JP3433456B2 - RAS mechanism of cooling monitoring system for computer - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an RAS mechanism of a cooling monitoring system for an electronic computer. In recent years, fuzzy theory has been industrially used as a theory for obtaining deterministic output from ambiguous input, as is well known in “Fuzzy Theory and Its Applications” (by Masaharu Mizumoto). In the fuzzy theory, in short, a person having expertise in inputting a membership function is defined and held as a knowledge base, and an ambiguous input is determined using the knowledge base. Here, the use of fuzzy logic when there is an ambiguous limit instead of an ambiguous input can be considered as a means for solving this problem. 2. Description of the Related Art In recent large-scale computer systems, high performance has been pursued, and high-density semiconductor elements and the like have been adopted with the increase in the degree of integration of arithmetic circuits. The cooling system has become indispensable as a solution to this problem. [0004] In the conventional cooling system, when an abnormality of the cooling system is detected by various sensors, a warning is issued to a computer, or in the worst case, the entire system is stopped (hereinafter, referred to as a system down). A conventional cooling monitoring system will be described. FIG. 4 is an example of a conventional computer, and is a configuration diagram of a conventional cooling monitoring system. In FIG. 1, reference numeral 1 denotes an electronic computer, which is an apparatus composed of high-density semiconductor elements 2 to be cooled by cooling system equipment. Reference numeral 2 denotes a high-density semiconductor element which performs calculations and consumes power to generate heat. Reference numeral 3 denotes a cooling system, which is a customer cooling system 4
And a refrigerant supply device 5. Reference numeral 4 denotes a customer air conditioner, which supplies cooling water (hereinafter referred to as customer refrigerant) supplied by the customer. Reference numeral 5 denotes a refrigerant supply device, which is an electronic computer 1
This is a device for cooling pure water (hereinafter referred to as an internal refrigerant) for cooling the high-density semiconductor element 2 using a customer refrigerant.
Since the internal refrigerant is expensive, it forms a closed circulation system. Reference numeral 6 denotes a monitoring sensor, which is a customer cooling facility 4
It monitors the operation of the cooling system such as the refrigerant supply device 5 and the computer 1, and includes a flow rate sensor, a temperature sensor, a fan rotation sensor, and the like. Reference numeral 7 denotes an environment cooling control unit, which is a central processing unit (hereinafter abbreviated as CPU) 8, a memory 9
, And status monitoring unit 10, fault tree analysis (fault tree analysis: hereinafter abbreviated as FTA) unit 11, configuration control unit
12 and a configuration control table 13. Reference numeral 8 denotes a CPU for executing a program in the memory 9 to perform environmental cooling control. Reference numeral 9 denotes a memory for storing programs and data. Reference numeral 10 denotes a state monitoring unit which detects an abnormality in the input of the monitoring sensor 6 and activates the FTA unit 11. 11
Is a function of determining the cause of the failure from the input of the monitoring sensor 6 indicating the failure by using the FTA method for analyzing the failure. The program in the memory 9 is executed by the CPU.
This is achieved by executing in step 8. Reference numeral 12 denotes a configuration control unit, which has a function of updating the configuration control table 13 describing the configuration of the system and changing the system configuration. A configuration control table 13 describes the system configuration of the computer 1. Reference numeral 14 denotes a configuration control device for connecting the devices of the computer 1 according to the configuration control table 13 to change the system configuration. In the conventional cooling monitoring system, the input of the monitoring sensor 6 is monitored by the state monitoring unit 10, and if an abnormality occurs, the abnormality is detected and the FTA is detected.
Unit 11 had been started. The FTA unit 11 describes in advance the relationship between the abnormal phenomenon and the cause of the failure in the form of a fault tree, and upon receiving the notification of the occurrence of the abnormal phenomenon, finds the cause of the failure from the description of the fault tree. FIG. 5 is a diagram illustrating a fault tree. In the figure, the node
51 is an abnormal phenomenon called "temperature abnormality of internal refrigerant",
This is a phenomenon detected by the monitoring sensor 6. The fact that the cause of this phenomenon is either the “sensor failure” of the node 52 or the “pump failure” of the node 53 is indicated by the connection of the node 61 with the OR symbol. The OR symbol of the node 63 indicates that the cause of the "pump failure" of the node 53 is either the content of the OR symbol of the node 64 or the "power failure" of the node 55. node
The contents of the OR symbol 64 are either “pomb leak” at node 56, “low pump pressure” at node 57, or “low pump capability” at node 58. Decomposition of abnormal phenomena from the top down to the cause is used in failure analysis, and failure tree analysis (fault tree analysis:
FTA). For example, in a high-speed computer, FIG.
Although the cooling of the high-density semiconductor element 2 is an important issue, it is difficult to directly measure the cooling of the high-density semiconductor element 2, so that the internal refrigerant that cools the high-density semiconductor element 2 and the internal refrigerant Cooling control is performed indirectly based on the temperature and flow rate of the customer refrigerant to be cooled. FIG. 6 is a diagram showing an operable temperature range of the electronic computer determined by the temperatures of the customer refrigerant and the computer refrigerant. In the figure, reference numeral 32 denotes an actual cooling function permissible curve. If the temperatures of the customer refrigerant and the computer refrigerant fall within this range, the operation of the high-density semiconductor element 2, ie, the computer 1
Operation is possible. Numeral 31 denotes a hypothetical cooling function allowable curve, which means that if the temperatures of the customer refrigerant and the computer refrigerant fall within this range, the operation of the electronic computer 1 is determined to be possible and the cooling control is performed. However, the range of the hypothetical cooling function permissible curve is virtual, and in fact, the actual cooling function permissible curve 32
The limit having the width indicated by is the limit of operation. Within this width (the shaded area in the figure), the accuracy of the operation is determined stochastically. Normal operation may not be possible. In the conventional cooling control system, this monitoring is performed by monitoring the internal refrigerant supply temperature in the refrigerant supply device 5 and the customer refrigerant temperature in the customer cooling facility 4 to determine whether the temperature is within the range of the cooling function allowable curve. The cooling function was monitored based on the binary logic as to whether or not the operation was deviated. If an abnormality is detected, the system is shut down at a certain level or higher according to the measured value, and a warning is sent to a higher-level computer via a communication means at a certain level or lower as a warning. These processes are performed by the FTA unit, and when it is necessary to change the configuration of the computer system for operation, the configuration control unit 12 rewrites the configuration control table 13. Configuration control device not shown
According to 13, the connection of the computer system is changed to change the configuration, the faulty device is disconnected, and the operation of the system is continued. It has been proposed to use fuzzy inference for such an FTA (see "Fuzzy Fault Tree Analysis").
And its applications ”(1st Knowledge Engineering Symposium)). However, as described above, in the conventional cooling control, it is determined whether or not the temperature is within the range of the cooling function allowable curve based on binary logic.
Since a process such as a down was performed, the system was stopped even if it was in an operable environmental temperature range or an environmental temperature range in which countermeasures could be taken, and the RAS (Reliability: Reliability, availability: availability, serviceability: maintainability). In view of the above, the present invention provides a means for improving the RAS function of a computer by defining the operable environmental temperature range by fuzzy logic, thereby increasing the tolerance for determining system shutdown. The purpose is to provide. The above object is achieved by a cooling monitoring system having fuzzy control configured as described below. FIG. 1 is a diagram showing the principle of the present invention. Monitoring sensor that monitors the cooling system that cools the computer system
6, an input recording unit 20 for receiving and recording the output, a fuzzy state monitoring unit 21 for judging the input using fuzzy logic to detect an abnormality, and when detecting the abnormality, starting a fuzzy fault tree analysis unit 22; A fuzzy fault tree analysis unit 22, which performs a fault tree analysis based on the record recorded in the input recording unit 20 and the input and determines the cause of the fault using fuzzy logic, the fuzzy state monitoring unit 21, and the fuzzy fault tree analysis unit A knowledge base 23 for storing rules used for inference with the configuration 22; and a configuration control table 13 for defining the configuration of the computer system.
And a configuration control unit 12 that changes the configuration control table 13 based on the determined cause. The input recording unit 20 receives and records the output of the monitoring sensor 6 for monitoring the cooling system for cooling the computer system. The fuzzy state monitoring unit 21 determines the input by using fuzzy logic and detects an abnormality. At this time, inference is performed using the knowledge base 23 provided in advance. Therefore, it is possible to perform detailed monitoring using expert knowledge. When an abnormality is detected, the fuzzy fault tree analysis unit 22 is activated. The fuzzy fault tree analysis unit 22 analyzes the fault tree from the output of the monitoring sensor 6 for monitoring the cooling system and the record recorded in the input recording unit 20 which is the past history, and determines the cause of the fault by fuzzy logic. Use and judge. At this time, inference is performed using the knowledge base 23. Therefore, it is possible to make a determination using expert's knowledge. The knowledge base 23 includes the fuzzy state monitoring unit.
The rules used for inference by the fuzzy fault tree analysis unit 21 and the fuzzy fault tree analysis unit 22 are stored in advance. The configuration control table 13 defines the configuration of the computer system, and the configuration control unit in the computer system sets the configuration according to this table. The configuration control unit 12 includes the fuzzy fault tree analysis unit.
The configuration control table 13 is changed so as to avoid the cause of the failure determined in step 22. With the above configuration, the abnormality of the cooling system of the computer system is detected,
By determining the cause, the configuration of the computer system can be changed so as to avoid the cause of the failure. FIG. 2 is a block diagram of a cooling monitoring system according to an embodiment of the present invention. In the figure, reference numeral 20 denotes an input recording unit which records an output of a monitoring sensor 6 for monitoring a cooling system or the like as an input. 14 is a configuration control device,
The configuration of the computer 1 is changed with reference to the configuration control table 13. Reference numeral 21 denotes a fuzzy state monitoring unit which activates a fuzzy fault tree analysis unit 22 when the output of the monitoring sensor 6 is determined using fuzzy logic and an abnormality is detected and detected. 22 is a fuzzy fault tree analysis unit,
The fault tree is analyzed from the output of the monitoring sensor 6 for monitoring the cooling system and the past history recorded in the input recording unit 20 to determine the cause of the failure using fuzzy logic. Reference numeral 23 denotes a knowledge base which stores rules used for inference by the fuzzy state monitoring unit 21 and the fuzzy fault tree analysis unit 22. In addition, FIG.
Items having the same reference numerals as the same are the same. The operation of the embodiment of the present invention will be described below.
Monitoring the output of the monitoring sensor 6 for monitoring the cooling system as an input is the same as in the related art, but the input recording unit 20 records this input as a history. Further, the fuzzy state monitoring section 21 judges whether or not the input state is abnormal by fuzzy inference based on the knowledge and the past history based on the knowledge base 23. The fuzzy inference can be realized by software, and there is a microprocessor called a fuzzy chip having a structure particularly suitable for fuzzy inference when it is desired to execute at a higher speed, and can be realized by using such a chip. The reason why the temperature state is determined by fuzzy inference will be described. FIG. 3 is an explanatory diagram of a temperature range in which the computer can operate, and the cooling function allowable curve shown in FIG. 3 is the same as FIG. A point O shown in the figure is a state specified by the customer refrigerant temperature and the internal refrigerant temperature, is outside the range of the hypothetical cooling function allowable curve 31, and conventionally, when the state shown by this point is reached, the system is brought down. Had gone. However, this point is within the width (shaded area) of the frame of the actual cooling function permissible curve 32. For example, when the point O changes from the point A to the point B to the point O, it takes 5 minutes. The position may be operated, but if the rules are based on expert knowledge that if the operation changes from point C to point D to point O, the operation must be stopped immediately to stop the heat generation and the element will be destroyed. There is. In such a case, it is not necessary to take a process of bringing the system down immediately by checking the past history and depending on the point from which the point O is reached, and change the system configuration while there is sufficient time. If the countermeasures against abnormal phenomena are taken, the operation of the system may be continued. Such expert's knowledge is stored in the knowledge base 23 and used when detecting an abnormality from the input state. When the fuzzy state monitor 21 detects an abnormality from the input state, the fuzzy fault tree analyzer 22 is activated. The fuzzy fault tree analysis unit 22 uses the conventional FT
Like A, the cause of failure is determined from the input state and the past history according to the rules stored in the knowledge base 23. At this time, for example, in FIG. 6, when three causes are considered from the node 64, in the conventional FTA, one of the three causes is determined from the input state. However, in the fuzzy fault tree analysis, each cause is weighted to determine a cause (for example, the node 56
0.2, node 57 is 0.5, and node 58 is 0.3) Expert knowledge stored in the knowledge base 23 based on the cause having this weight (for example, "judgment by listening to the motor sound at that time")
), And a definitive cause determination is made by fuzzy inference. The determined cause is the configuration control unit 12
And the configuration control unit 12 rewrites the configuration control table 13 according to a predetermined rule in accordance with the notification to change the system configuration. Change of the system configuration is performed by the configuration controller 14
Changes the connection of the computer system according to the configuration control table 13 to change the configuration, and disconnects the faulty device. The system software of the computer 1 continues operating the system in the changed system configuration. The cause of the failure determined in the above system is based on the knowledge of an expert, and a more sophisticated judgment can be made for cooling monitoring than before so that the system can be prevented from being down, and the RAS function of the computer system can be avoided. Can be improved. As is clear from the above description, according to the present invention, the operating environment is described using fuzzy logic and the situation is determined, so that a flexible determining process can be performed unlike the conventional one. The system can be prevented from being stopped, and there is a remarkable industrial effect in improving the RAS of the high-speed computer.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a principle diagram of the present invention; FIG. 2 is a configuration diagram of a cooling monitoring system according to an embodiment of the present invention; FIG. 3 is an explanatory diagram of a temperature range in which an electronic computer can operate; ] Configuration diagram of conventional cooling system [Fig. 5] Descriptive diagram of fault tree [Fig. 6] Diagram showing operable temperature range of electronic computer [Description of symbols] 1 electronic computer 2 high-density semiconductor element 3 cooling system equipment 4 customer Cooling equipment 5 Refrigerant supply device 6 Monitoring sensor 7 Environmental cooling control device 8 CPU 9 Memory 10 State monitoring unit 11 FTA unit 12 Configuration control unit 13 Configuration control table 14 Configuration control device 20 Input recording unit 21 Fuzzy state monitoring unit 22 Fuzzy fault tree Analysis part 23 Knowledge base 31 Hypothesis cooling function permissible curve 32 Actual cooling function permissible curve 51-58, 61-64 nodes

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-3-209533 (JP, A) JP-A-4-178841 (JP, A) JP-A-3-80338 (JP, A) 70035 (JP, A) JP-A-2-159635 (JP, A) JP-A-61-18011 (JP, A) JP-A-59-231609 (JP, A) (58) Fields investigated (Int. Cl. ^7, DB name) G06F 1/20 G06F 11/22 - 11/26

Claims (1)

(57) [Claims] [Claim 1] Infer the presence or absence of a failure and determine the cause of the failure
Memorize rules for deriving and inferring weights by tree analysis
Of the knowledge base and the output of the monitoring sensor for monitoring the cooling system of the computer
An input recording unit for recording the history of the monitoring sensor, and fuzzy output of the monitoring sensor using the knowledge base.
Fuzzy state monitoring unit that determines the presence / absence of a failure by using the logic
When the fuzzy state monitoring unit determines that a failure has occurred,
From the output of the sensor and the history recorded in the input recording unit.
Fuzzy logic based on fault tree analysis using knowledge base
A cooling monitoring system for an electronic computer, comprising: a fuzzy fault tree analysis unit for further determining a failure cause; and a configuration control unit for separating a faulty device of the determined failure cause.
RAS mechanism.