JPH0991254A - Power consumption reduction control system and method therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the power consumption of the whole of a cluster system in accordance with the load state of the cluster system. SOLUTION: An operation mode table and an optimum node number table are stored in a front end processor 300, and incorporation or disconnection of a node selected by a node selection circuit 400 is instructed by an incorporation instructing circuit 600 or a disconnection instructing circuit 500, and the node is incorporated to switch the system to the power saving state or is disconnected to switch the system to the normal power consumption state by an incorporation control circuit 130 or a disconnection control circuit 120 which are connected to each node.

Description

Detailed Description of the Invention

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a power consumption reduction control system, and more particularly to a power consumption reduction control system for a cluster system in which a plurality of computers execute distributed processing.

[0001]

2. Description of the Related Art A cluster system is used for the purpose of realizing performance improvement and availability improvement of the whole system by distributing processing requests by configuring a computer system with a plurality of computers (hereinafter referred to as nodes). Referring to FIG. 8, the cluster system to which the present invention is applied is
It has a plurality of nodes 110, a shared storage 200 shared by these nodes, and a front-end processor 300 that receives a transaction request from the outside and allocates it to each node 110. For example,
"Summer 1991, Digital Technical Journal,
Volume 3, Issue 3 (Digital Technical Journal, Vol.3,
No.3, Summer 1991) ", a technique for sharing data on a magnetic disk device or magnetic tape device among multiple loosely coupled nodes to distribute the load to multiple nodes for processing, and a cluster system. There is described a technique of disconnecting a node having a failure when a failure occurs while continuing the process of (1) and further re-installing the node according to an operator's instruction after recovery.

On the other hand, as a conventional technique for reducing the power consumption of a computer, for example, "September 13, 1993, Nikkei Electronics, No. 590, pages 104 to 133" describes:
The operating status of each peripheral device is grasped by monitoring the input / output processing request made to the peripheral device such as the connected display and magnetic disk in the desktop personal computer, and the operating status of each peripheral device is adjusted according to the operating status. There is described a technique for controlling power consumption to reduce power consumption.

Further, this document describes a method for reducing the power consumption of the entire system including a plurality of computers and peripheral devices. According to it, in order to reduce the power consumption of the entire system composed of multiple computers and peripheral devices, it is not enough to reduce the power consumption of each device alone, and it is necessary to coordinate the power control between the devices. It has been pointed out. Furthermore, in order for the user to be able to use the devices of different manufacturers in combination, the following standard power control protocol, that is, an interface is required between the devices. First, between the computer and the display,
An interface is required to control from the computer when the display enters the standby state.
Secondly, between the computers connected via the network, an interface for returning the computer to the operating state when the computer in the standby state is accessed is required. Thirdly, an interface for executing power control on the expansion board is required between the computer body and the expansion board. And fourth, an interface is required between the computer body and the power supply for switching to a small capacity auxiliary power supply when the computer enters the standby state. Among them, it is pointed out that the interface between the first computer and the display is realized, but the other second to fourth interfaces are not realized.

[0004]

As described above, in the conventional cluster system, the disconnection of the node is triggered by the failure, and the subsequent integration is performed based on the instruction from the operator. Therefore, in order to reduce the power consumption in the conventional cluster system, it is unavoidable to rely on the individual power saving function of each node, not by disconnecting the node. In this case, even if the load on the cluster system as a whole is light, there is a possibility that there is always processing being executed in the node, and the power saving function cannot be fully utilized.

An object of the present invention is to reduce the power consumption of the entire cluster system according to the load state of the cluster system.

[0006]

In order to solve the above problems, a power consumption reduction control system of the present invention is a cluster system that receives a processing request from the outside and allocates the processing request to a node. The optimal number of nodes is calculated from the frequency of processing requests, and it is determined whether the number of nodes needs to be increased or decreased according to the operating status of each node, and then the nodes are disconnected from the cluster system to transition to the power saving state or Incorporate a node and make a transition to the normal power consumption state.

Further, another power consumption reduction control system of the present invention is a cluster system comprising a plurality of nodes, and means for receiving a processing request from the outside and allocating the processing request to the node, Means for calculating an optimal number of nodes from the frequency of received processing requests; means for holding whether each node is in operation or in a dormant state; means for judging whether or not the number of nodes needs to be increased; The node selecting means for selecting a target node when it is determined that the increase of the node is necessary, and the means for incorporating the node selected by the node selecting means to make a transition to a normal power consumption state.

Further, another power consumption reduction control system of the present invention is a cluster system comprising a plurality of nodes, and means for receiving a processing request from the outside and allocating the processing request to the node, Means for calculating an optimal number of nodes from the frequency of received processing requests; means for holding whether each node is in operation or in a dormant state; means for determining whether or not the number of nodes needs to be reduced; And a node selecting unit that selects a target node when it is determined that the reduction is required, and a unit that disconnects the node selected by the node selecting unit and shifts to a power saving state.

Another power consumption reduction control system of the present invention is, in a cluster system composed of a plurality of nodes, means for receiving a processing request from the outside and allocating the processing request to the node, and receiving the processing request from the outside. A means for calculating the optimum number of nodes from the frequency of processing requests, a means for holding whether each of the nodes is in operation or a dormant state, a means for deciding whether or not to increase or decrease the number of nodes, and an increase or decrease of nodes Node selecting means for selecting a target node when it is determined that the node is necessary, means for incorporating the node selected by the node selecting means to make a transition to a normal power consumption state, and the node selecting means. Means for disconnecting the selected node and transitioning to a power saving state.

Further, in another power consumption reduction control system of the present invention, when the node selected by the node selection circuit cannot transit to an operating state, the circuit for incorporating the node is a node having a fault. Register as.

Further, in another power consumption reduction control system of the present invention, each of the nodes has means for controlling incorporation or disconnection of the node.

The power consumption reduction control method of the present invention is
In a cluster system that receives a processing request from the outside and allocates the processing request to a node, a step of calculating an optimum number of nodes from the frequency of processing requests received from the outside, and the optimum number of nodes are currently operating. When the number of nodes is larger than the number of nodes, the step of putting the nodes in the dormant state into the operating state, and when the optimal number of nodes is smaller than the number of the nodes currently in operation, putting the node in the operating state into the dormant state and saving Transitioning to a power state.

According to the present invention, node control according to the operating status of the entire cluster system is performed by selecting a node for which operation is suspended or resumed according to the operating status obtained by means for monitoring the operating status of the entire cluster system. . A first configuration control unit that disconnects the target node that has been determined to be suspended from the cluster system, and a node in the operating state that has been disconnected from the cluster system by this first configuration control unit transits to a sleep state in which power consumption is low. The first transition means for reducing power consumption according to the operating status of the entire cluster system. By the second transition means for transitioning the target node in the dormant state whose operation is decided to be resumed by selection to the operating state, and the second configuration control means for incorporating the node transitioned to the operating state by this second transition means into the cluster system. , Power consumption is reduced according to the operating status of the entire cluster system.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION Next, an embodiment of a power consumption reduction control system of the present invention will be described in detail with reference to the drawings.

Referring to FIG. 1, a power consumption reduction control system according to an embodiment of the present invention includes a plurality of nodes 110.
A disconnection control circuit 120 that controls the disconnection of each node 110, an embedded control circuit 130 that controls the integration of each node 110, a shared storage 200 shared by each node, and a transaction request from the outside. A front-end processor 300 assigned to the node 110, a node selection circuit 400 for investigating the operating status of the cluster system and selecting a node to be disconnected or incorporated, a disconnection instruction circuit 500 for instructing node disconnection, and a node incorporation And a built-in instruction circuit 600 for instructing.

The node selection circuit 400 investigates the operation status of the cluster system, selects a node suitable for changing the operation state, cuts off the result, and instructs the instruction circuit 500 and the built-in instruction circuit 600. Cut-off instruction circuit 50
0 and the built-in instructing circuit 600 instruct each node 110 to change the operating state of the node in order to change the configuration of the cluster system based on the instruction from the node selecting circuit 400. The disconnection instruction circuit 500 is provided in each node 110.
A disconnection control circuit 120 that transitions the node from the operating state to the dormant state based on an instruction from the embedded instruction circuit 600 and an embedded control circuit 130 that transitions the node from the dormant state to the operating state are connected.

The front-end processor 300 has means for storing the number of processing requests (hereinafter referred to as NR) received as a cluster system per a predetermined time.

On the other hand, the node selection circuit 400 has an operation mode table that holds the operation modes of all the nodes of the cluster system, and an optimum number of built-in nodes preset according to the characteristics of the cluster system and the processing characteristics. An optimum node number table that holds a range of the number of nodes (hereinafter referred to as NT).

Referring to FIG. 2, the operation mode table holds the correspondence between node names and their operation modes. As the operation mode, "operation state" indicating that the operation is in progress, "failure state" indicating that the device is disconnected due to occurrence of a failure, or "disconnected for power saving" Hibernation ". The node selection circuit 400 can recognize the number of operating nodes (hereinafter referred to as NA) based on this operation mode table.

Referring to FIG. 3, the optimum node number table is as follows.
The optimum number of nodes NT is held corresponding to the value of the number of processing requests NR. The node selection circuit 400 examines the optimum node number NT based on this optimum node number table, and compares it with the current operating state node number NA to determine the increase or decrease of the node number.

Next, the operation of the above embodiment of the power consumption reduction control system of the present invention will be described in detail with reference to the drawings.

Referring to FIGS. 1 and 4, first, the node selection circuit 400 checks the operating status of the cluster system (S201). Based on the result of this investigation, the node selection circuit 400 keeps all the nodes in the current state, disconnects one of the nodes incorporated in the cluster system and shifts it to the dormant state, or puts it in the dormant state. It is determined whether any one of the nodes is in the operating state (S202, S203).

If it is decided to keep all the nodes in the current state (S202), the control is terminated as it is.

When it is determined that any one of the nodes should be transited to the sleep state, the node selection circuit 400
Selects an appropriate node for sleep and instructs the disconnection instruction circuit 500 to transition the node to the sleep state. If there is no suitable node here, the control ends (S208). The disconnection instruction circuit 500 disconnects the cluster system so that a new processing request is not made from the front-end processor 300 to the node (S209). The cutoff instruction circuit 500 waits for the completion of the cutoff and then makes a transition request to the cutoff control circuit 120 connected to the node. The disconnection control circuit 120 of the node that has received the transition request transitions the node to the sleep state and ends the control (S210).

When it is determined that any one of the nodes should transit to the built-in state, the node selection circuit 40
0 selects an appropriate node to be incorporated and instructs the incorporation instruction circuit 600 to make a transition of that node. If there is no suitable node here, the control ends (S204). The built-in instruction circuit 600 requests the built-in control circuit 130 connected to the corresponding node to transition to the operating state. In response to this request, the embedded control circuit 130 transitions the node to the operating state, and reports this to the embedded instruction circuit 600 after the transition is completed (S206). At this time, if the node fails in the transition to the mode, the built-in instruction circuit 60
0 indicates that the node is in a failure state in the operation mode table,
Control is executed again from node selection (S207). Built-in instruction circuit 600 that has received the completion report of transition to the operating state
Causes the front end processor 300 to make a processing request to the node (S206).

The above control is executed at predetermined time intervals.

With reference to FIGS. 1 and 5, the investigation of the operating status of the cluster system and the selection of nodes are performed as follows.

First, the node selection circuit 400 inquires of the front-end processor 300 about the current processing request number NR. In response to this, the front end processor 3
00 transmits the processing request number NR to the node selection circuit 400 (S301). The node selection circuit 400 refers to the optimum node number table to investigate the optimum node number NT for the processing request number NR. In addition, the number of active nodes NA is checked in the operation mode table (S302). This allows
It is checked whether or not the optimum number of nodes NT and the current number of operating state nodes NA match (S303). If they match, it is determined that the number of operating nodes is not increased or decreased, and the control ends. The optimum number of nodes NT is the number of operating nodes NA
When the number is larger (NT> NA), it is determined that the number of operating nodes should be increased. When the optimum number of nodes NT is smaller than the number of operating nodes (NT <NA), it is determined that the number of operating nodes should be reduced (S30).
4). If you decide to increase the number of active nodes,
It is checked with reference to the operation mode table whether there is a hibernating node (S308). If there is no dormant node, the control is terminated as it is. 1 dormant node
If more than one device exists, one device is randomly selected from the nodes in the dormant state and the embedded processing control of the node is started (S309, S500). On the other hand, when it is determined that the number of operating nodes is reduced, it is checked by referring to the operating mode table whether there are two or more operating nodes (S30).
5). If there are no more than two nodes in operation, control ends as it is. When there are two or more hibernating nodes, one is randomly selected from the operating nodes and the hibernation control of the node is started (S30).
6, S400).

Here, a specific operation example will be described.
Referring to FIG. 2, the operation mode table holds that node A is in an operating state, node B is in a dormant state, and node C is in a failure state. Further, referring to FIG. 3, in the optimum node number table, one node is suitable if the processing request number NR is 0 to 99, two nodes are suitable if the processing request number NR is 100 to 149, and 150 to 199. If so, it is held that 3 nodes are appropriate. The node selection circuit 400 is
It is assumed that the front-end processor 300 is inquired about the processing request number NR, and as a result, NR = 120 is obtained (S2
01). At this time, referring to the optimum node number table, it can be seen that the optimum node number NT = 2 (S202). On the other hand, from the examination of the operation mode table, the current operating state number N of nodes
A = 1 and does not match NT (S303). NT> N
Since it is A, it is determined that the number of operating nodes needs to be increased (S304). At this time, referring to the operation mode table, the node B is in the dormant state, and therefore, the node B is selected and brought into the operating state (S309). The result is transmitted to the built-in instruction circuit 600. It should be noted that the selection of the node in the dormant state can be performed in the same manner, and the description thereof will be omitted.

In the above description, the operating node is selected by grasping the operating status by using the number of processing requests to the cluster system per fixed time as an index, but similarly, the processing request being executed may be used as an index of the operating status. The number, or the operating rate of the processor, memory or I / O bus of each node may be considered. It can be easily inferred that the same selection can be made for these as well by storing the optimum number of operating nodes and the threshold value for increasing or decreasing this number as in the present embodiment. Also, in the case of a cluster system in which nodes with different configurations such as the number of processors are mixed as the nodes of the cluster system, an index having the same unit as the above-mentioned index when selecting the operating node for each node is calculated for each node. It is possible to select an operating node similar to that in the above-mentioned embodiment by comparing the total of the indexes and the operating state by storing the same.

Referring to FIGS. 1 and 6, the control for disconnecting the active node to make it transit to the inactive state and the configuration control for bringing the inactive node into the active state and incorporating it into the cluster system are performed as follows.

First, when bringing a node in an operating state into a dormant state, it is necessary to make a transition to a disconnected state in which a processing request to the cluster system is not made. When receiving the instruction from the node selection circuit 400 to suspend the operating node A, the disconnect instruction circuit 500 changes the node A column in the operating mode table of the front-end processor 300 to the suspending state (S401). Cut-off instruction circuit 500
Determines from the information from the front-end processor 300 whether or not all the processes performed in the node A to be disconnected have been completed (S402). Node A
Disconnection instruction circuit 500 that detects completion of processing in
Is the node A for the front end processor 300.
To the disconnect control circuit 120 connected to the control circuit 120 to instruct the disconnect. In response to this, the front-end processor 300 removes the node A from the fault monitoring members and reports to the disconnection instruction circuit that it should be disconnected (S4).
03). In response to this report, the disconnection instruction circuit 500 requests the disconnection control circuit 120 connected to the node A to transit to the sleep state (S404). The disconnection instruction circuit 500 instructs the operating system of the node A to transit to the hibernation state, and the operating system lowers the operating frequency of the node A and stops the spindle motor of the connected disk (not shown) (S405). .

Referring to FIGS. 1 and 7, when the node in the idle state is put into the operating state, the built-in instruction circuit 600
Requests the embedded control circuit 130 connected to the node A to transition to the operating state (S501). The embedded control circuit 130 instructs the operating system of the node A to shift the node to the operating state. The operating system raises the clock of the node A to be the same as the operating state, and also operates the spindle motor of the connected disk (not shown) (SS502).
When the transition to the operating state has failed (S503), the embedded control circuit 130 reports the failure to the embedded instruction circuit 600. Upon receiving this report, the built-in instruction circuit 600 reports the occurrence of a failure to the front end processor 300 and the node selection circuit 400. The node selection circuit 400 sets the state of the corresponding node in the operation mode table to the fault state and starts over from the selection of the embedded node. When the transition to the operating state is completed normally, the fact is reported to the built-in instruction circuit 600 (S504). Upon receipt of this report, the embedded instruction circuit 600 notifies the front-end processor 300 to add the node to be incorporated as a member of fault monitoring (S505). The embedded instruction circuit 600 changes the state of the node A in the operation mode table of the front-end processor 300 to the operation state (S506).

As described above, according to the power consumption reduction control system of the embodiment of the present invention, the operation mode table and the optimum node number table are stored in the front-end processor 300, and selected by the node selection circuit 400. A cluster system by instructing the incorporation or disconnection of a node by the incorporation instruction circuit 600 or the disconnection instruction circuit 500, and controlling the incorporation or disconnection of the node by the embedded control circuit 130 or the disconnection control circuit 120 connected to each node. It is possible to reduce the power consumption according to the operating status of.

In the above embodiment, the front-end processor 300, the node selection circuit 400, the disconnection instruction circuit 5
00 and the built-in instruction circuit 600 have been described as independent circuits, but the functions of the front-end processor 300 and the node selection circuit 400, or the functions of the cutoff instruction circuit 500 and the embedded instruction circuit 600 may be given to a specific node. Needless to say, it is okay.

Further, these front end processors 3
00 or the function of the node selection circuit 400 or the function of the disconnection instruction circuit 500 or the built-in instruction circuit 600 may be distributed to each node so that any one of the nodes operates according to the load condition.

[0037]

As is apparent from the above description, according to the present invention, the power consumption is controlled according to the operating condition of the entire cluster system, so that the power consumption is reduced when the load of the cluster system is light. You can

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a power consumption reduction control system of the present invention.

FIG. 2 is a diagram showing the contents of an operation mode table in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 3 is a diagram showing the contents of an optimum node table in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 4 is a diagram showing an outline of an operation in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 5 is a diagram showing a detailed operation in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 6 is a diagram showing a cut-off control operation in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 7 is a diagram showing an operation of embedded control in the power consumption reduction control system according to the embodiment of the present invention.

FIG. 8 is a block diagram showing a configuration of an example of a conventional power consumption reduction control system.

[Explanation of symbols]

 110 nodes 120 disconnection control circuit 130 embedded control circuit 200 shared storage 300 front-end processor 400 node selection circuit 500 disconnection instruction circuit 600 embedded instruction circuit

Claims (5)

[Claims] 1. In a cluster system that receives a processing request from the outside and allocates the processing request to a node, the optimum number of nodes is calculated from the frequency of the processing request received from the outside, and the operating status of each node is calculated. A power consumption reduction control system characterized by determining whether or not to increase or decrease the number of nodes according to the above, and then disconnecting the node from the cluster system to make a transition to a power saving state or incorporating a node to make a transition to a normal power consumption state. . 2. In a cluster system comprising a plurality of nodes, means for receiving a processing request from the outside and allocating the processing request to the node, and calculating the optimum number of nodes from the frequency of the processing request received from the outside. Means, a means for holding whether each of the nodes is in operation or a dormant state, a means for determining whether or not the number of nodes needs to be increased, and a target when it is determined that the number of nodes needs to be increased. A power consumption reduction control system, comprising: a node selecting means for selecting a node to be formed, and a means for incorporating the node selected by the node selecting means to make a transition to a normal power consumption state. 3. In a cluster system comprising a plurality of nodes, means for receiving a processing request from the outside and allocating the processing request to the node, and calculating the optimum number of nodes from the frequency of the processing request received from the outside. Means, a means for holding whether each of the nodes is in operation or a dormant state, a means for deciding whether or not to reduce the number of nodes, and a target when it is determined that the number of nodes needs to be reduced. A power consumption reduction control system comprising: a node selecting means for selecting a node to be formed, and a means for disconnecting the node selected by the node selecting means to make a transition to a power saving state. 4. In a cluster system including a plurality of nodes, means for receiving a processing request from the outside and allocating the processing request to the node, and calculating the optimum number of nodes from the frequency of the processing request received from the outside. Means, a means for holding whether each of the nodes is in operation or a dormant state, a means for determining whether or not the number of nodes needs to be increased or decreased, and a target when it is determined that the number of nodes needs to be increased or decreased. Node selecting means for selecting a node to be selected, means for incorporating the node selected by this node selecting means to transition to a normal power consumption state, and disconnecting the node selected by the node selecting means to transition to a power saving state A power consumption reduction control system comprising: 5. In a cluster system that receives a processing request from the outside and allocates the processing request to a node, a step of calculating an optimal number of nodes from the frequency of processing requests received from the outside, and the optimal number of nodes. The number of nodes in operation is greater than the number of nodes currently in operation, the step of putting the nodes in the dormant state into operation, and if the optimal number of nodes is less than the number of nodes currently in operation, A power-saving reduction control method, comprising a step of changing to a power-saving state by setting a sleep state.