can be allocated to a fixed CPU as a function of the OS or middleware.

  As is apparent from FIG. 1, the data transfer thread is assigned to CPU 1 and the call processing thread is assigned to CPU 2. By dividing each call processing thread and data transfer thread into different CPUs, the memory contents are committed (confirmed), and all call processing services including calls that perform memory transfer are performed without interruption. The copy data of the memory is transferred to the standby system asynchronously with the commit timing of the memory contents of the call processing service. On the hot standby system, the data reception thread started from the call processing process in the hot standby state is started to activate the active system. The copy data of the memory transferred from is transferred to the actual data area. Here, as shown in the figure, it is assumed that there is no possibility that the memory is rewritten in the section (ΔT) in which the call processing thread 1 performs data transfer. This is because if the data is rewritten during the transfer process during this period Δ, the identity of the active system and the hot standby system is lost.

  As described above, according to this embodiment, by dividing the call processing thread into two threads of call processing and data transfer, call processing can be continued without affecting the transfer delay between servers across the network. become.

  FIG. 2 is a diagram showing a configuration example of the second invention. The same components as those in FIG. 10 are denoted by the same reference numerals. In the figure, reference numeral 10 denotes an active server, and the CPU 6 shows a case where three CPUs 1 to 3 are provided. The call processing thread 1 and the call processing thread 2 are assigned to the CPU 3, the transfer data copy thread is assigned to the CPU 2, and the data transfer thread is assigned to the CPU 1. The operation of the system configured as described above will be described as follows.

  In the server 10 that performs a call processing service in an active / hot standby configuration, in an active system for memory replication, a thread that performs a call processing service for a call processing process (call processing thread) and a thread that performs a memory copy (transfer data) A copy thread) and a memory transfer thread (data transfer thread), which are allocated to different CPUs.

  In this way, the memory contents are committed in synchronization with the commit timing of the memory contents of the call processing service in the hot standby system without interrupting the call processing service except for the call for performing the memory transfer. The copy is transferred, and in the hot standby system, the data reception thread activated from the call processing process in the hot standby state is activated, and the memory copy data transferred from the active system is relocated to the actual data area. It has become.

  As described above, according to this embodiment, the call processing, data copying, and data transfer are divided into three threads, and the call processing is stopped during the data copying. Will be unaffected and call processing will continue. According to this embodiment, even if the contents of the memory are rewritten during the continuous call processing, the copy is already taken into another buffer by the transfer data copy thread, so that it is not affected.

  FIG. 3 is a diagram showing a configuration example of the third aspect of the invention, and shows a configuration on the hot standby side. The same components as those in FIG. 10 are denoted by the same reference numerals. In the figure, reference numeral 10 denotes a server on the hot standby side, which includes a CPU 4. As this CPU 4, CPU1 and CPU2 are provided. The data reception thread is assigned to CPU 1 and the copy data rearrangement thread is assigned to CPU 2. The operation of the system configured as described above will be described as follows.

  In the server that performs the call processing service in the active / hot standby configuration, the data is received from the call processing process in the hot standby state in the hot standby system that has received the data from the active system of the first invention and the second invention. The thread and the copy data rearrangement thread are activated, and at least the data reception thread is allocated to a CPU different from the rearrangement thread.

  By doing this, the copy data of the memory transferred from the active system is temporarily copied to the buffer, and after queuing, it is rearranged in the actual data area. If the data being received cannot receive the original data length, the data being received is discarded to prevent rearrangement of abnormal / insufficient data.

According to this embodiment, incompleteness of received data can be eliminated on the hot standby side.
FIG. 4 is a diagram showing a configuration example of the fourth invention, showing the operation of the active server. The same components as those in FIG. 10 are denoted by the same reference numerals. In the figure, 10 is an active server, and 6 is a CPU provided in the active server 10. In the figure, an example in which three CPUs are provided is shown. Here, CPU1 is a data transfer thread, and CPU2 is a call processing thread. CPUi is a CPU with a low usage rate. The operation of the system configured as described above will be described as follows.

  In the second invention described above, the CPU to which a transfer data copy thread for copying transfer data is assigned is not assigned to a fixed CPU, but the CPU that is the most CPU of any one except for the CPU assigned to the data transfer thread. The call processing thread searches for a CPU with a low usage rate and assigns it to that CPU (here, CPUi). In this way, CPU resources can be effectively used without being exclusively used for copy processing.

  According to this embodiment, the allocation CPU of the data copy process according to claim 2 is not fixed, but is allocated to an empty CPU, and the copy process 1 CPU whose processing time is shorter than that of data transfer is not occupied. By utilizing the idle time, the CPU can be used effectively and the call processing performance can be improved.

  FIG. 5 is a diagram showing an operation flow of the first invention. The same components as those in FIG. 10 are denoted by the same reference numerals. First, the call processing thread side operation in the active server 10 will be described. When the thread is assigned to the CPU and the call-related data is rewritten and the memory is transferred to the hot standby system (S1), an address list of the data to be transferred is created (S2). Here, the address list is a list in which a plurality of sets of functional unit IDs, addresses, and data lengths (lengths) are combined. Next, the data transfer thread queue (transfer queue) 20 is instructed to transfer (S3). Thereafter, the call processing is continued (S4).

  Next, the operation on the data transfer thread side will be described. The data transfer side thread periodically takes out the address list from the transfer queue 20 (S5). Then, the server name of the transfer destination and the file version number are added to the head of the extracted address list and sent to the hot standby system (S6). Next, the data length from the designated address is sent to the hot standby system from the top of the address list (S7). Here, when data is transferred, the data is read from the memory and transferred. Next, it is checked whether or not the predetermined number of turns has been made (S8). Here, “turned by the number” means the number of address lists waiting in turn in the transfer queue 20. If it has been rotated, the process returns to step S5 to perform an address list extraction operation from the queue. If not, the process returns to step S7 to send data.

  Next, the operation of the hot standby server will be described. The transfer destination server name, the file version number, and the data are transmitted to the hot standby system via the LAN 15 (see FIG. 10) and are queued in the reception queue 21. The data receiving thread periodically retrieves data from the queue (S9) and checks whether the head of the list is found (S10). If not found, the process returns to step S9. If found, it is checked whether the transfer destination server name matches the file version (S11). If they do not match, the process returns to step S9, and the data is extracted from the queue.

  If they match, the data length of the address list is copied to the relocation address (S12), and the address list for the number is taken out (S13). Next, it is checked whether or not the predetermined number has been rotated (S14). If not, the process returns to step S12 to take out the address list, and if it has been rotated, the process returns to step S9.

  FIG. 7 is a diagram showing an operation flow on the active server side of the second invention. The same components as those in FIG. 10 are denoted by the same reference numerals. First, the operation on the call processing thread side will be described. When the thread is assigned to the CPU and the call-related data is rewritten and the memory is transferred to the hot standby system (S1), an address list of the data to be transferred is created (S2). Here, the address list is a list in which a plurality of sets of functional unit IDs, addresses, and data lengths are combined. Next, the transfer data copy thread is instructed to copy (S3). Thereafter, the call processing is interrupted and sleeps (S4).

  Next, the operation on the transfer data copy thread side will be described. When a copy instruction is received from the call processing thread side, an address list is taken out in response to the instruction (S5). Next, the data length from the designated address is copied to the memory transfer buffer from the top of the address list (S6). Next, it is checked whether or not the number has been turned (S7). If not, the process returns to step S6 to perform the copy operation.

  If the number of rotations has been reached, the memory transfer buffer is queued in the transfer queue 20 (S8). Then, the end of the copy process is notified to the call processing thread side (S9). Upon receiving this notification, the call processing thread continues the call processing (S10).

  Next, the data transfer thread side operation will be described. The data transfer thread periodically fetches the memory transfer buffer from the transfer queue 20 (S11), adds the server name of the transfer destination and the file version number to the head of the fetched memory transfer buffer, and sends them to the hot standby system. (S12). The operation on the hot standby system side is the same as that of the first invention of FIG.

  FIG. 8 is a diagram showing an operation flow on the hot standby server side according to the third aspect of the invention. 5 and 10 are denoted by the same reference numerals. First, the operation on the data reception thread side will be described. When data is sent from the active server via the LAN 15, the data is accumulated in the reception queue 21. The data reception thread periodically retrieves data from the reception queue 21 (S1).

  Next, it is checked whether the head of the list has been found (S2). If the head of the list cannot be found, the process returns to step S1 to retrieve data from the queue. When the head of the list is found, it is checked whether the file version is correct (S3). If not, the process returns to step S1. If the file versions are correct, the address list corresponding to the number waiting in the reception queue 21 is taken out (S4). Then, the data length of the address list is queued in the rearrangement buffer (S6). As a result, the data length of the address list is queued in the relocation queue 22. The data reception thread side returns to the beginning (S7).

  Next, the operation on the data rearrangement thread side will be described. Periodically, data is taken out from the relocation queue 22 (S9), and the data length of the address list is copied to the relocation address (S9). Then, it is checked whether or not the number of waiting in the rearrangement queue 22 has been reached (S10). If not, the process returns to step S9 to perform the copy operation. If it is rotating, the process returns to step S8, and data is extracted from the rearrangement queue 22.

  FIG. 9 is a diagram showing an operation flow on the active server side of the fourth invention. The same components as those in FIGS. 5 and 10 are denoted by the same reference numerals. First, the call processing thread side operation will be described. It is assumed that when the thread is assigned to the CPU, the call-related data is rewritten and the opportunity to transfer the memory to the hot standby system is reached (S1). At this time, an address list of data to be transferred is created. Here, the address list is a list in which a function unit ID, an address, and a set of data length are combined.

  Next, a CPU having a low CPU usage rate is searched, and a transfer data copy thread is assigned to the CPU (S3). Next, the transfer data copy thread is instructed to copy (S4). Thereafter, the call processing is interrupted and a sleep state is entered (S5).

  Next, the operation on the transfer data copy thread side will be described. Upon receiving a copy instruction from the call processing thread, the address list is taken out in response to this instruction (S6), and the data length from the designated address is copied to the memory transfer buffer from the head of the address list (S7). . Then, it is checked whether or not the number of address lists has been rotated by a predetermined number (S8).

  If the predetermined number of times has not been turned, the process returns to step S7 to perform a copy operation. If the predetermined number of turns, the memory transfer buffer is queued in the transfer queue 20 (S9). Then, the end of the copy process is notified to the call processing thread side (S10). On the call processing thread side, call processing is continued (S11).

  Next, the operation on the data transfer thread side will be described. The memory transfer buffer is periodically taken out from the transfer queue 20 (S12), the transfer destination server name and the file version number are added to the head of the taken out memory transfer buffer and sent to the hot standby system (S13). The data is transmitted to the reception queue of the hot standby server via the LAN 15.

In the conventional technique, the function in charge for each CPU is not assigned, and the transfer process is embedded in the call process. This causes the operation of the call processing thread to be inhibited at each transfer timing. In the present invention,
Paying attention to the relationship between data transfer time between servers >> memory copy processing time> thread switching time, the effect of this transfer speed and copy processing time is separated from the call processing thread, and during I / O By making the call processing thread sleep and causing the other call processing threads to use the CPU, it is possible to reduce the occupied time by the I / O of the CPU and effectively use it.

  As described above, according to the present invention, it is possible to improve the call processing performance per hardware and reduce the cost by increasing the use efficiency of the CPU. The present invention is a very effective method for porting software on dedicated hardware such as an exchange configured with one existing CPU to a server (changing the exchange to a general-purpose server without changing as much as possible). It is.

  As described above in detail, according to the present invention, it is possible to provide an uninterrupted memory replication method that does not require interruption of a call processing service even during data transfer processing.

It is a figure which shows the structural example of 1st invention. It is a figure which shows the structural example of 2nd invention. It is a figure which shows the structural example of 3rd invention. It is a figure which shows the structural example of 4th invention. It is a figure which shows the operation | movement flow of 1st invention. It is a figure which shows a transfer destination address list. It is a figure which shows the operation | movement flow by the side of the active system server of 2nd invention. It is a figure which shows the operation | movement flow by the side of the hot standby type | system | group server of 3rd invention. It is a figure which shows the operation | movement flow by the side of the active system server of 4th invention. It is a figure which shows the system configuration | structure used as the object of this invention. It is explanatory drawing of the memory area distributed on a server. It is a figure which shows the boot flow of a call processing server. It is a figure which shows switch overflow of a call processing server. It is explanatory drawing of the conventional problem.

Explanation of symbols

4 CPU
6 CPU
10 servers

Claims (4)

In a server that performs a call processing service in an active / hot standby configuration, in the active system for memory replication, a call processing process is divided into a thread that performs a call processing service and a thread that performs memory transfer.
Assign each one to a different CPU,
Transfer memory copy data asynchronously with the commit timing of the memory contents of the call processing service to the hot standby system without interrupting all call processing services including calls that perform memory transfer and perform memory transfer.
In the hot standby system, the data reception thread started from the call processing process in the hot standby state is started, and the memory copy data transferred from the active server is relocated to the actual data area.
An uninterrupted memory replication method characterized by the above. In a server that performs a call processing service in an active / hot standby configuration, for the purpose of memory replication, in an active system, a call processing process is divided into a thread that performs a call processing service, a thread that performs memory copying, and a thread that performs memory transfer. ,
By assigning each to a different CPU, the memory contents are committed, and the call processing services except for calls that perform memory transfer are not interrupted, and the hot standby system is synchronized with the memory processing commit timing of the call processing services. Transfer a copy of the memory
In the hot standby system, the data reception thread activated from the call processing process in the hot standby state is activated, and the copy data of the memory transferred from the active system is relocated to the actual data area.
An uninterrupted memory replication method characterized by the above. In the server that performs the call processing service in the active / hot standby configuration, the data reception thread and the copy data relocation thread are activated from the call processing process in the hot standby state on the hot standby server that has received the data from the active server. ,
By allocating at least the data reception thread to a CPU different from the rearrangement thread, the memory copy data transferred from the active system is temporarily copied to the buffer, queued, and then rearranged in the actual data area.
The non-disruptive memory replication method according to claim 1 or 2, characterized in that it is configured as described above.   When starting a thread for copying the memory, instead of allocating to a fixed CPU, a CPU with the lowest CPU usage rate is searched for from any one of the CPUs other than the CPU allocated to the data transfer thread. 3. The non-disruptive memory replication method according to claim 2, wherein the allocation is made to a CPU.

Images