JPH04340155A - Fault recovery device for shared link structure in computer system - Google Patents

Abstract

PURPOSE:To restrict influence to be exerted upon the whole system to its minimum by removing a fault generated when a processor runs down in a multiprocessor system having a shared link in a shared memory. CONSTITUTION:A connection changing procedure for forward links 20 to 23 and back-ward links 30 to 33 is previously regulated as moving operation for acquiring and returning elements 11 to 13 constituting the link structure in the shared memory. The operation of the processors is monitored during the operation of the system, and at the time of detecting processor down, the stored connection changing procedure is compared with the status of the link structure at that time, a section in which moving operation is executed by the down processor is specified and link connection among the elements in the section of the moving operation is changed to repair the line structure.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a failure recovery device for a shared link structure in a computer system having a plurality of processors and a memory shared by the processors having a shared link structure.

0002

[Prior Art] Generally, in data processing in computer systems, especially operating systems, memory is divided into small elements, and a link structure is created by sequentially inserting a pointer to the next element in each element, and ordering is achieved. Often involves managing certain data. In addition, in order to quickly delete each element from the middle of an arbitrary link structure, in addition to the pointer to the next element (forward link), the pointer to the immediately previous element (backward link) is placed inside the element, so that it can be used in both directions. It often has a link structure. The structure of this bidirectional link is shown in FIG.

In FIG. 6, elements 11 to 13 are the front link 2.
0 to 23 and rear links 30 to 33, they are sequentially connected to the terminal 10, which is the starting point of the link structure. Note that the terminal 10 and the elements 11 to 13 may be collectively referred to as nodes 10 to 13. Element 11 is the first element of the link structure, and element 13 is the last element of the link structure. Forward links 21 and 22 are the first element 11
The rear links 33 and 32 point to the previous element in order from the last element 13, but the last front link 23 and the last rear link 31 have a circular structure pointing to the terminal 10, so-called double-sided links. It is a circular link. This bidirectional circulation link structure has the following characteristics in a stable state where no link operation is performed.

Features of links 1: The forward link of a specific node points to the backward link of a node that points to the specific node itself. Link characteristic 2: The forward link of the node pointed to by the backward link of a specific node points to the specific node itself. By the way, conventionally, in these multiprocessor systems, when considering a control structure using the above-mentioned bidirectional link structure on the shared memory, normally, during the link operation of each processor, the link operation is performed while performing some kind of mutual exclusive control. We are trying to prevent conflicts from occurring.

[0005]

[Problem to be Solved by the Invention] However, in conventional multiprocessor systems, if any processor goes down in a transient state during a link operation, other processors cannot use that link again due to exclusive control. state. Even if the mutual monitoring mechanism for processors down is used to release the exclusive control performed by the downed processor, the link structure itself is in a transient state, so if other processors continue to perform normal processing, unexpected results will occur. It turns out.

Furthermore, even if a mechanism is provided to release the exclusive state when a processor goes down, the processing load on the system increases under normal conditions due to the effects of the processor going down. In order to prevent such an increase in load under normal conditions, conventional systems do not take any special measures, resulting in the following problems.

(1) If a processor that is operating a link structure goes down, the impact is significant, and the target that was controlled by the link structure that was being operated by the processor that went down cannot be properly controlled by other processors. becomes. (2) Even if the system is made operational by restarting only the processor that has gone down while the system remains in an operating state, the effectiveness of the restart will be reduced because the failure caused by the downtime will continue after the system is restarted.

The present invention has been made to solve the above problems, and its purpose is to enable other processors to repair the link structure even if the processor operating the link structure goes down. Therefore, it is an object of the present invention to provide a failure recovery device for a shared link structure in a computer system, which can minimize the influence of processor failure and improve reliability.

[0009]

[Means for Solving the Problems] In order to achieve the above object, a first invention provides a procedure for changing connections of forward links and backward links in a movement operation for acquiring and returning elements constituting a link structure on a shared memory. means for defining and storing a shared memory having a shared link structure; means for monitoring operations of a plurality of processors accessing a shared memory having a shared link structure; and means for detecting down of a processor; A method for identifying a section in which a movement operation is being performed by comparing the connection change procedure currently being performed with the state of the link structure at that time; It is characterized by comprising a means for repairing.

[0010] The second invention is such that the procedure for changing the connections of forward links and backward links in the movement operation of acquiring and returning elements constituting a link structure on a shared memory is such that all elements to be moved are always connected to some element. means for defining and storing operation procedures for changing connections in a linked state; means for monitoring operations of a plurality of processors accessing a shared memory having a shared link structure; and detecting failure of a processor; When down is detected,
A means for identifying a section during a movement operation by comparing a stored connection change procedure with the state of the link structure at that time, and changing link connections between elements within the specified section during the movement operation. The present invention is characterized by comprising a means for repairing the link structure and inserting the element whose connection was being changed into any link structure depending on its characteristics.

[0011]

[Operation] In the first invention, a procedure for changing connections between forward links and backward links is defined and stored in advance as a movement operation for acquiring and returning elements constituting a link structure on a shared memory. During operation of the system, the operation of multiple processors accessing a shared memory having a shared link structure is monitored to detect a down processor. When a processor down is detected, the stored connection change procedure is compared with the state of the link structure at that time to identify the section that was being moved by the down processor, and the section that was being moved is The link connections between elements within are changed to repair the link structure.

[0012] In the second invention, the procedure for changing the connection of forward links and backward links in the movement operation of acquiring and returning elements constituting a link structure on the shared memory always connects all the elements to be moved to one of the elements. It is defined and stored as an operation procedure for changing the connection in a state where the link connection is made with. During operation of the system, the operation of multiple processors accessing a shared memory having a shared link structure is monitored to detect a down processor. When a processor down is detected, the stored connection change procedure is compared with the state of the link structure at that time to identify the section that was being moved by the down processor, and the section that was being moved is The link structure is repaired by changing the link connections between the elements within the link structure, and the element whose connection was being changed is inserted into one of the link structures depending on its characteristics.

[0013]

Embodiments Hereinafter, embodiments of the present invention will be explained with reference to the drawings. FIG. 1 is a diagram showing the operation procedure of a link structure in a first embodiment of the first invention. In the figure,
Operations from state 1 to state 5 are linked to element 11 from the link structure.
, 12 is shown below. In state 1, the link structure is stable, and in states 2 to 4, link connections are sequentially changed, so they are transient states. The link structure becomes stable again in state 5 when the operation is completed. The specific operating procedure is specified as follows.

State 1 → 2: The first element to be deleted (
The backward link 31 of element 11) is used as a pointer to the last element to be deleted (element 12). State 2 → 3: Backward link 33 of the next node (element 13) of the last element to be deleted (element 12) to the node (terminal 10) immediately before the first element to be deleted (element 11) be a pointer to State 3 → 4: The forward link 22 of the last element to be deleted (element 12) is replaced with the forward link 22 of the last element to be deleted (element 1).
1). State 4 → 5: The forward link 20 of the node (terminal 10) immediately before the first element to be deleted (element 11) is
The next node (
Pointer to element 13).

Furthermore, when inserting an element into the link structure, the procedure from state 5 to state 1 is performed in reverse order to these deletion procedures. By defining the operation procedure in this way, if the processor operating the link structure goes down, it will be limited to any of states 2 to 4 in a transient state, and will have the following characteristics. Note that "movement" in the characteristics is a general term for deletion and insertion.

Characteristic 1: The leading element being moved is specified by the forward link of the node immediately before the section being moved in the link structure. Characteristic 2: The backward link of the first element being moved points to the last element being moved. Characteristic 3: The backward link of the node immediately after the moving section in the link structure points to the last moving element or terminal.

From these characteristics, the following overall characteristics can be obtained. Comprehensive characteristic 1: Check "link characteristic 1" in the order of forward links starting from the terminal, and if there is a node that violates the rule, that node is the node immediately before the moving head element, that is, the node is A forward link points to the first element being moved. Also, the backward link of the first element being moved points to the last element being moved. Furthermore, examine each element in the order of backward links starting from the terminal,
A node whose backward link content points to a terminal or the last element being moved is the node immediately following the last element being moved.

In this embodiment, the link structure is repaired using this comprehensive characteristic 1, and the details of the process are shown in the flowchart of FIG. The symbols in the figure are defined as follows. p~s: Address of each node p→flink: Contents p of the forward link of the node indicated by p
→blink: Contents of the backward link of the node indicated by p A=
B: Assignment of B to A A==B: Matching judgment of A and B (YES when matching, NO when mismatching)

In the figure, processing is started when a processor down is detected, and "Link Feature 1" is applied to each node in order from the first terminal to its forward link.
is checked to see if there is a node that violates the rule (step 301).
~304). If there is no violation (step 302 YES)
, the link structure is normal, that is, state 1 or state 5 in FIG. 1, and the process ends (step 305
). If there is a violation (step 302 NO), it means that the nodes following that node are in the transient state to be separated or inserted. That is, the link structure corresponds to any of states 2 to 4 in FIG.

Next, the addresses of nodes p, q, and s are newly set (steps 306 to 308), and the nodes immediately after the last element to be separated or inserted are examined in the order of backward links (steps 309 to 308). 311), identify the nodes before and after the element to be separated or inserted. Next, based on the address of the identified element, the link structure in any of states 2 to 4 in FIG. 1 is restored to state 5, and the process is completed (steps 312 to 315). Note that the nodes p to s in the figure become as follows when link repair is completed.

p...Address of the first node of the separated moving section q...Address r of the last node of the separated moving section
...The address s of the node that was located immediately before the node indicated by p before separating the section being moved...The address s of the node located immediately before the node indicated by q before separating the section being moved node address

[0022] The elements separated by these processes can be formally inserted again into an appropriate link structure depending on the nature of the link structure. For example, if the link structure that performed the repair operation was one that manages free memory blocks, the element that was being moved from the link structure will be moved from the memory block that is no longer needed by the downed processor to the free memory block. It can be assumed that it is either something that was intended to be returned (inserted into the link structure) or something that was intended to be acquired (removed from the link structure) for future use. If it is preferable to return the separated elements to the original link structure, they are inserted into the link structure again (link reconfiguration).

In this way, in the first embodiment, by defining the operation procedure for the link structure of the nodes 10 to 13 shown in FIG. 1, even if the processor goes down during the operation of the link structure, The process in FIG. 2 specifies which state the moving elements 11 and 12 are in.
By separating the elements, it becomes possible to reconfigure the separated elements into a healthy link structure. In this way, the impact on the entire system due to processor failure can be reduced. Note that the operational procedure defined in the embodiment shown in FIG. 1 is just an example, and the order of operations can be defined differently from the embodiment. Furthermore, although this operating procedure has been shown for the case of separating two elements, it can be similarly applied to elements other than two.

Furthermore, in the embodiment, when a processor down is detected, the operating section is detected and the elements of the operating section are separated from states 2 to 4 to state 5. However, depending on the system conditions, etc. It may be preferable to return to state 1 from states 2 to 4. In that case, the process in Figure 2 includes a process to determine whether to return to state 1 or state 5, and state 1.
Add processing to return to.

Next, a second embodiment of the second invention will be explained with reference to FIGS. 3, 4 and 5. Figure 3
, 4 shows the operation procedure of the link structure. This illustrated example is the first
The difference from the above embodiment is that the moving elements 11 and 12 are not isolated and are always connected to one of the terminals. In state 1, there is a terminal 10 and a destination terminal 50, and elements 11 to 13 are sequentially connected to the terminal 10 by a front link and a rear link to form a bidirectional circulation link. Note that no element is connected to the terminal 50 at first. Status 1-4
The operations of nodes 10-13 in are the same as in the first embodiment shown in FIG.

After state 5, moving elements 11 and 12
is separated from terminal 10 and terminal 50
In the final state 10, the terminal 50 and the elements 11 and 12 are sequentially connected by the front link and the rear link to form a bidirectional circulation link. When the elements 11 and 12 are separated from the terminal 50 and inserted into the original terminal 10, the procedure from state 10 to state 1 is performed in reverse order.

Here, the characteristics starting from the terminals 10 and 50 inherit the "general characteristic 1" in the first embodiment, and furthermore, the operation procedure of the link structure is changed between the link structures shown in FIGS. 3 and 4. A unified method for moving elements is defined. This gives the following overall characteristics. Comprehensive property 2: All elements, including those in motion,
After repairing the same procedure as in the example, the element is successfully included in any link structure of all terminals to which it may be connected, or is an orphan link that is not connected to any terminal. included in the structure.

According to this characteristic, the necessary number of elements are acquired from a link structure connecting empty elements, necessary information is written therein, and the process is performed by inserting the necessary information into the link structure that forms the processing request queue. In cases where you are requesting a For example, if you prepare a temporary terminal for each processor, even if a processor goes down, you can perform the same process as shown in Figure 2 to properly link all the elements that are being processed. Can be incorporated into structures.

FIG. 5 specifically shows these processes, and shows the terminal 110 and elements 111 to 114.
is a link structure connecting empty elements, terminal 120 and elements 121 and 122 are temporary terminals and elements connected to them, and terminal 130 is a link structure connecting empty elements.
and elements 131 and 132 are processing request queues.

Here, when a specific processor issues a processing request, the necessary number of elements are first moved from the link structure including the terminal 110 to the link structure including the terminal 120 according to the procedure shown in FIGS. Writes the contents of the processing request into the moved element within the link structure containing the . Thereafter, the element to which the processing request has been written is again moved to the link structure including the terminal 130 in the procedure shown in FIGS. 3 and 4. The processed elements in the link structure of the terminal 130 are moved as free elements to the link structure including the terminal 110 in the procedure shown in FIGS. 3 and 4.

In this processing example, all elements are moved between link structures according to the procedure shown in FIGS. 3 and 4, so even if the processor that is moving the elements goes down at any stage, the By performing processing similar to the processing shown in 2, the link can be restored to a normal state. The recovery procedure is performed as follows.

(1) First, the same process as shown in FIG. 2 is performed on the terminal 110 to which empty elements are connected, and if there is a separated element, it is added to the link structure of the terminal 110 as an empty element. insert. Note that the separated element here is used as an empty element in the terminal 120.
is in the process of being moved to the link structure of Terminal 130 and has not yet been filled with content.
Since the processing has already been completed in the link structure of , there is no real harm even if it is treated as an empty element.

(2) Next, a process similar to that shown in FIG. 2 is performed on the terminal 120 to insert the separated element, if any, into the link structure of the terminal 110 as an empty element. Since the separated elements have already been removed from the link structure of terminal 110 to the link structure of terminal 120 in the operation (1), they are transferred from the link structure of terminal 120 to the link structure of terminal 1.
There is generally no real harm if there are only 30 links in transit and are not processed due to requests from a down processor. In addition, from Terminal 120 to Terminal 1
30, the contents of the request have already been written and may be inserted into the terminal 130.

(3) When the operations (1) and (2) above are completed, all elements moving between link structures have been removed. The element that a downed processor that depends on the terminal 120 was trying to write can be written by putting the processor identification number in the element when moving from the terminal 110 to the terminal 120, or by preparing a terminal 120 for each processor. It can be returned to the link structure of terminal 110. Additionally, requests for downed processors remaining in terminal 130 can be removed in a similar manner.

In this way, in the second embodiment, FIG.
By prescribing the operation procedure for the link structure consisting of the nodes 10 to 13 and 50 shown in FIG. 4, if the processor goes down during the operation of the link structure, the elements being operated This makes it possible to reconfigure the link structure without isolating any element, reducing the impact on the entire system due to processor failure. can be kept to a minimum.

[0036]

[Effect of the invention] As described above, according to the first invention,
When a processor is detected to be down, the pre-stored connection change procedure is compared with the state of the link structure at that time, the section that was being moved by the downed processor is identified, and the By changing the link connections between elements within the section and repairing the link structure, the fault caused by the link structure that was under the control of the down processor is removed, and healthy operation of the system is restored.

According to the second invention, when a processor down is detected, a pre-stored connection change procedure is compared with the state of the link structure at that time, and it is determined whether a move operation has been performed by the down processor. The section where the
The link connections between elements within the section being moved are changed, the link structure is repaired, and the element whose connection was being changed is inserted into one of the link structures depending on its characteristics, thereby improving the system. The healthy operation of the down processor is restored, and the elements of the link structure that were under the control of the down processor are fully connected to one of the link structures and restored, minimizing the impact of the down processor on the system. can.

[Brief explanation of drawings]

FIG. 1 is an explanatory diagram of an operating procedure in a first embodiment.

FIG. 2 is a flowchart showing processing of the first embodiment.

FIG. 3 is an explanatory diagram of the operating procedure in the second embodiment.

FIG. 4 is an explanatory diagram of the operating procedure in the second embodiment.

FIG. 5 shows an example of a link structure in a system to which the second embodiment is applied.

FIG. 6 is an explanatory diagram conceptually showing the structure of a bidirectional circulation link.

[Explanation of symbols]

10 Terminal (node) 11-13 Elements (nodes) 20-23 Forward link 30-33 Back link 50 Terminal (node) 110 Terminal (node) 111-114 Elements (nodes) 120 Terminal (node) 121, 122 Element (node) 130 Terminal (node) 131, 132 Element (node)

Claims (2)

[Claims] 1. A means for defining and storing a connection change procedure for forward links and backward links in a movement operation for acquiring and returning elements constituting a link structure on a shared memory, and a shared memory having a shared link structure. means for monitoring operations of a plurality of accessing processors and detecting down of a processor, and when down of a processor is detected, comparing a stored connection change procedure with a state of a link structure at that time; means for identifying a section during a movement operation; and means for repairing a link structure by changing link connections between elements within the specified section during a movement operation;
A failure recovery device for a shared link structure in a computer system, characterized by comprising: [Claim 2] The procedure for changing the connection of forward links and backward links in a movement operation that acquires and returns elements constituting a link structure on a shared memory is carried out in a state in which all elements to be moved are always linked and connected to any element. means for defining and storing operation procedures for changing connections in a shared link structure; means for monitoring operations of multiple processors accessing a shared memory having a shared link structure; means for identifying the section during the movement operation by comparing the stored connection change procedure with the state of the link structure at that time, and link connections between elements within the specified section during the movement operation. of a shared link structure in a computer system, comprising means for repairing a link structure by changing the link structure, and inserting an element whose connection was being changed into any link structure according to its characteristics. Disaster recovery equipment.