Energy-saving strategy of cloud data center based on adaptive service rate adjustment mechanism

Abstract

The invention relates to an energy-saving strategy of a cloud data center based on an adaptive service rate regulation mechanism, and the method comprises the steps: taking a physical machine in the cloud data center and a virtual machine deployed on the physical machine as a system, and setting a task counter and a sleep timer to monitor the state of the system, the method comprises the following steps: constructing an energy-saving strategy based on a self-adaptive service rate adjustment mechanism, depicting a working principle of the energy-saving strategy based on the self-adaptive service rate adjustment mechanism, evaluating a system performance index, analyzing a quantitative behavior of a cloud data center under the energy-saving strategy based on the self-adaptive service rate adjustment mechanism, and searching an optimal solution. And the power consumption of the cloud data center, the staying time of the task and the system loss rate are balanced. According to the method, the CPU clock frequency of the virtual machine is dynamically adjusted in real time according to the number of tasks in the system, the running speed of the virtual machine is changed, idle energy consumption can be effectively reduced, and the performance of the cloud data center is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of cloud computing technology, and in particular to an energy-saving strategy for a cloud data center based on an adaptive service rate adjustment mechanism. Background Art

[0002] In recent years, companies such as Huawei, Alibaba, Tencent, Google, and Amazon have invested huge amounts of money to build ultra-large-scale cloud data centers. Studies have shown that cloud data centers consume an alarming amount of electricity. The total amount of electricity consumed by cloud data centers worldwide will reach 13% of global electricity consumption. Although more and more large-scale data centers have been established, the resource utilization rate of cloud computing centers is very low (less than 10%). Cloud data centers consume a lot of energy. One of the important reasons is that the servers in cloud data centers are always awake. Therefore, putting idle servers to sleep during low-peak periods of cloud data center load and waking up dormant servers during peak periods of cloud data center load is one of the effective measures to save electricity. However, there is a trade-off between switching idle servers to sleep to save power and keeping servers awake to reduce service delays. Therefore, it is urgent to design energy-efficient solutions for cloud computing while ensuring user performance.

[0003] In addition, cloud computing provides resources based on virtualization technology. Usually, a number of virtual machines are deployed on a physical machine to process cloud services. In order to achieve a trade-off between energy consumption and performance, designing an energy-saving strategy with adjustable service rate has become an effective way for cloud data centers to improve their competitiveness. On the one hand, there are some random factors in the cloud data transmission process, such as communication noise and device failures. These random factors will change the arrival process and service process of cloud task requests; on the other hand, short-term business flows with high rates and large data volumes are common in cloud data centers. These micro-burst business flows with high burstiness consume most of the network bandwidth, resulting in serious deterioration of system performance, such as increased latency, increased packet loss rate, and reduced throughput. In existing research, Poisson processes are usually used to simulate the arrival of tasks without considering correlation characteristics. However, business flows in real cloud data centers generally have correlation characteristics, such as common micro-burst traffic.

[0004] In view of this, there is an urgent need to provide an energy-saving strategy for cloud data centers based on an adaptive service rate adjustment mechanism. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an energy-saving strategy for a cloud data center based on an adaptive service rate adjustment mechanism, which can realize real-time dynamic adjustment of the CPU clock frequency of the virtual machine according to the number of tasks in the system, change the operating rate of the virtual machine, reduce idle energy consumption by hibernation of the virtual machine, improve the performance of the cloud data center by service rate adjustment, and help cloud service providers select the optimal solution for system settings according to different service levels.

[0006] To solve the above problems, the technical solution adopted by the present invention is:

[0007] A cloud data center energy saving strategy based on an adaptive service rate adjustment mechanism includes:

[0008] Step S1: In a cloud data center, a plurality of virtual machines are deployed on a physical machine using virtualization technology, and the plurality of virtual machines share a buffer. The physical machine and the plurality of virtual machines deployed thereon are regarded as a system, and the virtual machines dynamically switch between a sleep state and a wake-up state as the load changes, and the service rate in the wake-up state;

[0009] Step S2, setting a task counter and a sleep timer: setting a task counter with two thresholds ξ and ξ for monitoring the number of tasks in the system, where ξ<ξ; setting a sleep timer, and starting a sleep timer with a random duration when the system enters a sleep state;

[0010] Step S3: Construct an energy-saving strategy based on an adaptive service rate adjustment mechanism:

[0011] Step S31, processing the task according to the current state of the system when the task reaches the system and whether the buffer overflows; wherein the current state is a dormant state or a wake-up state;

[0012] Step S32: In the sleep state, the system stops processing tasks to save energy. When the sleep timer times out, the system decides whether to remain in the sleep state or switch to the wake-up state according to the value of the task counter.

[0013] Step S33: In the wake-up state, the system dynamically adjusts the service rate of the virtual machine according to the value of the task counter;

[0014] Step S4, describing the working principle of the energy-saving strategy based on the adaptive service rate adjustment mechanism;

[0015] Step S5: evaluating the performance indicators under the energy-saving strategy based on the adaptive service rate adjustment mechanism;

[0016] Step S6, analyzing the quantitative behavior of the cloud data center under the energy-saving strategy based on the adaptive service rate adjustment mechanism: from the perspectives of cloud users and cloud service providers, respectively, the minimum system expected cost under the joint optimality of the sleep parameter α and the threshold ξ is obtained, thereby weighing the overall power consumption of the cloud data center and the service quality of the task.

[0017] As an implementation mode of the present invention, step S31 includes:

[0018] When a task arrives at the system and the system is in sleep state, if the buffer overflows, the arrived task is discarded; if the buffer does not overflow, the arrived task is queued in the buffer;

[0019] When a task arrives at the system and the system is in the awake state, if the buffer overflows, the arriving task is discarded; if the buffer does not overflow and at least one virtual machine is idle, the arriving task is assigned to any idle virtual machine; if the buffer does not overflow and there is no idle virtual machine, the arriving task is queued in the buffer.

[0020] As an implementation of the present invention, in step S32, the system determines whether to maintain the sleep state or switch to the wake-up state according to the value of the task counter as follows:

[0021] The system monitors the value of the task counter. If the value of the task counter is 0, the system remains in sleep mode and starts a new sleep timer. If the value of the task counter is greater than 0 and less than ξ, the system switches to wake-up state I, the virtual machines occupied by the task run at a normal rate, and the virtual machines not occupied by the task are in an idle state. If the value of the task counter reaches ξ, the system switches to wake-up state II and runs at a high rate.

[0022] As an implementation mode of the present invention, step S33 includes:

[0023] Step S331: In the wake-up state I, all tasks are served at a normal rate, and the task counter is monitored in real time. If the value of the task counter rises to ξ, the system switches to the wake-up state II and runs at a high rate; if all tasks are processed and the system is idle, the system switches to the sleep state and starts a sleep timer with a random duration;

[0024] Step S332: In wake-up state II, all tasks are served at a high rate and the value of the task counter is continuously monitored; when the value of the task counter drops to , the system enters the awake state I and all tasks are serviced at normal rates.

[0025] As an implementation mode of the present invention, step S4 includes:

[0026] S41,The arrival process of cloud services in a random cloud environment is described as an R-phase Markov arrival process (R-MAP) with environment dependence;

[0027] In wake-up state I and wake-up state II, the service time of the task follows the exponential distribution with parameters μ1 and μ2 respectively, and μ1<μ2, and the sleep cycle follows the exponential distribution with parameter α. A R-MAP / M / N / N+K queuing model with multi-server synchronous multiple vacations and adaptive service rate adjustment mechanism is constructed, where N is the number of virtual machines deployed on the physical machine, and N<ξ, and K is the buffer size;

[0028] S42. The random factors in the cloud environment are described as the state space E r ={1,2,...,R}, the infinitesimal generator is a homogeneous irreducible continuous-time Markov chain {r(t),t≥0}, H is described as follows:

[0029]

[0030] S43, in the environmental phase r, r∈E r , the arrival of the task obeys the infinitesimal representative matrix: The arrival of tasks is guided by the random process {ν(t), t≥0}, where {ν(t), t≥0} is a homogeneous irreducible continuous-time Markov chain, called the underlying Markov chain of MAP, and its state space is E ν = {0,1,...,W}, the average arrival rate of tasks in phase r is λ (r) ;

[0031] S44. Considering the random behavior of microburst services in the cloud environment, an irreducible four-dimensional continuous-time Markov chain with a state space of Ω and an infinitesimal generator of Q is constructed:

[0032] {x(t),y(t),h(t),z(t),t≥0};

[0033] Where x(t)∈{0,1,2} is the state of the virtual machine at time t, t≥0, x(t)=0 indicates that the virtual machine is in sleep state, x(t)=1 indicates that the virtual machine is in wake-up state I running at normal rate, and x(t)=2 indicates that the virtual machine is in wake-up state II running at high rate; y(t)∈{0,1,...,N+K} is the number of tasks in the queuing system, including the services being processed in the virtual machine and the services waiting in the cache, also called the system level; h(t)∈E r is the environmental phase of the random cloud environment at time t; z(t)∈E ν , is the state of MAP at time t;

[0034] Ω={(j,i,r,v)|j∈{0,1,2},i∈{0,1,...,N+K},r∈E r ,v∈E v};

[0035] S45. Using the quasi-birth-death process and Gauss-Seidel method, the steady-state distribution Π of the system is obtained:

[0036] Π=(π0,π1,π2);

[0037] Among them, π j is the steady-state probability of the virtual machine in state j, and satisfies

[0038] π0=(π 0，0 ,π 0，1 ,...,π 0,K );

[0039] π1=(π 1，1 ,π 1，2 ,...,π 1,ξ-1 );

[0040]

[0041] Among them, π 0,i is the steady-state probability vector of the virtual machine state being in sleep state and the system level being at i, i = {0, 1, ..., K}, π 0,i It is expressed as follows:

[0042] π 0,i =(π 0,i,1,0 ,π 0,i,1,1 ,...,π 0,i,1,W ,π 0,i,2,0 ,π 0,i,2,1 ,...,π 0,i,2,W ,...,π 0,i,R,0 ,π 0,i,R,1 ,...,π 0,i,R,W );

[0043] π 1,i is the steady-state probability vector of the virtual machine state being in awake state I and the system level being in i, i = {1, 2, ..., ξ-1}, π 1,i It is expressed as follows:

[0044] π 1,i =(π 1,i,1,0 ,π 1,i,1,1 ,...,π 1,i,1,W ,π 1,i,2,0 ,π 1,i,2,1 ,...,π 1,i,2,W ,...,π 1,i,R,0,π 1,i,R,1 ,...,π 1,i,R,W ).

[0045] π 2,i The virtual machine state is in wake-up state II and the system level is in i. The steady-state probability vector, π 2,i It is expressed as follows:

[0046] π 2,i =(π 2,i,1,0 ,π 2,i,1,1 ,...,π 2,i,1,W ,π 2,i,2,0 ,π 2,i,2,1 ,...,π 2,i,2,W ,...,π 2,i,R,0 ,π 2,i,R,1 ,...,π 2,i,R,W ).

[0047] As an implementation mode of the present invention, step S5 includes:

[0048] S51. From the perspective of cloud users, solve the service quality indicators related to service quality. The service quality indicators include: the average stay time E[T] of the task and the system loss rate φ;

[0049] (1) Use the following formula to calculate the average stay time E[T] of the task:

[0050]

[0051] where λ eff is the effective arrival rate, λ eff The calculation is as follows:

[0052]

[0053] in, μ1=f1 run / (mθ), where f1 run , is the clock frequency of the virtual machine in normal rate operation state and high rate operation state, m is the average data size of the task, and θ is the number of cycles required for the virtual machine CPU to process 1 bit of data;

[0054] (2) Use the following formula to solve the system loss rate φ:

[0055]

[0056] in,

[0057] S52. From the perspective of the cloud service provider, the following formula is used to solve the system energy consumption indicator system power consumption γ:

[0058]

[0059] Among them, P0 sle , P1 idle , P1 run , P2 run It is the CPU dynamic power consumption of the virtual machine in sleep state, idle state, normal speed running state and high speed running state;

[0060]

[0061] Among them, f0 sle , f1 idle , f1 run , f2 run The clock frequency of the virtual machine in the sleep state, idle state, normal speed running state and high speed running state. is a constant related to the chip specifications of the physical machine, and ∈ is an adaptive coefficient related to power consumption.

[0062] As an implementation mode of the present invention, step S6 includes:

[0063] S61. Establish a cost function as the expected cost of the system to balance the performance indicators:

[0064] F(α,ξ)=C1γ+C2E[T]+C3φ;

[0065] Among them, C1 is the system power consumption weight factor, C2 is the task average stay time weight factor, and C3 is the system loss rate weight factor;

[0066] S62. Using the improved Firehawk optimization algorithm, the minimum system expected cost under the joint optimality of the sleep parameter α and the threshold ξ is obtained, and this is taken as the optimal solution.

[0067] The beneficial effects of adopting the above technical solution are:

[0068] The present invention provides an energy-saving strategy for an adaptive service rate regulation mechanism in a cloud data center, and the method includes: regarding a physical machine in the cloud data center and a virtual machine deployed on the physical machine as a system, setting a task counter and a sleep timer to monitor the system status, constructing an energy-saving strategy based on the adaptive service rate regulation mechanism, characterizing the working principle of the energy-saving strategy based on the adaptive service rate regulation mechanism, evaluating system performance indicators, analyzing the quantitative behavior of the cloud data center under the energy-saving strategy based on the adaptive service rate regulation mechanism, finding the optimal solution, and weighing the power consumption of the cloud data center, the average stay time of the task, and the system loss rate.

[0069] The energy-saving strategy of the adaptive service rate adjustment mechanism in the cloud data center provided by the present invention aims at the problem of the sharp increase in energy consumption caused by the rapid growth of the global data center industry. It considers switching the idle virtual machine state to the sleep mode according to the system load change of the server of the cloud data center to effectively reduce the idle energy consumption, thereby reducing the energy consumption of the cloud data center; considering the micro-burst of cloud business flow, the service rate is changed by changing the CPU clock frequency of the virtual machine to improve the service quality. In order to balance the service quality and energy consumption level of the task, an energy-saving strategy based on the adaptive service rate adjustment mechanism is proposed, and a queuing model based on the multi-phase Markov arrival process is constructed for the energy-saving strategy.

[0070] The present invention designs an energy-saving strategy for an adaptive service rate regulation mechanism in a cloud data center. The strategy adopts a quasi-birth-death process and a Gauss-Seidel method to obtain indicators such as the energy consumption of the system in a steady state, the average residence time of tasks, and the loss rate of the system. By weighing the power consumption of the system, the average residence time of tasks, and the system loss rate, the improved Firehawk algorithm is used to propose the best solution for cloud service providers, thereby helping cloud service providers to choose the solution that represents the best compromise between different objectives according to their preferences. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is a schematic diagram of the system architecture of the energy-saving strategy of the cloud data center based on the adaptive rate adjustment mechanism of the present invention;

[0072] Figure 2 It is a flow chart of the energy-saving strategy of the cloud data center based on the adaptive rate adjustment mechanism of the present invention. DETAILED DESCRIPTION

[0073] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is clearly and completely described below in conjunction with specific embodiments.

[0074] The embodiment of the present invention provides an energy-saving strategy for a cloud data center based on an adaptive service rate adjustment mechanism, such as Figure 1 and Figure 2 As shown, the method includes:

[0075] Step S1: In a cloud data center, a plurality of virtual machines are deployed on a physical machine using virtualization technology, and the plurality of virtual machines share a buffer. The physical machine and the plurality of virtual machines deployed thereon are regarded as a system, and the virtual machines dynamically switch between a sleep state and a wake-up state as the load changes, and the service rate in the wake-up state;

[0076] The task of obtaining services in the cloud is first submitted to the cloud switch and then assigned to the virtual machines deployed on the physical machines.

[0077] Step S2, setting a task counter and a sleep timer: setting a task counter with two thresholds ξ and ξ to monitor the number of tasks in the system, where ξ<ξ; setting a sleep timer, when the system enters the sleep state, starting a sleep timer of a random duration; the system wakes up after several consecutive sleep cycles.

[0078] Step S3: Construct an energy-saving strategy based on an adaptive service rate adjustment mechanism:

[0079] Step S31, processing the task according to the current state of the system when the task reaches the system and whether the buffer overflows; wherein the current state is a dormant state or a wake-up state;

[0080] Specifically, when a task arrives at the system and the system is in sleep state, if the buffer overflows, the arriving task is discarded; if the buffer does not overflow, the arriving task is queued in the buffer;

[0081] When a task arrives at the system and the system is in the awake state, if the buffer overflows, the arriving task is discarded; if the buffer does not overflow and at least one virtual machine is idle, the arriving task is assigned to any idle virtual machine; if the buffer does not overflow and there is no idle virtual machine, the arriving task is queued in the buffer.

[0082] Step S32: In the sleep state, the system stops processing tasks to save energy. When the sleep timer times out, the system monitors the value of the task counter and decides whether to maintain the sleep state or switch to the wake-up state according to the value of the task counter.

[0083] The system decides whether to remain in the sleep state or switch to the wake-up state according to the value of the task counter as follows:

[0084] If the value of the task counter is 0, the system remains in sleep state and starts a new sleep timer; if the value of the task counter is greater than 0 and less than ξ, the system switches to wake-up state I, the virtual machines occupied by the task run at a normal rate, and the virtual machines not occupied by the task are in idle state; if the value of the task counter reaches ξ, the system switches to wake-up state II and runs at a high rate.

[0085] Among them, for any task, the service time of the task follows an exponential distribution with parameter μ, and its corresponding service rate is equal to the parameter of the exponential distribution, that is, the service rate is also μ. In the present invention, the normal service rate is μ1, and the high speed rate is μ2. μ1 and μ2 are calculated from the clock frequency of the virtual machine, and the calculation formula is: μ1=f1 run / (mθ), where f1 run , is the clock frequency of the virtual machine in the normal rate running state and the high rate running state, m is the average data size of the task, and θ is the number of cycles required for the virtual machine CPU to process 1 bit of data; the clock frequency of the virtual machine at the normal service rate and the high rate can be obtained according to the prior art, and the embodiment of the present invention does not make any specific limitation on this.

[0086] Step S33: In the wake-up state, the system dynamically adjusts the service rate of the virtual machine according to the value of the task counter;

[0087] This step includes:

[0088] Step S331: In the wake-up state I, all tasks are served at a normal rate, and the task counter is monitored in real time. If the value of the task counter rises to ξ, the system switches to the wake-up state II and runs at a high rate; if all tasks are processed and the system is idle, the system switches to the sleep state and starts a sleep timer with a random duration;

[0089] Step S332: In wake-up state II, all tasks are served at a high rate and the value of the task counter is continuously monitored; when the value of the task counter drops to ξ, the system enters wake-up state I and all tasks are served at a normal rate. This can prevent the system from jittering due to frequent switching rates.

[0090] Step S4, describing the working principle of the energy-saving strategy based on the adaptive service rate adjustment mechanism;

[0091] S41,The arrival process of cloud services in a random cloud environment is described as an R-phase Markov arrival process (R-MAP) with environment dependence;

[0092] As mentioned above, the normal service rate is μ1 and the high-speed rate is μ2. Therefore, in wake-up state I and wake-up state II, the service time of the task obeys the exponential distribution with parameters μ1 and μ2 respectively, and μ1<μ2. The sleep cycle follows the exponential distribution with parameter α. A R-MAP / M / N / N+K queuing model with multi-server synchronous multiple vacations and adaptive service rate adjustment mechanism is constructed, where N is the number of virtual machines deployed on the physical machine, and N<ξ, and K is the buffer size;

[0093] S42. The random factors in the cloud environment are described as the state space E r = {1,2,…,R}, the infinitesimal generator is a homogeneous irreducible continuous-time Markov chain {r(t),t≥0}, H is described as follows:

[0094]

[0095] S43, in the environmental phase r, r∈E r , the arrival of the task obeys the infinitesimal representative matrix: The arrival of tasks is guided by the random process {ν(t), t≥0}, where {ν(t), t≥0} is a homogeneous irreducible continuous-time Markov chain, called the underlying Markov chain of MAP, and its state space is E ν = {0,1,...,W}, the average arrival rate of tasks in phase r is λ (r) ;

[0096] S44. Considering the random behavior of microburst services in the cloud environment, an irreducible four-dimensional continuous-time Markov chain with a state space of Ω and an infinitesimal generator of Q is constructed:

[0097] {x(t),y(t),h(t),z(t),t≥0};

[0098] Where x(t)∈{0,1,2} is the state of the virtual machine at time t, t≥0, x(t)=0 indicates that the virtual machine is in sleep state, x(t)=1 indicates that the virtual machine is in wake-up state I running at normal rate, and x(t)=2 indicates that the virtual machine is in wake-up state II running at high rate; y(t)∈{0,1,...,N+K} is the number of tasks in the queuing system, including the business being processed in the virtual machine and the business waiting in the cache, also known as the system level; h(t)∈E r is the environmental phase of the random cloud environment at time t; z(t)∈E ν , is the state of MAP at time t;

[0099] Ω={(j,i,r,v)|j∈{0,1,2},i∈{0,1,...,N+K},r∈E r ,v∈E v};

[0100] S45. Using the quasi-birth-death process and Gauss-Seidel method, the steady-state distribution Π of the system is obtained:

[0101] Π=(π0,π1,π2);

[0102] Among them, π j is the steady-state probability of the virtual machine in state j, and satisfies:

[0103] π0=(π 0，0 ,π 0，1 ,...,π 0,K );

[0104] π1=(π 1，1 ,π 1，2 ,...,π1,ξ-1 );

[0105]

[0106] Among them, π 0,i is the steady-state probability vector of the virtual machine state being in sleep state and the system level being at i, i = {0, 1, ..., K}, π 0,i It is expressed as follows:

[0107] π 0,i =(π 0,i,1,0 ,π 0,i,1,1 ,...,π 0,i,1,W ,π 0,i,2,0 ,π 0,i,2,1 ,...,π 0,i,2,W ,...,π 0,i,R,0 ,π 0,i,R,1 ,...,π 0,i,R,W );

[0108] π 1,i is the steady-state probability vector of the virtual machine state being in awake state I and the system level being in i, i = {1, 2, ..., ξ-1}, π 1,i It is expressed as follows:

[0109] π 1,i =(π 1,i,1,0 ,π 1,i,1,1 ,...,π 1,i,1,W ,π 1,i,2,0 ,π 1,i,2,1 ,...,π 1,i,2,W ,...,π 1,i,R,0 ,π 1,i,R,1 ,...,π 1,i,R,W ).

[0110] π 2,i The virtual machine state is in wake-up state II and the system level is in i. The steady-state probability vector, π 2,i It is expressed as follows:

[0111] π 2,i =(π 2,i,1,0 ,π 2,i,1,1 ,...,π 2,i,1,W ,π 2,i,2,0 ,π 2,i,2,1 ,...,π 2,i,2,W ,...,π 2,i,R,0 ,π 2,i,R,1 ,...,π 2,i,R,W )

[0112] S5. Evaluate the performance indicators under the energy-saving strategy based on the adaptive service rate adjustment mechanism:

[0113] It includes:

[0114] S51. From the perspective of cloud users, solve the service quality indicators related to service quality. The service quality indicators include: the average stay time E[T] of the task and the system loss rate φ;

[0115] (1) Use the following formula to calculate the average stay time E[T] of the task:

[0116]

[0117] where λ eff is the effective arrival rate, λ eff The calculation is as follows:

[0118]

[0119] in, μ1=f1 run / (mθ), where f1 run , is the clock frequency of the virtual machine in normal rate operation state and high rate operation state, m is the average data size of the task, and θ is the number of cycles required for the virtual machine CPU to process 1 bit of data;

[0120] (2) Use the following formula to solve the system loss rate φ:

[0121]

[0122] in,

[0123] S52. From the perspective of the cloud service provider, the following formula is used to solve the system energy consumption indicator system power consumption γ:

[0124]

[0125] Among them, P0 sle , P1 idle , P1 run , It is the CPU dynamic power consumption of the virtual machine in sleep state, idle state, normal speed running state and high speed running state;

[0126]

[0127] Among them, P0 sle , f1 idle , f1 run , The clock frequency of the virtual machine in the sleep state, idle state, normal speed running state and high speed running state. is a constant related to the physical machine chip specifications, and ∈ is an adaptive coefficient related to power consumption.

[0128] S6. Analyze the quantitative behavior of the cloud data center under the energy-saving strategy based on the adaptive service rate adjustment mechanism:

[0129] S61. Establish a cost function as the expected cost of the system to balance the performance indicators:

[0130] F(α,ξ)=C1γ+C2E[T]+C3φ;

[0131] Among them, C1 is the system power consumption weight factor, C2 is the task average stay time weight factor, and C3 is the system loss rate weight factor. Specifically, the system power consumption is considered from the perspective of the cloud service provider, and the task average stay time and system loss rate are considered from the perspective of the service quality obtained by the cloud user.

[0132] In actual applications, cloud service providers can set weight factors of different sizes based on target preferences.

[0133] S62. Using the improved Firehawk optimization algorithm, the minimum system expected cost under the joint optimality of the sleep parameter α and the threshold ξ is obtained, and this is taken as the optimal solution.

[0134] Therefore, cloud service providers can balance service quality and energy consumption costs to the greatest extent possible to obtain the minimum expected cost and enhance their market competitiveness.

[0135] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An energy-saving strategy for a cloud data center based on an adaptive service rate adjustment mechanism, characterized in that: It includes: Step S1: In a cloud data center, a plurality of virtual machines are deployed on a physical machine using virtualization technology, and the plurality of virtual machines share a buffer. The physical machine and the plurality of virtual machines deployed thereon are regarded as a system, and the virtual machines dynamically switch between a sleep state and a wake-up state as the load changes, and the service rate in the wake-up state; Step S2, setting the task counter and sleep timer: setting a timer with two thresholds ξ and The task counter is used to monitor the number of tasks in the system, where Set a sleep timer. When the system enters sleep mode, start a sleep timer with a random duration. Step S3: Construct an energy-saving strategy based on an adaptive service rate adjustment mechanism: Step S31, processing the task according to the current state of the system when the task reaches the system and whether the buffer overflows; wherein the current state is a dormant state or a wake-up state; Step S32: In the sleep state, the system stops processing tasks to save energy. When the sleep timer times out, the system decides whether to remain in the sleep state or switch to the wake-up state according to the value of the task counter. Step S33: In the wake-up state, the system dynamically adjusts the service rate of the virtual machine according to the value of the task counter; Step S4, describing the working principle of the energy-saving strategy based on the adaptive service rate adjustment mechanism; Step S5: evaluating the performance indicators under the energy-saving strategy based on the adaptive service rate adjustment mechanism; Step S6, analyzing the quantitative behavior of the cloud data center under the energy-saving strategy based on the adaptive service rate adjustment mechanism: from the perspectives of cloud users and cloud service providers, respectively, the minimum system expected cost under the joint optimality of the sleep parameter α and the threshold ξ is obtained, thereby weighing the overall power consumption of the cloud data center and the service quality of the task.

2. According to claim 1, a cloud data center energy saving strategy based on an adaptive service rate adjustment mechanism is characterized in that: Step S31 includes: When a task arrives at the system and the system is in sleep state, if the buffer overflows, the arrived task is discarded; if the buffer does not overflow, the arrived task is queued in the buffer; When a task arrives at the system and the system is in the awake state, if the buffer overflows, the arriving task is discarded; if the buffer does not overflow and at least one virtual machine is idle, the arriving task is assigned to any idle virtual machine; if the buffer does not overflow and there is no idle virtual machine, the arriving task is queued in the buffer.

3. According to claim 1, a cloud data center energy saving strategy based on an adaptive service rate adjustment mechanism is characterized in that: In step S32, the system determines whether to maintain the sleep state or switch to the wake-up state according to the value of the task counter as follows: The system monitors the value of the task counter. If the value of the task counter is 0, the system remains in sleep mode and starts a new sleep timer. If the value of the task counter is greater than 0 and less than ξ, the system switches to wake-up state I, the virtual machines occupied by the task run at a normal rate, and the virtual machines not occupied by the task are in an idle state. If the value of the task counter reaches ξ, the system switches to wake-up state II and runs at a high rate.

4. According to claim 1, a cloud data center energy saving strategy based on an adaptive service rate adjustment mechanism is characterized in that: Step S33 includes: Step S331: In the wake-up state I, all tasks are served at a normal rate, and the task counter is monitored in real time. If the value of the task counter rises to ξ, the system switches to the wake-up state II and runs at a high rate; if all tasks are processed and the system is idle, the system switches to the sleep state and starts a sleep timer with a random duration; Step S332: In wake-up state II, all tasks are served at a high rate and the value of the task counter is continuously monitored; when the value of the task counter drops to , the system enters the awake state I and all tasks are serviced at normal rates.

5. According to claim 1, a cloud data center energy saving strategy based on an adaptive service rate adjustment mechanism is characterized in that: Step S4 includes: S41,The arrival process of cloud services in a random cloud environment is described as an R-phase Markov arrival process (R-MAP) with environment dependence; In wake-up state I and wake-up state II, the service time of the task follows the exponential distribution with parameters μ1 and μ2 respectively, and μ1<μ2, and the sleep cycle follows the exponential distribution with parameter α. A R-MAP / M / N / N+K queuing model with multi-server synchronous multiple vacations and adaptive service rate adjustment mechanism is constructed, where N is the number of virtual machines deployed on the physical machine, and N<ξ, and K is the buffer size; S42. The random factors in the cloud environment are described as the state space E r = {1,2,…,R}, the infinitesimal generator is a homogeneous irreducible continuous-time Markov chain {r(t),t≥0}, H is described as follows: S43, in the environmental phase r, r∈E r , the arrival of the task obeys the infinitesimal representative matrix: The arrival of tasks is guided by the random process {ν(t), t≥0}, where {ν(t), t≥0} is a homogeneous irreducible continuous-time Markov chain, called the underlying Markov chain of MAP, and its state space is E ν = {0,1,...,W}, the average arrival rate of tasks in phase r is λ (r) ; S44. Considering the random behavior of microburst services in the cloud environment, an irreducible four-dimensional continuous-time Markov chain with a state space of Ω and an infinitesimal generator of Q is constructed: {x(t),y(t),h(t),z(t),t≥0}; Where x(t)∈{0,1,2} is the state of the virtual machine at time t, t≥0, x(t)=0 indicates that the virtual machine is in sleep state, x(t)=1 indicates that the virtual machine is in wake-up state I running at normal rate, and x(t)=2 indicates that the virtual machine is in wake-up state II running at high rate; y(t)∈{0,1,...,N+K} is the number of tasks in the queuing system, including the services being processed in the virtual machine and the services waiting in the cache, also called the system level; h(t)∈E r is the environmental phase of the random cloud environment at time t; z(t)∈E ν , is the state of MAP at time t; Ω={(j,i,r,v)j∈{0,1,2},i∈{0,1,...,N+K},r∈E r ,v∈E v }; S45. Using the quasi-birth-death process and Gauss-Seidel method, the steady-state distribution Π of the system is obtained: Π=(π0,π1,π2); Among them, π j is the steady-state probability of the virtual machine in state j, and satisfies π0＝(π 0，0 ,π 0，1 ,...,π 0,K )； π1＝(π 1，1 ,π 1，2 ,...,π 1,ξ-1 )； Among them, π 0,i is the steady-state probability vector of the virtual machine state being in sleep state and the system level being at i, i = {0, 1, ..., K}, π 0,i It is expressed as follows: π 0,i ＝(π 0,i,1,0 ,π 0,i,1,1 ,...,π 0,i,1,W ,π 0,i,2,0 ,π 0,i,2,1 ,...,π 0,i,2,W ,...，,π 0,i,R,0 ,π 0,i,R,1 ,...,π 0,i,R,W )； π 1,i is the steady-state probability vector of the virtual machine state being in awake state I and the system level being in i, i = {1, 2, ..., ξ-1}, π 1,i It is expressed as follows: π 1,i ＝(π 1,i,1,0 ,π 1,i,1,1 ,...,π 1,i,1,W ,π 1,i,2,0 ,π 1,i,2,1 ,...,π 1,i,2,W ,...,π 1,i,R,0 ,π 1,i,R,1 ,...,π 1,i,R,W )； π 2,i The virtual machine state is in wake-up state II and the system level is in i. The steady-state probability vector, π 2,i It is expressed as follows: π 2,i ＝(π 2,i,1,0 ,π 2,i,1,1 ,...,π 2,i,1,W ,π 2,i,2,0 ,π 2,i,2,1 ,...,π 2,i,2,W ,...,π 2,i,R,0 ,π 2,i,R,1 ,...,π 2,i,R,W )。 6. The energy-saving strategy of a cloud data center based on an adaptive service rate adjustment mechanism according to claim 1, characterized in that: Step S5 includes: S51. From the perspective of cloud users, solve the service quality indicators related to service quality. The service quality indicators include: the average stay time E[T] of the task and the system loss rate φ; (1) Use the following formula to calculate the average stay time E[T] of the task: where λ eff is the effective arrival rate, λ eff The calculation is as follows: in, in is the clock frequency of the virtual machine in normal rate operation state and high rate operation state, m is the average data size of the task, and θ is the number of cycles required for the virtual machine CPU to process 1 bit of data; (2) Use the following formula to solve the system loss rate φ: in, S52. From the perspective of the cloud service provider, the following formula is used to solve the system energy consumption indicator system power consumption γ: in, It is the CPU dynamic power consumption of the virtual machine in sleep state, idle state, normal speed running state and high speed running state; in, The clock frequency of the virtual machine in the sleep state, idle state, normal speed running state and high speed running state. is a constant related to the chip specifications of the physical machine, and ∈ is an adaptive coefficient related to power consumption.

7. The energy-saving strategy of a cloud data center based on an adaptive service rate adjustment mechanism according to claim 1, characterized in that: Step S6 includes: S61. Establish a cost function as the expected cost of the system to balance the performance indicators: F(α,ξ)=C1γ+C2E[T]+C3φ; Among them, C1 is the system power consumption weight factor, C2 is the task average stay time weight factor, and C3 is the system loss rate weight factor; S62. Using the improved Firehawk optimization algorithm, the minimum system expected cost under the joint optimality of the sleep parameter α and the threshold ξ is obtained, and this is taken as the optimal solution.

Images