System and method for establishing consistent memory contents in redundant systems

Abstract

In a fault-tolerant system which is constructed from two identical control devices, there is a requirement to establish consistent memory contents in both devices in order to guarantee uninterrupted operation in the event of a hardware defect in one of the two control devices. To this end, the memory contents are transferred from an active control device to an inactive control device. In this situation, the active control device remains in operation and the contents of the memory of the active control device can be continually updated. According to the invention, the transfer of the memory contents is performed by a copying device, whereby the copying device manages the memory areas of the memory of the active control device which are to be transferred and which are possibly changing by means of a memory monitoring module and a marking memory.

Description

CLAIM FOR PRIORITY

[0001] This application claims the benefit of priority from German Application No. 01120254.6 filed Aug. 23, 2001.

TECHNICAL FIELD OF THE INVENTION

[0002] The invention relates to a system and method for establishing consistent memory contents in redundant systems.

BACKGROUND OF THE INVENTION

[0003] In a fault-tolerant system which is constructed from two identical control devices, there is a requirement to establish consistent, i.e. identical, memory contents in both devices in order to guarantee uninterrupted operation in the event of a hardware defect in one of the two control devices. To this end, the memory contents must be transferred from the active control device to the inactive control device. In this situation, the active control device remains in operation and the contents of the memory of the active control device can be continually updated.

[0004] Until now, a hardware-based method or a software-based method has been used as a solution to this problem. With regard to the hardware-based method, all the data which is written to the memory of the active control device is transferred by way of an interface from the active control device to the inactive control device. To achieve this, it is necessary to sample the data externally on the memory interface of the active control device or internally in the north bridge of the active control device and forward it to the interface with the inactive control device.

[0005] Sampling and forwarding must either be able to occur quickly such that no data is lost, i.e. the bandwidth of the interface with the inactive control device must be at least as high as that of the memory interface, or there must be a capability to reduce the speed at which new data is written to the memory of the active control device, the speed of the CPU therefore, such that no loss of data results.

[0006] The following problems are encountered with the known hardware-based method:

[0007] The processing speed of the CPU is reduced as a result of the slowing down of the write data rate of the CPU.

[0008] In order to slow down the CPU, control of the CPU from the hardware side and thus access to the CPU bus are required. Access to the CPU bus may be undesirable or even impossible if, for example, the functionality of the north bridge is integrated in the CPU—a situation which will be encountered more frequently in future.

[0009] If the sampling of the write data takes place in the north bridge, as a rule, no conventional north bridge can be used. The development effort for a special north bridge which is then required is considerable.

[0010] With regard to the software-based method, memory area tables are used for the CPU, and in that situation particularly the dirty bit which indicates whether write access has been made to a memory area. A task which is running in the background periodically checks the entries in the memory area tables and initiates the copying of memory areas to which write access has been made, i.e. their associated dirty bit is set.

[0011] The following problems are encountered with the known software-based method:

[0012] The background task which evaluates the entries in the memory area tables consumes CPU power.

[0013] It is necessary to ensure that the dirty bits can be utilized. Modem operating systems use the memory management unit of the CPU, which is responsible for management of the memory with the assistance of the memory area tables, and frequently manipulate the memory area tables during the process. The software-based method functions if the operating system does not use the memory management unit of the CPU, or is designed such that the method is supported directly by the operating system.

SUMMARY OF THE INVENTION

[0014] In one embodiment of the invention, consistent memory contents are established and carried out with the aid of simple devices such as a memory monitoring module, a copying device, and a marking memo, and is controlled by the copying device. No specially produced north bridge is therefore required for sampling the memory contents. Additionally, it is not necessary to slow down the write data rate and thus the processing speed of the CPU since the data to be transferred is not sampled directly by the CPU but is read from the memory. The link which is used for transferring the memory contents to the inactive control device can have a smaller bandwidth than the memory interface between CPU and memory. The method works independently of the dirty bits in the memory area tables of the CPU—MMU, Memory Management Unit—whereby new dirty bits are formed in the marking memory. This is advantageous because the use of operating systems is not subject to any restrictions regarding memory management and that no adaptation of the operating system is required. Furthermore, control of the method being effected by the copying device ensures that no CPU power is consumed for establishing consistent memory contents.

[0015] In one aspect of the invention, information concerning memory areas which are to be omitted during the transfer to the inactive system is additionally stored in the marking memory in which the dirty bits that indicate a write access to the memory area in question are stored. The advantage in this situation is that memory areas which are frequently modified do not include any required information which does not impede convergence of the method.

BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The invention will be described below, with reference to the drawings.

[0017] FIG. 1 shoes an active control unit and an inactive control unit of a fault-tolerant system.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0018] FIG. 1 shows an active control unit SEo and an inactive control unit SE1 of a fault-tolerant system. Both control units SEo and SE1 are of similarly constructed and each include a processing unit CPUo, CPU1, an interface unit or north bridge NBo, NB1, and a memory MEM0, MEM1. The functionality of the processing units CPUo, CPU1, and of the north bridges NBo, NB1, can, as shown, be implemented in two separate devices, or combined in a single device, which is not shown.

[0019] In addition, for each of the two control devices SEo, SE, FIG. 1 shows a copying device Ko, K1, each having a connected marking memory, or dirty page tag RAM DPTRo, DPTR1 and a memory monitoring module, or snooper So, S1. In a first embodiment, this is used in order to synchronize the contents of the memory MEMo of the active control device SEo with the memory MEM1 of the inactive control device SE1 for the first time, for example following the replacement of a hardware unit belonging to the inactive control device SE1. After memory contents have been established, the inactive control device SE1 is also activated. Both control devices then operate in synchronism, whereby the synchronous operation of the memory contents of the two control devices is monitored.

[0020] The snooper So of the active control device SEo observes the accesses of the north bridge NBo of the active control device SEo to the memory MEMo of the active control device SEo. To this end, the snooper So of the active control device SEo is connected to the control lines of the memory interface SIo of the active control device SEo, which provides information concerning whether, for example, a write cycle, a read cycle or a refresh cycle is taking place. Write cycles are of significance to the snooper So of the active control device SEo. The snooper So of the active control device SEo is additionally connected to the address lines and control lines of the memory interface SIo of the active control device SEo such that it is possible to determine by means of the snooper So of the active control device SEo on which memory area a write access is taking place.

[0021] This information, i.e. the memory address to which a write access is being performed is transferred by the snooper So of the active control device SEo as address information AIo by way of an interface to a copying device Ko of the active control device SEo, which evaluates this information. With regard to this copying device Ko of the active control device SEo, this is preferably a field programmable gate array FPGA or an application specific integrated circuit ASIC. However, it is also possible to implement the function of the copying device Ko of the active control device SEo in a program-controlled fashion by using a micro-controller.

[0022] The copying device Ko of the active control device SEo monitors the memory areas managed by it of the memory MEMO of the active control device SEo for modifications and to read any modified contents and to send them by way of a link L to the copying device K1 of the inactive control device SE. The copying devices Ko, K1 are advantageously connected in one embodiment by a standard interface—e.g. PCI bus or AGP bus—to the north bridges NBo, NB1.

[0023] The copying device Ko of the active control device SEo is connected to a dirty page tag RAM DPTPo of the active control device SEo. An associated bit exists in the dirty page tag RAM DPTRo of the active control device SEo for each memory area of the memory MEMo of the active control device SEo. The bit assigned to a memory area in the dirty page tag RAM DPTRo of the active control device SEo is set by the copying device Ko of the active control device SEo if a write access has been performed to the corresponding memory area. This causes the corresponding memory area to be identified as modified “dirty”.

[0024] The number of memory areas into which the memory MEMO of the active control device SEo is divided and the size of the particular memory area which is managed by a respective bit in the dirty page tag RAM DPTRo of the active control device SEo do not necessarily play a significant role in this situation for the method according to the invention. The memory areas can be equal in size or can have different sizes.

[0025] When a bit has been set in the dirty page tag RAM DPTRo of the active control device SEo, it is not buffered by the snooper So of the active control device SEo. In other words, write accesses to an address or an address range is registered by the snooper So of the active control device SEo and signaled to the copying device Ko of the active control device SEo irrespective of whether the corresponding memory area has already been marked as modified by the corresponding bit set in the dirty page tag RAM DPTRo of the active control device SEo.

[0026] When the operation to carry out the first synchronization of the memory contents of the active control device SEo and the inactive control device SE1 is initiated, the copying device Ko of the active control device SEo utilizes the dirty page tag RAM DPTRo of the active control device SEo, e.g. beginning at the first bit of the dirty page tag RAM DPTR0 of the active control device SEo, to check whether the corresponding memory area has been modified. If this is the case, the corresponding bit in the dirty page tag RAM DPTRo of the active control device SEo is reset, and the copying device Ko of the active control device SEo reads the memory area, identified as modified by this bit, of the memory MEMO of the active control device SEo and transfers the memory contents by way of the link L to the copying device K1 of the inactive control device SE1.

[0027] The copying device K1 of the inactive control device SE1 transfers the received data to the memory MEM1 of the inactive control device SE1. In this situation, the data is stored by the copying device K1 of the inactive control device SE1 at the address in the memory MEM1 of the inactive control device SE1 at which it is also stored in the memory MEMo of the active control device SEo. The copying operation takes place in the background concurrently with the activity of the processing unit CPUo such that memory areas can be modified again until the copying device Ko of the active control device SEo has processed the memory areas identified by the corresponding bits of the dirty page tag RAM DPTRo of the active control device SEo.

[0028] If write access is performed to memory areas which have already been transferred while the operation is running to effect the first synchronization of the memory contents, the associated bits are set again in the dirty page tag RAM DPTRo of the active control device SEo, as a result of which these memory areas are again identified as modified. If the memory area currently being copied is also affected by this, the copying operation in progress can be continued or aborted in this situation.

[0029] After the copying operation for a memory area has been successfully completed or has been aborted because of modifications occurring in the interim, the next bit in the dirty page tag RAM DPTRo of the active control device SEo is checked by the copying device Ko of the active control device SEo. If this is set, the relevant bit in the dirty page tag RAM DPTRo of the active control device SEo is reset and the corresponding memory area is likewise copied from the memory MEMo of the active control device SEo into the memory MEM1 of the inactive control device SE.

[0030] Checking of the dirty page tag RAM DPTRo of the active control device SEo is performed for each individual bit. On reaching the last bit in the dirty page tag RAM DPTRo of the active control device SEo, the operation is started again for the first bit by the copying device Ko of the active control device SEo.

[0031] A counter for the number of memory areas not yet copied or modified is present in the copying device Ko of the active control device SEo. When the count reaches or fails to reach a pre-defined value, the processing unit CPUo is briefly prevented from performing further write transactions in the memory MEMO of the active control device SEo, for example by an interrupt triggered by the copying device Ko, with a corresponding interrupt handling routine. During this time, the remaining modified memory areas of the active control device SEo are copied to the inactive control device SE by the copying device Ko.

[0032] The memory areas identified by corresponding bits in the dirty page tag RAM DPTRo of the active control device SEo are transferred from the active control device SEo to the inactive control device SE1. With that, the method according to the first embodiment of the present invention is completed, the control device SE1 can be activated and both control devices SEo and SE can continue to operate synchronously.

[0033] The convergence of the method, i.e. the speed at which the number of memory areas not yet copied reduces, depends heavily on the application software running on the processing unit CPUo of the active control device SEo, in particular on the locality and frequency of the write memory accesses. Generally speaking, write accesses made by an application have a restricted local effect. However, the nature of an application may be such that it modifies memory areas over extended periods of time or without interruption more quickly than these can be copied by copying device Ko of the active control device SEo. In this case, the application software is slowed down in order to force convergence. This can be done in an advantageous embodiment, for example, by an interrupt triggered by the copying device Ko of the active control device SEo with a corresponding interrupt handling routine, as a result of which the processing unit CPUo of the active control device SEo is increasingly slowed down further but without any intervention in the actual application software.

[0034] In an advantageous embodiment of the invention, the dirty page tag RAM DPFRo for each memory area can include a further bit which is set in order to indicate to the copying device Ko whether the memory area is to be synchronized. If a memory area is not to be synchronized, the associated bit in the dirty page tag RAM DPTRo is ignored by the copying device Ko. This is appropriate for memory areas where it is known that these are frequently modified but do not include any important or required information.

[0035] The invention is not restricted to the embodiment. For example, the method can likewise be used in order to continually add changes made in a memory MEMo of an active control device SEo to a memory MEM1 of an inactive control device SE1 with the objective that, in the event of failure of the active control device SEo, operation can be continued by the inactive control device SE1 with the relevant current memory contents.

[0036] To this end, the method described above is adapted to the effect that the checking of the dirty page tag RAM DPTRo of the active control device SEo is carried out continually and that there no interruption of the processing unit CPUo occurs if a particular counter value is not reached for the number of memory areas not yet copied or modified.

[0037] The dirty page tag RAM DPTRo of the active control device SEo can advantageously include a further bit for each memory area which is set in order to indicate to the copying device Ko of the active control device SEo whether the memory area is to be synchronized—not shown.

[0038] Advantageously, an interruption can be provided between two complete checks of the dirty page tag RAM DPTRo of the active control device SEo, for example in order to prevent excessive loading of the memory interface by the copying device Ko of the active control device SEo.

Claims

1. A method for establishing consistent memory contents in a redundant system, comprising: an active control unit and an inactive control unit, each having a processing unit with an interface unit and a memory, wherein the memory of the active control unit is monitored by a memory monitoring module of the active control unit, the memory having address information relating to memory areas to which a write access has been performed and which is forwarded by the memory monitoring module of the active control unit to a copying device of the active control unit, the address information supplied by the memory monitoring module of the active control unit is synchronized by the copying device of the active control unit with a marking memory of the active control unit, in which the memory areas to which a write access has been performed are entered, data from the memory areas entered in the marking memory of the active control unit is sent by the copying device of the active control unit by a link to the copying device of the inactive control unit, whereby corresponding bits are reset in the marking memory of the active control unit, and the data received by the copying device of the inactive control device by the link is stored at the address in the memory of the inactive control device at the same address stored in the memory of the active control device.

2. The method according to claim 1, wherein one or more bits are used in the memory area in the marking memory of the active control unit for marking the memory area as having been modified by a write access.

3. The method according to claim 2, wherein at least one additional bit is used per memory area in the marking memory of the active control unit for marking the memory area as omitted during transfer to the memory of the inactive control device.

4. The method according to claim 1, wherein the method is executed until the memory areas of the memory of the active control device which have been modified by a write access have been transferred to the memory of the inactive control device.

5. The method according to claim 4, further comprising: entering a number of remaining memory areas of the memory of the active control device which have been modified by a write access; and defining a limit value for the number such that when the number fails to reach the value, additional write accesses to the memory of the active control device are prevented until the remaining memory areas of the memory of the active control device which have been modified by the write access have been transferred to the memory of the inactive control device.

6. The method according to claim 1, further comprising providing an interruption between two complete checks of the marking memory of the active control unit and the transfer of data to the memory of the inactive control device.

7. The method according to claim 1, wherein at least one of a field programmable gate array, an application specific integrated circuit or a micro-controller is provided for each of the copying devices, and the connection of the copying devices to the interface unit including a memory interface or to the CPU with an integrated interface unit is implemented by a PCI bus system or by an AGP bus system.

8. A system for establishing consistent memory contents in a redundant system, comprising: an active control unit and an identical inactive control unit, each having a processing unit with an integrated or interface unit, a memory, and a memory monitoring module for monitoring the memory of the active control unit which forwards address information to a copying device of the active control unit, the copying device of the active control unit synchronizes the address information with a marking memory of the active control unit, the marking memory having memory areas to store information to which a write access has been performed, and which, by means of the marking memory of the active control unit, transfers marked memory areas of the memory of the active control device to a copying device of the inactive control unit and resets corresponding bits in the marking memory prior to transfer, and the copying device of the inactive control unit, which receives data from the copying device of the active control unit by way of a link, enters data in the memory of the inactive control device at the same address at which the data is stored in the memory of the active control device.

9. A memory monitoring module, comprising: a first device to monitor a memory interface of a memory; and a second device to provide address information regarding memory areas to which a write access has been performed.

10. A copying device of a redundant system, comprising: a first device to read data from a memory of a control device; a second device to write the data to the memory of the control device; a third device to receive address information about memory areas to which a write access has been performed; a fourth device to synchronize the address information with a marking memory having memory areas storing information to which a write access has been performed, by sending data from the memory areas of the memory which are entered in the marking memory and the address information of the data to a first copying device; a fifth device to receive the data and the address information of the data from a second copying device; and a sixth device to enter the data in the memory of the control device in accordance with the address information received.

11. The copying device according to claim 10, wherein the copying device is at least one of a field programmable gate array, an application specific integrated circuit or a micro-controller, and the copying device is connected by one of a PCI bus system or by an AGP bus system to the interface unit including the memory interface or to a CPU with an integrated interface unit.

12. The copying device according to claim 10, wherein the copying device includes a memory monitoring module with a first unit to monitor the memory interface of the memory and a second unit to provide address information about the memory areas to which a write access has been performed, and the marking memory in which the memory areas to which a write access has been performed are entered.