CN103034539A - Server system and power management method thereof - Google Patents
Server system and power management method thereof Download PDFInfo
- Publication number
- CN103034539A CN103034539A CN2011102962125A CN201110296212A CN103034539A CN 103034539 A CN103034539 A CN 103034539A CN 2011102962125 A CN2011102962125 A CN 2011102962125A CN 201110296212 A CN201110296212 A CN 201110296212A CN 103034539 A CN103034539 A CN 103034539A
- Authority
- CN
- China
- Prior art keywords
- cpu
- state
- node
- processing unit
- central processing
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
Disclosed are a server system and a power management method thereof. The power management method of the server system comprises the following steps: a usage rate corresponding to at least one central processing unit (CPU) in a specific node is detected so as to calculate an average service rate of the specific node. A state of the specific node is judged according to the average service rate. When the state is lower than a normal load state, an operation parameter of the CPU is adjusted so as to reduce processing speed of the CPU, and when the state is higher than the normal load state, the operation parameter of the CPU is adjusted so as to improve the processing speed of the CPU, and therefore the state of the specific node can be in the normal load state. According to the power management method, operating efficiency and working period of the CPU are adjusted dynamically based on the usage rate of the CPU of the specific node so that power supply consumption is saved without affecting work efficiency of a virtual host machine.
Description
Technical field
The present invention relates to a kind of electric power management mechanism, relate in particular to a kind of server system and method for managing power supply thereof.
Background technology
High in the clouds computing (cloud computing) technology be by world-wide web (Internet) in conjunction with a large amount of servomechanisms to form superelevation computing and super large storage capacity.The high in the clouds computing can be shared the software and hardware of these servomechanisms or divide the work, and operates the webpage of these services by network and browser, and the interface above using carries out various computings and work.Because the quantity of these servomechanisms is huge, a servomechanism group of planes (server group) that is comprised of servomechanism can adopt electric power management mechanism to reduce electrical source consumption usually.
Shown in Figure 1, Fig. 1 is a kind of synoptic diagram of server system 100.Have a large amount of node PM (Physical Machine) in server system 100, each node PM is the entity main frame in the server system 100, and these nodes PM all links to each other with network 110.And, all set up one or more fictitious host computer VM (virtual machine) on each node PM, to carry out the high in the clouds calculation function by network 110.Management servomechanism 120 comes these nodes PM is carried out power management by network 110.In part embodiment, management servomechanism 120 can be certain node PM in the server system 100, and it can carry out specific function in the fictitious host computer to carry out electric power management mechanism.
Electric power management mechanism in the past can be transferred to the lighter node PM of total amount load with the heavier fictitious host computer VM of load among the part node PM, load with each node PM of balance, this action also can be described as the instant transfer (live migration) of fictitious host computer VM, uses the load balance (load balance) of realizing each node.
Immediately after shifting, when if the average total amount load of each node PM is still on the low side, just the fictitious host computer VM among some node PM can be transferred to totally among other nodes PM, and close and do not have fictitious host computer VM node PM therein, just can reduce the quantity of node PM in the running, also can make the node PM in the running carry out all fictitious host computers in the most effective power supply service efficiency, above-mentioned action can be described as node and closes (node power-off).Relatively, when being difficult to load if the average total amount load of each node PM is higher, just can open/restart the not node PM of running, and carry out the instant transfer of fictitious host computer VM, to keep the power supply service efficiency, above-mentioned action also can be described as node and starts (node power-on).
Although above-mentioned electric power management mechanism can be saved more power supply in theory, fictitious host computer VM will expend a large amount of network bandwidths in its transfer process, thereby reduces the usefulness of whole high in the clouds computing.The load of each fictitious host computer VM also can be because of factors such as the kinds of user's quantity, application software and is changed at any time, thereby is difficult to estimate the generation time point of these variations.Therefore, if node PM can be regular carry out the power-off of instant transfer, node PM of fictitious host computer VM when the electric power management mechanism such as restarting, efficient that will the computing of decrease high in the clouds.Therefore, in the computing, how to adopt better electric power management mechanism at the server system 100 with numerous nodes is to want at present direction to be furtherd investigate beyond the clouds.
Summary of the invention
The invention provides a kind of server system and method for managing power supply thereof, its CPU (central processing unit) according to node (CPU) utilization rate is dynamically adjusted the operation efficiency of CPU and the state of work period, to save electrical source consumption under the work efficiency that does not affect fictitious host computer.
The invention provides a kind of method for managing power supply of server system, above-mentioned server system includes a plurality of nodes.The method for managing power supply of server system comprises the following steps.Detect each node (for example, a specific node) corresponding utilization rate of interior CPU (central processing unit) to calculate the CPU average service rate of this specific node.According to this average service rate to judge the state of this specific node.When this state is lower than the normal load state, just adjust the CPU operating parameter in this specific node, reducing the processing speed of CPU, and make this state get back to the normal load state and save electrical source consumption.Similarly, when this state is higher than the normal load state, just adjusts CPU operating parameter in this specific node promoting the processing speed of CPU, and make the state of this CPU get back to the normal load state.
In other words, the invention provides a kind of method for managing power supply of server system, wherein this server system comprises a plurality of nodes, the method for managing power supply of this server system comprises the following steps: to detect in the specific node the corresponding at least one utilization rate of at least one CPU (central processing unit) to calculate an average service rate of this specific node, wherein this specific node one that is those nodes; Judge a state of this specific node according to this average service rate; When this state is lower than a normal load state, adjust an operating parameter of this CPU (central processing unit) in this specific node to reduce the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state; And when this state is higher than this normal load state, adjust this operating parameter to promote the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state.
In one embodiment of this invention, the method for managing power supply of server system also comprises the following steps.When the operating parameter that increases above-mentioned CPU still can't make above-mentioned state get back to the normal load state (for example, when the state of CPU be fully loaded load condition and CPU the operation efficiency adjusted to the work period adjusted of maximal value and CPU to maximal value), then carry out the instant branching program of a fictitious host computer, the load of being born with these nodes of balance.
In one embodiment of this invention, the method for managing power supply of server system also comprises the following steps.When the processing speed that reduces CPU still can't make above-mentioned state get back to the normal load state, just carry out a node bolt down procedure and cooperate the instant branching program of above-mentioned fictitious host computer, so that the state of each node in the running is got back to the normal load state.Relatively, when the processing speed that promotes CPU still can't make above-mentioned state get back to the normal load state, just carry out a node start-up routine and cooperate the instant branching program of above-mentioned fictitious host computer, so that the state of each node in the running is got back to the normal load state.
In one embodiment of this invention, the state of judging above-mentioned specific node includes the following step.When average service rate is higher than a maximum CPU threshold value, above-mentioned setting state is fully loaded load condition.When average service rate is higher than a height CPU threshold value and is lower than maximum CPU threshold value, then be the height load condition with above-mentioned setting state.When average service rate is higher than a low CPU threshold value and be lower than height CPU threshold value, be the normal load state with above-mentioned setting state then.And, when average service rate is lower than above-mentioned low CPU threshold value, be low load condition with above-mentioned setting state.In addition, when average service rate is lower than an idle CPU threshold value, above-mentioned setting state is idle load condition, wherein, above-mentioned idle CPU threshold value should be lower than low CPU threshold value.
On the other hand, the invention provides a kind of server system, include a plurality of nodes and a management servomechanism.Each node can be regarded as an entity main frame, so each node all comprises at least one CPU (central processing unit) (CPU).Each node and management servomechanism all are coupled to network.The management servomechanism can detect the utilization rate of CPU of each node (for example, a specific node) calculating the average service rate of this specific node, and judges the state of specific node according to average service rate.When above-mentioned state was lower than the normal load state, the management servomechanism was just adjusted the operating parameter of CPU in the specific node to reduce the processing speed of CPU.And when above-mentioned state was higher than the normal load state, the management servomechanism was just adjusted CPU operating parameter in the specific node to promote the processing speed of CPU, so that above-mentioned state is got back to the normal load state.In addition, all the other implementation details of server system please refer to above-mentioned explanation, do not add at this and give unnecessary details.
In other words, the invention provides a kind of server system, comprising: a plurality of nodes and a management servomechanism.Each node is coupled to a network, and each node comprises at least one CPU (central processing unit); The management servomechanism is coupled to this network, this management servomechanism detects in the specific node the corresponding utilization rate of this CPU (central processing unit) to calculate an average service rate of this specific node, judge a state of this specific node according to this average service rate, when this state is lower than a normal load state, adjust an operating parameter of this CPU (central processing unit) in this specific node to reduce the processing speed of this CPU (central processing unit), and when this state is higher than this normal load state, adjust this operating parameter to promote the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state, this specific node one that is those nodes wherein.
Based on above-mentioned, because electric power management mechanism in the past can occupy in the process of implementation than multi-system resource, also expend the more processing time.By this, the embodiment of the invention can according to the overall load of each node (or, the average service rate of the CPU (central processing unit) of each node (CPU)) dynamically adjusts the running speed of CPU (central processing unit) (CPU), so that each node can turn round in the normal load state, with in the work efficiency that does not affect each node, do not occupy and save electrical source consumption under the system resource.
By this, instant transfer and node that the embodiment of the invention can be avoided regularly carrying out fictitious host computer are closed/the node startup, have exempted the reduction of high in the clouds operation efficiency.Be on the point of rate (P) and work period (T) still can't make the server system running in better consumption efficiency the time when the running of adjusting CPU, carry out again electric power management mechanism in the past, just can reach best power management effect.
For above-mentioned feature and advantage of the present invention can be become apparent, embodiment cited below particularly, and cooperate accompanying drawing to be described in detail below.
Description of drawings
Fig. 1 is a kind of synoptic diagram of server system.
Fig. 2 is the process flow diagram according to the method for managing power supply of one embodiment of the invention explanation server system.
Fig. 3 is the synoptic diagram according to one embodiment of the invention explanation server system.
Fig. 4 is the synoptic diagram of BIOS in the specific node, chipset and one of them entity CPU.
Fig. 5 is the CPU average service rate state table in the management servomechanism.
Fig. 6 is the described CPU operating parameter of embodiment of the invention table.
Reference numeral:
100,300: server system
110,310: network
120,320: the management servomechanism
330: Basic Input or Output System (BIOS) (BIOS)
410: chipset
412,421～460: register
413: storer
414:SMI instruction output/input port
415:SMI data output/input port
This group register that CT1～CTN:CPU Thread includes
CPU: CPU (central processing unit)
PM: node/entity main frame
VM: fictitious host computer
S210～S280: step
The operation efficiency of P:CPU
The work period of T:CPU
Embodiment
Now will in detail with reference to example embodiment of the present invention, the example of described example embodiment be described in the accompanying drawings.In addition, all possibility parts, the element/member of use same numeral/identical or similar portions of symbology in drawings and the embodiments.
Please refer to Fig. 2, Fig. 2 is the process flow diagram according to the method for managing power supply of one embodiment of the invention explanation server system 300.Fig. 3 is the synoptic diagram according to one embodiment of the invention explanation server system 300.In present embodiment, at these server system 300 method for managing power supply that come execution graph 2 to show with server system shown in Figure 3 300.
In this explanation is that the entity main frame PM of indication can have a plurality of CPU slots herein, therefore can set up many multi-core (core) CPU that have, and for example sets up a plurality of four core CPU in entity main frame PM.Each core in the middle of each CPU also can include one or more Threads (thread).Therefore, present embodiment is called entity CPU with multi-core CPU, and each core among each entity CPU is called core CPU.And, also can include a plurality of Threads in each core CPU, and each CPU Thread is called logic CPU.By this, mentioned CPU can be entity CPU in present embodiment and Fig. 1 and Fig. 3, can be core CPU, also can be logic CPU.
At the method for managing power supply of this explanation server system 300, in present embodiment, this method for managing power supply can be every through after a while (for example, 60 seconds) just carry out once.For the purpose of simplifying the description, at this one of them node PM being called is a specific node.Therefore, in step S210, management servomechanism 320 detects the corresponding utilization rate of CPU in the specific node, and calculates by this average service rate of this specific node.The corresponding utilization rate of CPU means Thread in the middle of each CPU at the loading level of carrying out instruction, so each CPU usage then can average the loading level of each Thread among this CPU to calculate and tries to achieve.The highest CPU usage is 100% (all Threads should be in full load condition among the CPU at this moment), and minimum CPU usage then is 0% (all Threads almost is in idle state among the CPU at this moment).
As above-mentioned, owing to all can include a plurality of logic CPU or logic CPU on each entity node PM.Therefore, in calculating specific node, during the average service rate of CPU, just can detect the corresponding utilization rate of each CPU in the specific node in advance.Then, the utilization rate of these CPU added up and divided by the quantity of CPU on the specific node, to calculate the CPU average service rate on the specific node.
What mention especially is that in present embodiment, management servomechanism 320 detects and control for CPU usage detection, the CPU operating parameter of each node PM, all can adopt the software (Software of Basic Input or Output System (BIOS) (BIOS); SW) system management interrupt (System management interrupt; SMI) mechanism cooperates the chipset among each node PM to manage, and does not need to carry out this method for managing power supply by baseboard management controller (BMC), therefore can reduce the cost that sets up of node PM.Certainly, in other embodiment, management servomechanism 320 also can utilize the baseboard management controller among the node PM to carry out the detection of CPU usage, detection, management and the control of operating parameter, and the present invention is not restricted to this.
Describe present embodiment in detail at this and how to utilize SMI handling procedure among the BIOS, use utilization rate, the CPU operating parameter of obtaining each CPU among the node PM.Fig. 4 is the synoptic diagram of BIOS 330 among the specific node PM, chipset 410 and one of them entity CPU.For convenience of description, the entity CPU described in the present embodiment for example is the logic CPU with N Thread, and N is a positive integer.Each Thread CT1～CTN includes one group of register, for ease of explanation, represents respectively position this group register in Thread CT1～CTN at this with CT1～CTN.In present embodiment, each node PM can produce smi signal by chipset (chipset) 410 wherein.
For instance, chipset 410 includes list and points to register 412, SMI instruction output/input port (SMI command I/O Port) 414 and SMI data output/input port (SMI Data I/O Port) 415.In present embodiment, SMI instruction output/input port 414 and SMI data output/input port 415 are to realize by register.SMI instruction output/input port 414 and SMI data output/input port 415 are in order to carry out SMI instruction and the data access of SW SMI mechanism, and list sensing register 412 stores a storage address, use the storeroom that points to many lists among storer (the memory buffer) 413, allow the SMI handling procedure (SMI handler) of BIOS 330 be able to reference to the following present utilization rate state table of CPU, the operation efficiency P of CPU and work period T scope table, the lists such as the operation efficiency P of CPU and work period T operating parameter table are used the required input that operation is provided, output parameter.In implementation, the address of supposing SMI instruction output/input port 414 is that the address of B2 and SMI data output/input port 415 is B3, and definition F0 value is the administration order of the CPU Thread of SW SMI mechanism.
Therefore, please refer to Fig. 4, when chipset 410 writes the address B2 of SMI instruction output/input port 414 with the F0 value, B2 will produce a smi signal to entity CPU, and entity CPU will skip to SMI handling procedure (SMI handler) corresponding among the BIOS 330 at once to process this smi signal.In detail, entity CPU will read among the address B2 of SMI instruction output/input port 414 value and the F0 value, this moment, entity CPU will carry out the supervisory routine of CPU Thread, and this program can further read value among the address B3 of SMI data output/input port 415 in order to learn that further what will be done works.In herein, the value among the address B3 of present embodiment definition SMI data output/input port 415 is as follows:
0--is from the obtained present CPU usage of CPU register.
1--is from the obtained CPU operation efficiency P of CPU register, maximal value and the minimum value of work period T.
2--is from the obtained present operation efficiency P of CPU register, the value of work period T.
The operation efficiency P that 3--sets the CPU register, the value of work period T.
Therefore, the supervisory routine of CPU Thread will have work as above to carry out, but the required input of these work, output parameter can be pointed among register 412 storer (the memory buffer) 413 pointed many lists by list and obtain.
When chipset 410 sends smi signal to CPU, Thread among the CPU all can be received this smi signal, then each Thread just can arrive and read each self-corresponding SMI handling procedure among the BIOS 330 and carry out, and uses the utilization rate that obtains corresponding CPU, and according to CPU corresponding to operating parameter control.
For convenience of explanation, this with entity CPU in one group of register CT1 corresponding to a Thread as an example, (for example, the framework in the middle of the register group CT2～CTN) is because its framework should be identical with register group CT1 and Fig. 4 has omitted other register groups.In present embodiment, each register group CT1～CTN includes System Management Mode substrate register (system management mode base register) 450; Reach P present situation register 421, P maximum value register 422 and the P minimum value register 423 of present situation, P maximal value and the P minimum value of temporary operation efficiency P in order to control; And, in order to control and to keep in T present situation register 431, T maximum value register 432 and the T minimum value register 433 of present situation, T maximal value and the T minimum value of work period T.SMM (system management mode) substrate register 450 is used the inlet point (entry) that stores the SMI handling procedure of required execution in each Thread in order to the address starting point among the temporary BIOS 330.In addition, register group CT1～CTN also comprises the CPU usage register 460 that can keep in this Thread utilization rate.
By above-mentioned framework, each Thread among the CPU is after receiving smi signal, convenient with address starting point stored in SMM (system management mode) the substrate register 450 to skip to the inlet point of SMI handling procedure, use and begin to carry out SMI handling procedure corresponding to each Thread, and in will be in the SMI handling procedure obtained register 421～460 that relevant information is temporary and control is corresponding.With present embodiment flow process, be to utilize in the middle of the CPU one of them to be called system to start banded processor (system boot strap processor; SBSP) Thread can be carried out normal SMI handling procedure, other Threads (being called application processor (application processor)) realize by signal flag (semaphore) mode and SBSP between synchronization action, finish together the SMI handling procedure by carrying out the RSM instruction again, and make CPU leave System Management pattern (SMM).The SMI mechanism of this CPU is the above-mentioned relevant action that how to utilize the operation of utilization rate, CPU operating parameter that SMI interface among the BIOS obtains each CPU among the node PM and the P that uses operating parameter control CPU and T namely.
On the other hand, the flow process of above-mentioned SMI handling procedure outlines as follows.At first, this SMI handling procedure can be analyzed first the smi signal that is sent by chipset 410, to judge the action of required execution.For example, present embodiment is divided into following four kinds with smi signal:
After having analyzed smi signal, the SMI handling procedure can point to the pointed various lists of register 412 by list in the chipset 410, and carries out corresponding action according to above-mentioned instruction.For example, learn by these lists why are the output of its instruction and input parameter.
Get back to Fig. 2 and Fig. 3, in step S215, management servomechanism 320 is then just judged the load condition of this specific node according to the average service rate that is calculated by step S210.In present embodiment, management servomechanism 320 can utilize built-in CPU average service rate state table to judge the load condition that each node is present, and this CPU average service rate state table also can be adjusted and be set by the user.
Fig. 5 is the CPU average service rate state table in the management servomechanism 320.As shown in Figure 5, because CPU usage is the highest to be 100%, therefore the average service rate when specific node (for example is higher than maximum CPU threshold value, 95%) and less than or equal to 100% the time, management servomechanism 320 just is fully loaded load condition with the setting state of specific node, in other words, the status code with specific node is set as 5.When average service rate was higher than height CPU threshold value (for example, 80%) and is lower than maximum CPU threshold value (95%), then the setting state with this specific node was height load condition (status code of setting specific node is 4).
The rest may be inferred, when average service rate (for example is higher than low CPU threshold value, 40%) and be lower than height CPU threshold value when (80%), represent that this specific node is arranged in better electrical source consumption power, and the fictitious host computer VM on the specific node all can normally carry out not affecting under its usefulness, is normal load state (status code of setting specific node is 3) with the setting state of this specific node then therefore.And, when average service rate be lower than respectively low CPU threshold value (40%) or even when being lower than idle CPU threshold value (10%), then the state with this specific node is set as respectively low load condition (status code of setting specific node is 2) or idle load condition (status code of setting specific node is 1).
Get back to Fig. 2, then enter step S220, management servomechanism 320 is judged respectively the load condition of each node PM.If the load condition of each node PM is all the normal load state, just can directly enter step S225 to finish this method for managing power supply.When the load condition of specific node is lower than normal load state (status code is 3) (for example, when specific node is positioned at low load condition (status code is 2) or idle load condition (status code is 1)), just enter step S230 by step S220, the management servomechanism 320 just adjust in this specific node the CPU operating parameter (for example, the operation efficiency P of CPU and work period (duty cycle) T) to reduce the processing speed of CPU, and expect to make this state get back to the normal load state, use the electrical source consumption of saving in this specific node.
Comparatively speaking, when the load condition of specific node is higher than normal load state (status code is 3) (for example, when specific node is positioned at fully loaded load condition (status code is 5) or height load condition (status code is 4)), just enter step S240 by step S220, the management servomechanism 320 adjust according to this in the specific node the CPU operating parameter (for example, the operation efficiency P of CPU or work period (duty cycle) T) improving the processing speed of CPU, and expectation makes this state get back to the normal load state.
How to utilize the CPU operating parameter of adjusting in the specific node to reduce/improve the CPU processing speed at this explanation present embodiment.In present embodiment, the adjustment of CPU processing speed is the SMI interface that management servomechanism 320 utilizes BIOS, and utilizing a CPU operating parameter table to adjust to operation efficiency P and the work period T of each CPU, Fig. 6 is the described CPU operating parameter of embodiment of the invention table.
In Fig. 6, operation efficiency P and the work period T of CPU operating parameter code in order to set each CPU.Operation efficiency P and work period T then are with number percent as an example in Fig. 4, for example, when the operating parameter code of a certain CPU is set to 1, the operation efficiency P that represents this CPU is set to maximal value (100%), and the work period T of this CPU also is set to maximal value (100%), so that CPU can come processing instruction with the processing speed of maximum amount.Comparatively speaking, when the operating parameter code of a certain CPU is set to 8, it is peaked 25% that the operation efficiency P that represents this CPU then is set to, and that the work period T of this CPU also is set to is peaked 12.5%, so that CPU can come processing instruction with MIN processing speed.This special instruction be, use the present embodiment person and can adjust data in the CPU operating parameter table of Fig. 6 according to its demand, the present invention should be as limit.
Utilize Fig. 6 examples description of step S240 at this.The load condition of supposing specific node is height load condition (status code is 4), and the operating parameter code of each CPU all is set as 4 in the specific node, that is in the specific node each CPU operation efficiency P be set as peakedly 70%, its work period, T also was set to peaked 62.5%.Because the load condition of specific node (status code of height load condition is 4) is higher than normal load state (status code is 3), therefore in step S240, management servomechanism 320 is at first obtained the previous CPU operating parameter code (4) of setting in the specific node, and reduce the numerical value (for example CPU operating parameter code being reduced to 3 by 4) of CPU operating parameter code in the specific node, to promote the processing speed of CPU.
Comparatively speaking, utilize Fig. 6 examples description of step S230 at this.The load condition of supposing specific node is low load condition (status code is 2), and the operating parameter code of each CPU also is set as 4 in the specific node.Because the load condition of specific node (status code of low load condition is 2) is lower than normal load state (status code is 3), therefore in step S230, management servomechanism 320 is at first obtained the previous CPU operating parameter code (4) of setting in the specific node, and reduce the numerical value (for example CPU operating parameter code being promoted to 5 by 4) of CPU operating parameter code in the specific node, reducing the processing speed of CPU, and expect that the load condition of specific node can get back to the normal load state by this.
In view of this, but the method for managing power supply time-based of the embodiment of the invention interdependent (time locality) principle, in each node, detect the load condition that the average CPU usage that obtains dopes each node next time according to this, and then the operating parameter of CPU adjusted, all can get back in the normal load state with the average service rate of expecting each node next time.
But, if the load condition of specific node has been fully loaded load condition (status code is 5), and the operating parameter code of each CPU has been set to 1 in this specific node, that is each CPU all is positioned at full load condition, and this moment, above-mentioned method for managing power supply can't more adjust to make each node to get back to the normal load state.Or, if the load condition of specific node has been idle load condition (status code is 1), and the operating parameter code of each CPU has been set to 8 in this specific node, that is each CPU almost is positioned at idle state in the specific node, and this moment, above-mentioned method for managing power supply can't more adjust to make each node to get back to the normal load state.Therefore, the electric power management mechanism that closes in the past of embodiment of the invention constipation improves this method for managing power supply.
Please continue with reference to Fig. 2, in step S250, when the load condition of specific node has been idle load condition (status code is 1), and in this specific node the processing speed of each CPU be down to when minimum (that is, the running parameter code of CPU has been set to 8), when then the processing speed of the CPU of the specific node of expression reduction can't make the load condition of specific node get back to the normal load state, then enter step S270, management servomechanism 320 carries out the instant branching program of fictitious host computer VM, the load of being born with these nodes of balance PM.
When if the instant branching program of fictitious host computer VM still can't make the load condition of specific node get back to the normal load state, then carry out the node bolt down procedure, the instant branching program that cooperates fictitious host computer VM, fictitious host computer VM among some node PM is transferred among other nodes PM totally, and close and do not have fictitious host computer VM node PM therein, to reduce the quantity of node PM in the running.Comparatively speaking, in step S250, if when reducing the processing speed of the CPU of specific node and can make state get back to the normal load state, just can get back to step S225 to finish this method for managing power supply from step S250.And after execution in step S270 is complete, also get back to step S225 to finish this method for managing power supply.
Please continue with reference to figure 2, in step S260, when the load condition of specific node has been fully loaded load condition (status code is 5), and in this specific node the processing speed of each CPU be promoted to when the highest (that is, the running parameter code of CPU has been set to 1), then enter step S280, management servomechanism 320 just carries out the instant branching program of fictitious host computer VM, the load of being born with these nodes of balance PM.
When if the instant branching program of fictitious host computer VM still can't make the load condition of specific node get back to the normal load state, then carry out the node start-up routine to open/to restart the not node PM of running, and the instant branching program that cooperates fictitious host computer VM carries out transfer between the node PM with fictitious host computer VM according to its loading level, to save power supply under the work efficiency that does not affect fictitious host computer VM.And, be similar to step S250 and S270, in step S260, if when reducing the processing speed of the CPU of specific node and can make state get back to the normal load state, just can get back to step S225 to finish this method for managing power supply from step S260.And after execution in step S280 is complete, also get back to step S225 to finish this method for managing power supply.
In sum, since in the past electric power management mechanism can occupy in the process of implementation than multi-system resource, also expend the more processing time.By this, the embodiment of the invention can dynamically be adjusted according to the overall load (that is utilization rate of the CPU (central processing unit) of each node (CPU)) of each node the running speed of CPU (central processing unit) (CPU), so that each CPU can turn round in the normal load state, with in the work efficiency that does not affect fictitious host computer, do not occupy and save electrical source consumption under the system resource.
By this, instant transfer and node that the embodiment of the invention can be avoided regularly carrying out fictitious host computer are closed/the node startup, have exempted the reduction of high in the clouds operation efficiency.When adjusting the CPU rotating speed and still can't make the server system running in better consumption efficiency, the electric power management mechanism by in the past is to reach best power management effect again.
Although the present invention discloses as above with embodiment, so it is not to limit the present invention, and any person of an ordinary skill in the technical field when can doing a little change and retouching, and does not break away from the spirit and scope of the present invention.
Claims (10)
1. the method for managing power supply of a server system, wherein this server system comprises a plurality of nodes, the method for managing power supply of this server system comprises the following steps:
Detect in the specific node the corresponding at least one utilization rate of at least one CPU (central processing unit) to calculate an average service rate of this specific node, wherein this specific node one that is those nodes;
Judge a state of this specific node according to this average service rate;
When this state is lower than a normal load state, adjust an operating parameter of this CPU (central processing unit) in this specific node to reduce the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state; And
When this state is higher than this normal load state, adjust this operating parameter to promote the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state.
2. the method for managing power supply of server system according to claim 1 wherein also comprises the following steps:
When this operating parameter of adjustment can't make this state get back to this normal load state, carry out the instant branching program of a fictitious host computer, with the load of those nodes of balance.
3. the method for managing power supply of server system according to claim 2 wherein also comprises the following steps:
When the processing speed that reduces this CPU (central processing unit) can't make this state get back to this normal load state, carry out a node bolt down procedure and cooperate the instant branching program of this fictitious host computer, so that this state of each node in the running is got back to this normal load state.
4. the method for managing power supply of server system according to claim 2 wherein also comprises the following steps:
When the processing speed that promotes this CPU (central processing unit) can't make this state get back to this normal load state, carry out a node start-up routine and cooperate the instant branching program of this fictitious host computer, so that this state of each node in the running is got back to this normal load state.
5. the method for managing power supply of server system according to claim 1 judges that wherein this state of this specific node comprises the following steps:
When this average service rate is higher than a maximum CPU (central processing unit) threshold value, be a fully loaded load condition with this setting state;
When this average service rate is higher than a height CPU (central processing unit) threshold value and is lower than this maximum CPU (central processing unit) threshold value, be a height load condition with this setting state;
When this average service rate is higher than a low CPU (central processing unit) threshold value and is lower than this height CPU (central processing unit) threshold value, be this normal load state with this setting state;
When this average service rate is lower than this minuent CPU (central processing unit) threshold value, be this minuent load condition with this setting state of this CPU (central processing unit),
Wherein, this fully loaded load condition and this height load condition are higher than this normal load state, and should be lower than this normal load state by the minuent load condition.
6. the method for managing power supply of server system according to claim 5 judges that wherein this state of this specific node also comprises the following steps:
When this average service rate is lower than an idle CPU (central processing unit) threshold value, this setting state is the load condition that should leave unused, wherein should be lower than this minuent CPU (central processing unit) threshold value by idle CPU (central processing unit) threshold value, and should be lower than this normal load state by idle load condition.
7. the method for managing power supply of server system according to claim 1, wherein this operating parameter of this CPU (central processing unit) is an operation efficiency or a work period of this CPU (central processing unit).
8. server system comprises:
A plurality of nodes, each node is coupled to a network, and each node comprises at least one CPU (central processing unit);
One management servomechanism, be coupled to this network, this management servomechanism detects in the specific node the corresponding utilization rate of this CPU (central processing unit) to calculate an average service rate of this specific node, judge a state of this specific node according to this average service rate, when this state is lower than a normal load state, adjust an operating parameter of this CPU (central processing unit) in this specific node to reduce the processing speed of this CPU (central processing unit), and when this state is higher than this normal load state, adjust this operating parameter to promote the processing speed of this CPU (central processing unit), so that this state is got back to this normal load state, this specific node one that is those nodes wherein.
9. server system according to claim 8, when the processing speed that reduces this CPU (central processing unit) can't make this state get back to this normal load state, this management servomechanism carries out the instant branching program of a fictitious host computer and a node bolt down procedure, so that this state of each node in the running is got back to this normal load state.
10. server system according to claim 8, when the processing speed that promotes this CPU (central processing unit) can't make this state get back to this normal load state, this management servomechanism carries out a node start-up routine and cooperates the instant branching program of a fictitious host computer, so that this state of each node in the running is got back to this normal load state.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102962125A CN103034539A (en) | 2011-09-30 | 2011-09-30 | Server system and power management method thereof |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102962125A CN103034539A (en) | 2011-09-30 | 2011-09-30 | Server system and power management method thereof |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103034539A true CN103034539A (en) | 2013-04-10 |
Family
ID=48021462
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011102962125A Pending CN103034539A (en) | 2011-09-30 | 2011-09-30 | Server system and power management method thereof |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103034539A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331520A (en) * | 2014-11-28 | 2015-02-04 | 北京奇艺世纪科技有限公司 | Performance optimization method and device of Hadoop cluster and node state recognition method and device |
| CN104375621A (en) * | 2014-11-28 | 2015-02-25 | 广东石油化工学院 | Dynamic weighting load assessment method based on self-adaptive threshold values in cloud computing |
| CN104679213A (en) * | 2013-11-26 | 2015-06-03 | 英业达科技有限公司 | Server and server control method |
| CN105808344A (en) * | 2014-12-31 | 2016-07-27 | 环旭电子股份有限公司 | Network equipment with rapid throttling trigger mechanism |
| CN106610870A (en) * | 2016-12-28 | 2017-05-03 | 北京奇艺世纪科技有限公司 | Method and device for adjusting quantity of processing nodes |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1885232A (en) * | 2005-06-24 | 2006-12-27 | 戴尔产品有限公司 | Power management of multiple processors |
| CN101661327A (en) * | 2009-10-14 | 2010-03-03 | 中兴通讯股份有限公司 | Method for adjusting basic frequency of central processing unit (CPU) and device thereof |
| CN101770273A (en) * | 2008-12-29 | 2010-07-07 | 中兴通讯股份有限公司 | Method for realizing energy saving of system with a plurality of central processing units of server and device therefor |
-
2011
- 2011-09-30 CN CN2011102962125A patent/CN103034539A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1885232A (en) * | 2005-06-24 | 2006-12-27 | 戴尔产品有限公司 | Power management of multiple processors |
| CN101770273A (en) * | 2008-12-29 | 2010-07-07 | 中兴通讯股份有限公司 | Method for realizing energy saving of system with a plurality of central processing units of server and device therefor |
| CN101661327A (en) * | 2009-10-14 | 2010-03-03 | 中兴通讯股份有限公司 | Method for adjusting basic frequency of central processing unit (CPU) and device thereof |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104679213A (en) * | 2013-11-26 | 2015-06-03 | 英业达科技有限公司 | Server and server control method |
| CN104679213B (en) * | 2013-11-26 | 2017-08-25 | 英业达科技有限公司 | The control method of server and server |
| CN104331520A (en) * | 2014-11-28 | 2015-02-04 | 北京奇艺世纪科技有限公司 | Performance optimization method and device of Hadoop cluster and node state recognition method and device |
| CN104375621A (en) * | 2014-11-28 | 2015-02-25 | 广东石油化工学院 | Dynamic weighting load assessment method based on self-adaptive threshold values in cloud computing |
| CN104331520B (en) * | 2014-11-28 | 2018-08-07 | 北京奇艺世纪科技有限公司 | Hadoop clustering performances optimization method and device and node state recognition methods and device |
| CN105808344A (en) * | 2014-12-31 | 2016-07-27 | 环旭电子股份有限公司 | Network equipment with rapid throttling trigger mechanism |
| CN106610870A (en) * | 2016-12-28 | 2017-05-03 | 北京奇艺世纪科技有限公司 | Method and device for adjusting quantity of processing nodes |
| CN106610870B (en) * | 2016-12-28 | 2019-10-15 | 北京奇艺世纪科技有限公司 | A kind of processing number of nodes method of adjustment and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10664039B2 (en) | Power efficient processor architecture | |
| US8321693B2 (en) | Parallel processing method and system, for instance for supporting embedded cluster platforms, computer program product therefor | |
| KR101029414B1 (en) | Method and apparatus for providing for detecting processor state transitions | |
| US9235252B2 (en) | Dynamic balancing of power across a plurality of processor domains according to power policy control bias | |
| Zikos et al. | Performance and energy aware cluster-level scheduling of compute-intensive jobs with unknown service times | |
| EP2513751B1 (en) | System and method for controlling central processing unit power in a virtualized system | |
| US20130080809A1 (en) | Server system and power managing method thereof | |
| US8810584B2 (en) | Smart power management in graphics processing unit (GPU) based cluster computing during predictably occurring idle time | |
| CN102087619A (en) | Method and apparatus to improve turbo performance for events handling | |
| US10528119B2 (en) | Dynamic power routing to hardware accelerators | |
| CN103034539A (en) | Server system and power management method thereof | |
| Garg et al. | Task deadline-aware energy-efficient scheduling model for a virtualized cloud | |
| CN109491780B (en) | Multi-task scheduling method and device | |
| US20020184290A1 (en) | Run queue optimization with hardware multithreading for affinity | |
| CN110399034A (en) | A kind of power consumption optimization method and terminal of SoC system | |
| US20140282587A1 (en) | Multi-core binary translation task processing | |
| CN114710563A (en) | Cluster energy saving method and device | |
| Goubaa et al. | Scheduling periodic and aperiodic tasks with time, energy harvesting and precedence constraints on multi-core systems | |
| Sakamoto et al. | Analyzing resource trade-offs in hardware overprovisioned supercomputers | |
| Ramesh et al. | Energy management in embedded systems: Towards a taxonomy | |
| Fan et al. | GreenSleep: a multi-sleep modes based scheduling of servers for cloud data center | |
| CN102385529B (en) | Multi-CPU (Central Processing Unit) domain mobile electronic device and operating method thereof | |
| Richling et al. | Towards a Holistic Approach addressing the Energy/Performance Tradeoff in Multi-Core Systems |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130410 |