CN111338757A - Energy-optimized virtual machine deployment method and system - Google Patents

Energy-optimized virtual machine deployment method and system Download PDF

Info

Publication number
CN111338757A
CN111338757A CN202010112558.4A CN202010112558A CN111338757A CN 111338757 A CN111338757 A CN 111338757A CN 202010112558 A CN202010112558 A CN 202010112558A CN 111338757 A CN111338757 A CN 111338757A
Authority
CN
China
Prior art keywords
virtual machine
goblet
deployment
fitness
load
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
Application number
CN202010112558.4A
Other languages
Chinese (zh)
Inventor
张小庆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan Polytechnic University
Original Assignee
Wuhan Polytechnic University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Wuhan Polytechnic University filed Critical Wuhan Polytechnic University
Priority to CN202010112558.4A priority Critical patent/CN111338757A/en
Publication of CN111338757A publication Critical patent/CN111338757A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Feedback Control In General (AREA)

Abstract

An energy optimized virtual machine deployment method and system are disclosed. The method comprises the following steps: step 1: determining a virtual machine deployment target and constraint conditions thereof; step 2: determining a deployment solution calculation initialization parameter; and step 3: performing population initialization to enable the deployment solution of the goblet sea squirts to correspond to the deployment solution of the virtual machines one by one; and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof; and 5: determining a food source, a leader and a follower to obtain an updated population; step 6: calculating the fitness of each individual of the ascidians in the updated population, and determining the food source; and 7: and 5-6, outputting the position of the food source in the current population as the final optimal solution for deploying the virtual machine. According to the invention, exploration behaviors and development behaviors in an iterative process are balanced through a group optimization model of the goblet sea squirts, and the deployment effect of the energy-optimized virtual machine is realized while the global property and diversity are ensured.

Description

Energy-optimized virtual machine deployment method and system
Technical Field
The invention relates to the field of cloud computing and intelligent group algorithms, in particular to a virtual machine deployment method and system for energy optimization.
Background
Cloud computing can overcome the defects of a localized computing mode of a traditional computing mode for a data center owner through the guarantee of flexibility, safety, high expandability and service quality. A large number of heterogeneous servers of different types can be configured, a large number of virtual machines can be further deployed on the servers by utilizing a virtualization technology, and application isolation during service provision is realized by the aid of the virtual machines. Virtual machines also have different load types and different demands on resources in each dimension, which may lead to a contradiction between resource utilization and high energy consumption of the host. The high energy consumption of the host computer may result in an increase in the operating cost of the cloud computing, and may also have an adverse effect on the environment. The high power consumption is closely related to the utilization rate of the host resources. Therefore, how to perform energy-efficient virtual machine deployment will be a problem that cloud computing must solve.
The intelligent colony algorithm is a typical algorithm for solving the optimization problem, wherein a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a bee colony algorithm and a gravity search algorithm are all applied to a virtual machine deployment problem in a cloud computing environment, an energy consumption problem has many stage achievements as one optimization target, but due to inherent limitations of the meta-heuristic algorithm, such as the fact that the meta-heuristic algorithm is trapped in local optimization and depends on excessive initial parameters, the virtual machine deployment optimization problem based on the energy consumption has certain limitation, and the meta-heuristic algorithm with better performance is needed for optimizing and solving the problem. Therefore, there is a need to develop an energy-optimized virtual machine deployment method and system.
The information disclosed in this background section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or any form of suggestion that this information forms the prior art already known to a person skilled in the art.
Disclosure of Invention
The invention provides an energy-optimized virtual machine deployment method and system, which can fully balance exploration behaviors and development behaviors in an iterative process of a group of goblet ascidians through a group optimization model of the goblet ascidians, and realize an energy-optimized virtual machine deployment effect while ensuring the global property and diversity.
According to an aspect of the present invention, an energy-optimized virtual machine deployment method is presented. The method may include: step 1: determining a virtual machine deployment target and constraint conditions thereof; step 2: determining a deployment solution calculation initialization parameter; and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one; and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof; and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population; step 6: calculating the fitness of each individual goblet ascidian in the updated population, determining the individual goblet ascidian with the largest fitness value, judging whether the fitness value of the individual goblet ascidian is greater than the fitness value of the food source, and if so, taking the individual goblet ascidian as a new food source; if not, the original food source is reserved; and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
Preferably, the calculating the initialization parameters of the deployment solution comprises: maximum number of iterations TmaxThe method comprises the following steps of A, a virtual machine set V, a physical host set H, a resource capacity vector of a host, a resource request vector of a virtual machine, and full power consumption and idle power consumption of the host.
Preferably, the virtual machine deployment targets are:
Figure BDA0002390529530000021
wherein, Pi,fullIndicates the host hiPower consumption in the fully loaded state, Pi,idleIndicates the host hiPower consumption in idle state, PiIndicates the host hiCurrent power consumption of Ui,CPUIs a main machine hiCPU utilization of.
Preferably, the constraint condition is:
Loadi,CPU≤Ci,CPU,i＝1,2,...,m (2)
Loadi,MEM≤Ci,MEM,i＝1,2,...,m (3)
Loadi,DISK≤Ci,DISK,i＝1,2,...,m (4)
Loadi,NETW≤Ci,NETW,i＝1,2,...,m (5)
wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiIn (2)Storage capacity, Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability, Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe overall memory load.
Preferably, the step (3) includes: randomizing the position of the individual goblet sea squirt by formula (6) to obtain an initialization population:
Figure BDA0002390529530000031
wherein q is 1,2, …, N, N represents the size of the ascidian goblet population, j is 1,2, …, N, N represents the total amount of virtual machines, c4Represents the interval [0,1]Inner random number, ubj,maxDenotes an upper limit value, lb, of a j-th spacej,minRepresenting a lower bound, x, of a j-space1,jRepresenting the leader's position in the j-dimensional space, xq,jRepresenting the position of the follower in the j-th dimension space,
Figure BDA0002390529530000032
indicating rounding up.
Preferably, the fitness is calculated by equation (7):
Figure BDA0002390529530000033
wherein, the fitness is the fitness.
Preferably, according to the fitness, the goblet and ascidian individual with the largest fitness value is taken as the current food source position, the first goblet and ascidian individual of the N-1 goblet and ascidian individuals in the sequence except the food source is taken as the leader, and the rest N-2 goblet and ascidians are all taken as the followers.
Preferably, the leader location is determined by equation (8):
Figure BDA0002390529530000041
wherein x is1,jRepresenting the position of the sea squirt of the leader goblet in dimension j, FjRepresenting the position, ub, of the food source in the j-spacej,maxDenotes an upper limit value, lb, of a j-th spacej,minA lower limit value, c, representing the j-th dimension2And c3Represents the interval [0,1]Random number of cells, c1In order to be a factor of convergence, the method comprises the following steps,
Figure BDA0002390529530000042
t denotes the current number of iterations, TmaxRepresenting the maximum number of iterations of the method.
Preferably, the follower position is determined by equation (9):
Figure BDA0002390529530000043
wherein q' is not less than 2, xq’,jRepresenting the position of the follower in the j-space, △ t representing time, v0Representing the initial velocity of the follower, acceleration a ═ vfinal-v0) /△ t, wherein vfinal＝xq’-1,j-xq’,j/△t，xq’-1,jRepresents the location of the q' -1 st ascidian in the j-dimension space.
According to another aspect of the present invention, an energy-optimized virtual machine deployment system is provided, which is characterized by comprising: a memory storing computer-executable instructions; a processor executing computer executable instructions in the memory to perform the steps of: step 1: determining a virtual machine deployment target and constraint conditions thereof; step 2: determining a deployment solution calculation initialization parameter; and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one; and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof; and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population; step 6: calculating the fitness of each individual goblet ascidian in the updated population, determining the individual goblet ascidian with the largest fitness value, judging whether the fitness value of the individual goblet ascidian is greater than the fitness value of the food source, and if so, taking the individual goblet ascidian as a new food source; if not, the original food source is reserved; and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
Preferably, the calculating the initialization parameters of the deployment solution comprises: maximum number of iterations TmaxThe method comprises the following steps of A, a virtual machine set V, a physical host set H, a resource capacity vector of a host, a resource request vector of a virtual machine, and full power consumption and idle power consumption of the host.
Preferably, the virtual machine deployment targets are:
Figure BDA0002390529530000051
wherein, Pi,fullIndicates the host hiPower consumption in the fully loaded state, Pi,idleIndicates the host hiPower consumption in idle state, PiIndicates the host hiCurrent power consumption of Ui,CPUIs a main machine hiCPU utilization of.
Preferably, the constraint condition is:
Loadi,CPU≤Ci,CPU,i＝1,2,...,m (2)
Loadi,MEM≤Ci,MEM,i＝1,2,...,m (3)
Loadi,DISK≤Ci,DISK,i＝1,2,...,m (4)
Loadi,NETW≤Ci,NETW,i＝1,2,...,m (5)
wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiIn (2)Storage capacity, Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability, Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe overall memory load.
Preferably, the step (3) includes: randomizing the position of the individual goblet sea squirt by formula (6) to obtain an initialization population:
Figure BDA0002390529530000052
wherein q is 1,2, …, N, N represents the size of the ascidian goblet population, j is 1,2, …, N, N represents the total amount of virtual machines, c4Represents the interval [0,1]Inner random number, ubj,maxDenotes an upper limit value, lb, of a j-th spacej,minRepresenting a lower bound, x, of a j-space1,jRepresenting the leader's position in the j-dimensional space, xq,jRepresenting the position of the follower in the j-th dimension space,
Figure BDA0002390529530000053
indicating rounding up.
Preferably, the fitness is calculated by equation (7):
Figure BDA0002390529530000061
wherein, the fitness is the fitness.
Preferably, according to the fitness, the goblet and ascidian individual with the largest fitness value is taken as the current food source position, the first goblet and ascidian individual of the N-1 goblet and ascidian individuals in the sequence except the food source is taken as the leader, and the rest N-2 goblet and ascidians are all taken as the followers.
Preferably, the leader location is determined by equation (8):
Figure BDA0002390529530000062
wherein x is1,jRepresenting the position of the sea squirt of the leader goblet in dimension j, FjRepresenting the position, ub, of the food source in the j-spacej,maxDenotes an upper limit value, lb, of a j-th spacej,minA lower limit value, c, representing the j-th dimension2And c3Represents the interval [0,1]Random number of cells, c1In order to be a factor of convergence, the method comprises the following steps,
Figure BDA0002390529530000063
t denotes the current number of iterations, TmaxRepresenting the maximum number of iterations of the method.
Preferably, the follower position is determined by equation (9):
Figure BDA0002390529530000064
wherein q' is not less than 2, xq’,jRepresenting the position of the follower in the j-space, △ t representing time, v0Representing the initial velocity of the follower, acceleration a ═ vfinal-v0) /△ t, wherein vfinal＝xq’-1,j-xq’,j/△t，xq’-1,jRepresents the location of the q' -1 st ascidian in the j-dimension space.
The method and apparatus of the present invention have other features and advantages which will be apparent from or are set forth in detail in the accompanying drawings and the following detailed description, which are incorporated herein, and which together serve to explain certain principles of the invention.
Drawings
The above and other objects, features and advantages of the present invention will become more apparent by describing in more detail exemplary embodiments thereof with reference to the attached drawings, in which like reference numerals generally represent like parts.
Fig. 1 shows a flow chart of the steps of an energy optimized virtual machine deployment method according to the present invention.
Detailed Description
The invention will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present invention are shown in the drawings, it should be understood that the present invention may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
Fig. 1 shows a flow chart of the steps of an energy optimized virtual machine deployment method according to the present invention.
In this embodiment, the energy-optimized virtual machine deployment method according to the present invention may include: step 1: determining a virtual machine deployment target and constraint conditions thereof; step 2: determining a deployment solution calculation initialization parameter; and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one; and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof; and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population; step 6: calculating the fitness of each goblet ascidian individual in the updated population, determining the goblet ascidian individual with the largest fitness value, judging whether the fitness value of the goblet ascidian individual is greater than the fitness value of the food source, and if so, taking the goblet ascidian individual as a new food source; if not, the original food source is reserved; and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
In one example, the deployment solution calculation initialization parameters include: maximum number of iterations TmaxVirtual machine set V, physical host set H, resource capacity vector of host, resource request vector of virtual machine, hostFull load power consumption and idle power consumption of the machine.
In one example, the virtual machine deployment targets are:
Figure BDA0002390529530000081
wherein, Pi,fullIndicates the host hiPower consumption in the fully loaded state, Pi,idleIndicates the host hiPower consumption in idle state, PiIndicates the host hiCurrent power consumption of Ui,CPUIs a main machine hiCPU utilization of.
In one example, the constraints are:
Loadi,CPU≤Ci,CPU,i＝1,2,...,m (2)
Loadi,MEM≤Ci,MEM,i＝1,2,...,m (3)
Loadi,DISK≤Ci,DISK,i＝1,2,...,m (4)
Loadi,NETW≤Ci,NETW,i＝1,2,...,m (5)
wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiMemory capacity of Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability, Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe overall memory load.
In one example, step (3) comprises: randomizing the position of the individual goblet sea squirt by formula (6) to obtain an initialization population:
Figure BDA0002390529530000082
wherein q is 1,2, …N, N represents the size of the ascidian group, j is 1,2, …, N, N represents the total number of virtual machines, c4Represents the interval [0,1]Inner random number, ubj,maxAn upper limit value representing a j-th dimension space, which is defined as the maximum available physical host number; lbj,minThe lower limit value representing the j-th dimension space is defined as the minimum amount of host utilization. By individual ubs of each goblet ascidianj,max、lbj,minAnd c4The randomly generated Hyssopus goblet population can be obtained, and thus, the deployment solution space is an Euclidean space of N × N, x1,jRepresenting the position of the leader on the jth dimension space, i.e. the jth virtual machine in the deployment solution represented by the leader goblet sea squirt is deployed to the sequence number x1,jOn the host computer of (2), xq,j(q>1) Representing the position of the follower in the jth dimension space, i.e. the jth virtual machine in the deployment solution represented by the follower goblet q is deployed to the serial number xq,jOn the host computer of the computer system (2),
Figure BDA0002390529530000091
indicating rounding up.
In one example, fitness is calculated by equation (7):
Figure BDA0002390529530000092
wherein, the fitness is the fitness.
In one example, according to the fitness, the individual of the goblet ascidian with the largest fitness value (representing the optimal deployment solution) is taken as the current food source position, the first individual of the N-1 individuals of the goblet ascidians in the sequence except the food source is taken as the leader, and the rest N-2 individuals of the goblet ascidians are taken as followers.
In one example, the leader location is determined by equation (8):
Figure BDA0002390529530000093
wherein x is1,jRepresenting the position of the sea squirt of the leader goblet in the space of dimension j, i.e. virtualSimulation machine vjDeployed to sequence number x1,jThe host computer of (2); fjRepresenting the position of the food source in the j-dimensional space; since the food source is unknown to the goblet sea squirts, the position of the leader in the initial state is defined as the position of the food source; ubj,maxAn upper limit value representing a j-th dimension space, which is defined as the maximum available host number; lbj,minA lower limit value representing a j-th dimension space, which is defined as the minimum host utilization number; c. C2And c3Represents the interval [0,1]Random numbers in between, for deciding the moving direction (forward or backward) and the moving step size of the next update position in the j-th dimension space; c. C1Called convergence factor, is used for the local development and the global exploration capability of the individual goblet sea squirts in the iterative process of the equilibrium method,
Figure BDA0002390529530000094
t denotes the current number of iterations, TmaxRepresenting the maximum number of iterations of the method, the convergence factor will decrease from 2 to 0 during the iteration.
In one example, the follower position is determined by equation (9):
Figure BDA0002390529530000101
wherein q' is not less than 2, xq’,jRepresenting the position of the follower in the j-space, i.e. deploying virtual machine vjHost serial number of (1) is xq’,j△ t denotes time, v0Representing the initial velocity of the follower, acceleration a ═ vfinal-v0) /△ t, wherein vfinal＝xq’-1,j-xq’,j/△t，xq’-1,jRepresenting the position of the q '-1 st goblet ascidian on the jth dimension space, namely the deployment solution represented by the position of the q' -1 st goblet ascidian, the jth virtual machine is deployed to the host xq’-1,jThe above.
Specifically, because of the heterogeneity of host resource capacity, the full load power consumption of the host resource capacity is different, and the dimension resources requested by the virtual machines are different, so that n virtual machines are on m physical hostsThe deployment scenario theoretically has mnIn this case, therefore, the virtual machine deployment problem itself is an NP-hard problem. The problem is solved by using a goblet sea squirt group algorithm, and the energy-optimized virtual machine deployment method according to the invention can comprise the following steps:
step 1: suppose that a cloud data center has m heterogeneous physical hosts, denoted as set H ═ H1,h2,…,hm}. Each physical host hiThere are four resource types: CPU, memory, storage and network bandwidth, i ═ 1,2, …, m. Through virtualization technology, multiple virtual machines can be deployed on the same physical host at the same time, and each virtual machine can run different tasks. In order to execute application tasks, cloud users can submit load execution demands to a cloud data center in the form of virtual machines at any time, and the number of virtual machine requests at this time is assumed to be n, and is represented as a set V ═ V1,v2,…,vn}. I.e., n virtual machines now need to be deployed on m physical hosts.
Each physical host hiIs defined as a vector Ci＝(Ci,CPU,Ci,MEM,Ci,DISK,Ci,NETW) Wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiMemory capacity of Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability.
A virtual machine vjThe capability for a host resource request is defined as a vector Rj＝(Rj,CPU,Rj,MEM,Rj,DISK,Rj,NETW) Wherein R isj,CPURepresenting a virtual machine vjRequested CPU capability, Rj,MEMRepresenting a virtual machine vjRequested memory capacity, Rj,DISKRepresenting a virtual machine vjRequested storage capacity, Rj,NETWRepresenting a virtual machine vjRequested network bandwidth capabilities.
Order Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe total memory load of (a) is one physical host hiThe CPU utilization is:
Figure BDA0002390529530000111
v → H represents the deployment solution function of the virtual machine on the physical host, and the deployment factor χ is defined as:
Figure BDA0002390529530000112
i.e. if virtual machine VbIs disposed in a host HgIn the above, the deployment factor χ ═ 1; otherwise, χ ═ 0. Then it is determined that,
Loadi,CPU＝∑Rj,CPUχ(V,H),i＝1,2,...,m,j＝1,2,...,n (12)
Loadi,MEM＝∑Rj,MEMχ(V,H),i＝1,2,...,m,j＝1,2,...,n (13)
Loadi,DISK＝∑Rj,DISKχ(V,H),i＝1,2,...,m,j＝1,2,...,n (14)
Loadi,NETW＝∑Rj,NETWχ(V,H),i＝1,2,...,m,j＝1,2,...,n (15)
the energy consumption of the host is mainly related to the resource utilization rate of the host, and the energy consumption of the host and the resource utilization rate of the host are mainly in a linear relation. For the four types of resources possessed by the host, the CPU resource consumes most of the total energy consumption, and the energy consumption generated by other resource types is ignored, so that in order to measure the energy consumption of the host, a host energy consumption model which has a linear relation with the CPU utilization rate is utilized. The host h can be obtained by using the formula (12)iThe load condition of CPU is shown in the formula (10) to obtain the host hiA CPU utilization, which is the sum of CPU requests of all virtual machines deployed on the host. A main machine hiThe power consumption of (a) is defined as:
Figure BDA0002390529530000113
wherein, Pi,fullIndicates the host hiThe power consumption in the full load state, i.e. the power consumption when 100% of the CPU is occupied, is given by the unit watt, Pi,idleIndicates the host hiThe power consumption in the idle state is generally 70% of the power consumption of the full load when the host is in the idle state. PiIndicates the host hiCurrent power consumption.
The goal of virtual machine deployment is to minimize power consumption for all hosts, and thus, the deployment goal can be expressed as equation (1) with constraints of equations (2) - (5), where equation (2) indicates host hiCannot cross its overall CPU capacity, and equation (3) indicates host hiCannot cross its overall memory capacity, and equation (4) indicates host hiCannot cross its overall storage capacity, and equation (5) indicates that host hiThe network bandwidth resource load on cannot exceed its overall network bandwidth capability.
Step 2: determining the deployment solution to calculate the initialization parameters comprises: goblet sea squirt population scale N, maximum iteration number TmaxA random variable c1、c2、c3And c4Upper limit value ubmaxAnd a lower limit value lbminVirtual machine set V, physical host set H, resource capability vector (C) of hostCPU、CMEM、CDISK、CNETW) Resource request vector (R) of virtual machineCPU、RMEM、RDISK、RNETW) Full load power consumption P of hostfullAnd idle power consumption Pidle；
And step 3: the group initialization is carried out through the goblet sea squirt algorithm, so that the goblet sea squirt individuals correspond to the deployment solution of the virtual machine one by one, the traditional goblet sea squirt group optimization algorithm belongs to the continuous goblet sea squirt group optimization process, and the goblet sea squirt individuals can continuously move on any point in the search space. For the virtual machine deployment problem, its deployment solution will be discretely constrained to different physical host sequence numbers in order to represent to which host a single virtual machine is deployed. Therefore, the individual positions of the goblet sea squirts in the goblet sea squirt group optimization algorithm in the virtual machine deployment problem can only appear in the form of discrete numerical sequence numbers between 1 and m, wherein m represents the maximum resource sequence number. Encoding individual positions of goblet ascidians into the following form:
Xq＝[xq,1,xq,2,...,xq,n](17)
wherein, XqRepresenting the position of the individual q of the sea squirt in the population, each element x in the position matrixq,jRepresenting the location of the goblet ascidian q in the jth dimension. At the same time, let vector element xq,jIs [1, m ]]The random value between q is 1,2, …, N, N represents the size of the ascidian goblet population, and j is 1,2, …, N, N represents the total amount of virtual machines to be deployed. The position of the cask sea squirt, i.e., the corresponding deployment solution for the virtual machine, may be defined within the n-dimensional search space.
For example: the location of q of the individual ascidian as Zun is Xq＝[3,2,5,3,2,1,3,2,6]The specific meaning of the virtual machine deployment solution represented by the method is as follows: virtual machine v1Virtual machine v4And a virtual machine v7Deployment to host h3Upper, virtual machine v2Virtual machine v5And a virtual machine v8Deployment to host h3Upper, virtual machine v3Deployment to host h5Upper, virtual machine v6Deployment to host h1Upper, virtual machine v9Deployment to host h6The above.
The position of all goblet ascidians is a two-dimensional matrix, which can be defined as X:
Figure BDA0002390529530000131
the matrix X defines the positions of all N individuals in the iterative evolution process of the goblet sea squirt population, each row in the matrix represents the position of one goblet sea squirt individual in the N-dimensional space, and the position of one goblet sea squirt individual represents a possible virtual machine deployment solution. Randomizing the positions of the individual goblet ascidians by a formula (6) to obtain an initialized population, namely an initial matrix X, continuously updating the positions of the N individuals in the iteration process until the maximum iteration times are reached, and then obtaining the position of the individual goblet ascidians with the maximum fitness in the matrix X as a final virtual machine deployment solution.
And 4, step 4: according to the deployment target of the virtual machine and the constraint conditions thereof, the individual fitness of each goblet ascidian in the initialized population is calculated through a formula (7), and as the deployment target of the virtual machine is to minimize the energy consumption of a host, the greater the fitness value of the goblet ascidians is, the better the deployment solution represented by the individual is;
and 5: the goblet ascidian population can be divided into two groups: a leader and a follower. The chain head of the turtle sea squirt is called a leader, and has the optimal judgment in the process of searching food sources, so that the movement of the whole population is guided. The remaining other goblet ascidians, except the leader, are called followers. Followers follow each other and are led directly or indirectly. The food source F in the search space is the search target of the Hyacinus goblet population. According to the fitness, the goblet ascidian individual with the largest fitness value is taken as the current food source position, the first goblet ascidian individual of the N-1 goblet ascidian individuals except the food source is taken as the leader, and the rest N-2 goblet ascidians are all taken as followers.
Updating the position of the leader, namely updating the deployment solution of the virtual machine represented by the individual goblet and sea squirt as the leader, determining the position of the leader through a formula (8), wherein the deployment solution of the virtual machine can be limited to [1, m ] finally]Integer values within the interval, i.e. x1,jMust be in the value of [1, m]Within the interval, the virtual machine deployment solution represented by the leader is as follows:
Figure BDA0002390529530000141
wherein mod represents a modular operation, i.e., a remainder; m represents the total number of hosts, i.e., the maximum host sequence number.
The follower position is determined by equation (9), since time is the difference between the number of iterations △ t is 1, and at the beginning of each iteration, the initial velocity v of the follower goblet's ascidian is determined00, so the location of the follower is updatedThe formula can be expressed as:
Figure BDA0002390529530000142
equation (20) shows that the update of the deployment host serial number of the jth virtual machine in the follower goblet ascidian q 'is equal to half of the sum of the deployment solution corresponding to the goblet ascidian in the previous iteration and the deployment host serial number of the jth virtual machine in the follower goblet ascidian q' -1. Based on the same considerations as the leader's location, the deployment solution represented by the follower's location is:
Figure BDA0002390529530000143
and then an updated population is obtained.
Step 6: calculating the fitness of each goblet ascidian individual in the updated population, determining the goblet ascidian individual with the largest fitness value, judging whether the fitness value of the goblet ascidian individual is greater than the fitness value of the food source, and if so, taking the goblet ascidian individual as a new food source; if not, the original food source is reserved;
and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
According to the method, the exploration behavior and the development behavior in the iteration process of the goblet sea squirt group are fully balanced through the group optimization model of the goblet sea squirts, and the virtual machine deployment effect of energy optimization is realized while the global property and diversity are ensured.
Application example
To facilitate understanding of the solution of the embodiments of the present invention and the effects thereof, a specific application example is given below. It will be understood by those skilled in the art that this example is merely for the purpose of facilitating an understanding of the present invention and that any specific details thereof are not intended to limit the invention in any way.
The energy-optimized virtual machine deployment method according to the present invention comprises:
step 1: target of virtual machine deploymentTo minimize power consumption of all hosts, the deployment target can be expressed as equation (1) with constraints of equations (2) - (5), where equation (2) indicates host hiCannot cross its overall CPU capacity, and equation (3) indicates host hiCannot cross its overall memory capacity, and equation (4) indicates host hiCannot cross its overall storage capacity, and equation (5) indicates that host hiThe network bandwidth resource load on cannot exceed its overall network bandwidth capability.
Step 2: determining the deployment solution to calculate the initialization parameters comprises: goblet sea squirt population scale N, maximum iteration number TmaxA random variable c1、c2、c3And c4Upper limit value ubmaxAnd a lower limit value lbminThe method comprises the following steps of A, a virtual machine set V, a physical host set H, a resource capacity vector of a host, a resource request vector of a virtual machine, and full-load power consumption and idle power consumption of the host;
and step 3: performing population initialization through a goblet ascidian algorithm, enabling the goblet ascidian individuals to correspond to the deployment solution of the virtual machine one by one, and randomizing the positions of the goblet ascidian individuals through a formula (6) to obtain an initialized population;
and 4, step 4: calculating the individual fitness of each goblet ascidian in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof by a formula (7);
and 5: according to the fitness, the goblet sea squirt individual with the largest fitness value (representing the optimal deployment solution) is used as the current food source position, the first goblet sea squirt individual in the N-1 goblet sea squirt individuals except the food source is used as a leader, the rest N-2 goblet sea squirts are all used as followers, the position of the leader is determined through a formula (8), and the position of the followers is determined through a formula (9), so that an updated population is obtained;
step 6: calculating the fitness of each goblet ascidian individual in the updated population, determining the goblet ascidian individual with the largest fitness value, judging whether the fitness value of the goblet ascidian individual is greater than the fitness value of the food source, and if so, taking the goblet ascidian individual as a new food source; if not, the original food source is reserved;
and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
In conclusion, the invention fully balances the exploration behavior and the development behavior in the iteration process of the goblet sea squirt group through the group optimization model of the goblet sea squirt, and realizes the virtual machine deployment effect of energy optimization while ensuring the global property and diversity.
It will be appreciated by persons skilled in the art that the above description of embodiments of the invention is intended only to illustrate the benefits of embodiments of the invention and is not intended to limit embodiments of the invention to any examples given.
According to an embodiment of the present invention, there is provided an energy-optimized virtual machine deployment system, characterized by comprising: a memory storing computer-executable instructions; a processor executing computer executable instructions in the memory to perform the steps of: step 1: determining a virtual machine deployment target and constraint conditions thereof; step 2: determining a deployment solution calculation initialization parameter; and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one; and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof; and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population; step 6: calculating the fitness of each goblet ascidian individual in the updated population, determining the goblet ascidian individual with the largest fitness value, judging whether the fitness value of the goblet ascidian individual is greater than the fitness value of the food source, and if so, taking the goblet ascidian individual as a new food source; if not, the original food source is reserved; and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
At one endIn one example, the deployment solution calculation initialization parameters include: maximum number of iterations TmaxThe method comprises the following steps of A, a virtual machine set V, a physical host set H, a resource capacity vector of a host, a resource request vector of a virtual machine, and full power consumption and idle power consumption of the host.
In one example, the virtual machine deployment targets are:
Figure BDA0002390529530000171
wherein, Pi,fullIndicates the host hiPower consumption in the fully loaded state, Pi,idleIndicates the host hiPower consumption in idle state, PiIndicates the host hiCurrent power consumption of Ui,CPUIs a main machine hiCPU utilization of.
In one example, the constraints are:
Loadi,CPU≤Ci,CPU,i＝1,2,...,m (2)
Loadi,MEM≤Ci,MEM,i＝1,2,...,m (3)
Loadi,DISK≤Ci,DISK,i＝1,2,...,m (4)
Loadi,NETW≤Ci,NETW,i＝1,2,...,m (5)
wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiMemory capacity of Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability, Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe overall memory load.
In one example, step (3) comprises: randomizing the position of the individual goblet sea squirt by formula (6) to obtain an initialization population:
Figure BDA0002390529530000172
wherein q is 1,2, …, N, N represents the size of the ascidian goblet population, j is 1,2, …, N, N represents the total amount of virtual machines, c4Represents the interval [0,1]Inner random number, ubj,maxDenotes an upper limit value, lb, of a j-th spacej,minRepresenting a lower bound, x, of a j-space1,jRepresenting the leader's position in the j-dimensional space, xq,jRepresenting the position of the follower in the j-th dimension space,
Figure BDA0002390529530000181
indicating rounding up.
In one example, fitness is calculated by equation (7):
Figure BDA0002390529530000182
wherein, the fitness is the fitness.
In one example, according to the fitness, the goblet ascidian individual with the largest fitness value is taken as the current food source position, the first goblet ascidian individual of the N-1 goblet ascidian individuals except the food source is taken as the leader, and the remaining N-2 goblet ascidians are all taken as followers.
In one example, the leader location is determined by equation (8):
Figure BDA0002390529530000183
wherein x is1,jRepresenting the position of the sea squirt of the leader goblet in dimension j, FjRepresenting the position, ub, of the food source in the j-spacej,maxDenotes an upper limit value, lb, of a j-th spacej,minA lower limit value, c, representing the j-th dimension2And c3Represents the interval [0,1]Random number of cells, c1In order to be a factor of convergence, the method comprises the following steps,
Figure BDA0002390529530000184
t denotes the current number of iterations, TmaxRepresenting the maximum number of iterations of the method.
In one example, the follower position is determined by equation (9):
Figure BDA0002390529530000185
wherein q' is not less than 2, xq’,jRepresenting the position of the follower in the j-space, △ t representing time, v0Representing the initial velocity of the follower, acceleration a ═ vfinal-v0) /△ t, wherein vfinal＝xq’-1,j-xq’,j/△t，xq’-1,jRepresents the location of the q' -1 st ascidian in the j-dimension space.
The system fully balances exploration behaviors and development behaviors in the iteration process of the goblet sea squirt group through the group optimization model of the goblet sea squirts, and realizes the deployment effect of the energy-optimized virtual machine while ensuring the global property and diversity.
It will be appreciated by persons skilled in the art that the above description of embodiments of the invention is intended only to illustrate the benefits of embodiments of the invention and is not intended to limit embodiments of the invention to any examples given.
Having described embodiments of the present invention, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments.

Claims (10)

1. A method for energy-optimized deployment of virtual machines, comprising:
step 1: determining a virtual machine deployment target and constraint conditions thereof;
step 2: determining a deployment solution calculation initialization parameter;
and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one;
and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof;
and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population;
step 6: calculating the fitness of each individual goblet ascidian in the updated population, determining the individual goblet ascidian with the largest fitness value, judging whether the fitness value of the individual goblet ascidian is greater than the fitness value of the food source, and if so, taking the individual goblet ascidian as a new food source; if not, the original food source is reserved;
and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
2. The energy-optimized virtual machine deployment method according to claim 1, wherein deploying solution computation initialization parameters comprises: maximum number of iterations TmaxThe method comprises the following steps of A, a virtual machine set V, a physical host set H, a resource capacity vector of a host, a resource request vector of a virtual machine, and full power consumption and idle power consumption of the host.
3. The energy-optimized virtual machine deployment method of claim 1, wherein the virtual machine deployment goal is:
Figure FDA0002390529520000011
wherein, Pi,fullIndicates the host hiPower consumption in the fully loaded state, Pi,idleIndicates the host hiPower consumption in idle state, PiIndicates the host hiCurrent power consumption of Ui,CPUIs a main machine hiCPU utilization of.
4. The energy-optimized virtual machine deployment method according to claim 1, wherein the constraints are:
Loadi,CPU≤Ci,CPU,i＝1,2,...,m (2)
Loadi,MEM≤Ci,MEM,i＝1,2,...,m (3)
Loadi,DISK≤Ci,DISK,i＝1,2,...,m (4)
Loadi,NETW≤Ci,NETW,i＝1,2,...,m (5)
wherein, Ci,CPURepresents a physical host hiCPU capability of Ci,MEMRepresents a physical host hiMemory capacity of Ci,DISKRepresents a physical host hiStorage capacity of Ci,NETWRepresents a physical host hiNetwork bandwidth capability, Loadi,CPURepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiLoad, Loadi,MEMRepresents a physical host hiThe overall memory load.
5. The energy-optimized virtual machine deployment method of claim 1, wherein said step (3) comprises: randomizing the position of the individual goblet sea squirt by formula (6) to obtain an initialization population:
Figure FDA0002390529520000021
wherein q is 1,2, …, N, N represents the size of the ascidian goblet population, j is 1,2, …, N, N represents the total amount of virtual machines, c4Represents the interval [0,1]Inner random number, ubj,maxDenotes an upper limit value, lb, of a j-th spacej,minRepresenting a lower bound, x, of a j-space1,jRepresenting the leader's position in the j-dimensional space, xq,jRepresenting the position of the follower in the j-th dimension space,
Figure FDA0002390529520000022
indicating rounding up.
6. The energy-optimized virtual machine deployment method according to claim 1, wherein fitness is calculated by equation (7):
Figure FDA0002390529520000031
wherein, the fitness is the fitness.
7. The energy-optimized virtual machine deployment method according to claim 1, wherein, according to the fitness, the ascidian goblet with the highest fitness value is used as the current food source position, the first ascidian goblet among the N-1 ascidian goblet individuals in the sequence except the food source is used as the leader, and all the remaining N-2 ascidians goblet are used as followers.
8. The energy-optimized virtual machine deployment method according to claim 1, wherein leader location is determined by equation (8):
Figure FDA0002390529520000032
wherein x is1,jRepresenting the position of the sea squirt of the leader goblet in dimension j, FjRepresenting the position, ub, of the food source in the j-spacej,maxDenotes an upper limit value, lb, of a j-th spacej,minA lower limit value, c, representing the j-th dimension2And c3Represents the interval [0,1]Random number of cells, c1In order to be a factor of convergence, the method comprises the following steps,
Figure FDA0002390529520000033
t denotes the current number of iterations, TmaxRepresenting the maximum number of iterations of the method.
9. The energy-optimized virtual machine deployment method according to claim 1, wherein follower location is determined by equation (9):
Figure FDA0002390529520000034
wherein q' is not less than 2, xq’,jRepresenting the position of the follower in the j-space, △ t representing time, v0Representing the initial velocity of the follower, acceleration a ═ vfinal-v0) /△ t, wherein vfinal＝xq’-1,j-xq’,j/△t，xq’-1,jRepresents the location of the q' -1 st ascidian in the j-dimension space.
10. An energy-optimized virtual machine deployment system, comprising:
a memory storing computer-executable instructions;
a processor executing computer executable instructions in the memory to perform the steps of:
step 1: determining a virtual machine deployment target and constraint conditions thereof;
step 2: determining a deployment solution calculation initialization parameter;
and step 3: performing population initialization through a goblet sea squirt algorithm to enable the goblet sea squirt individuals to correspond to the deployment solution of the virtual machine one by one;
and 4, step 4: calculating the individual fitness of each bottle sea squirt in the initialized population according to the deployment target of the virtual machine and the constraint conditions thereof;
and 5: determining a food source, a leader and a follower according to the fitness, and further determining the position of the leader and the position of the follower to obtain an updated population;
step 6: calculating the fitness of each individual goblet ascidian in the updated population, determining the individual goblet ascidian with the largest fitness value, judging whether the fitness value of the individual goblet ascidian is greater than the fitness value of the food source, and if so, taking the individual goblet ascidian as a new food source; if not, the original food source is reserved;
and 7: and judging whether the current iteration times are smaller than the maximum iteration times, if so, repeating the steps 5-6, and if not, outputting the position of the food source in the current population as a final optimal solution for deployment of the virtual machine.
CN202010112558.4A 2020-02-24 2020-02-24 Energy-optimized virtual machine deployment method and system Pending CN111338757A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010112558.4A CN111338757A (en) 2020-02-24 2020-02-24 Energy-optimized virtual machine deployment method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010112558.4A CN111338757A (en) 2020-02-24 2020-02-24 Energy-optimized virtual machine deployment method and system

Publications (1)

Publication Number Publication Date
CN111338757A true CN111338757A (en) 2020-06-26

Family

ID=71183681

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010112558.4A Pending CN111338757A (en) 2020-02-24 2020-02-24 Energy-optimized virtual machine deployment method and system

Country Status (1)

Country Link
CN (1) CN111338757A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537833A (en) * 2021-08-12 2021-10-22 北方工业大学 Emergency vehicle dispatching method, device and equipment for traffic accidents

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105843689A (en) * 2016-04-12 2016-08-10 合肥工业大学 Virtual machine deployment method and system
US20160306644A1 (en) * 2015-04-14 2016-10-20 Dell Products L.P. Performance optimization of workloads in virtualized information handling systems

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160306644A1 (en) * 2015-04-14 2016-10-20 Dell Products L.P. Performance optimization of workloads in virtualized information handling systems
CN105843689A (en) * 2016-04-12 2016-08-10 合肥工业大学 Virtual machine deployment method and system

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
MIRJALILI S. ET AL: "Salp Swarm Algorithm: A bio-inspired optimizer for engineering design problems", pages 163 - 191 *
SHAYEM SALEH ALRESHEEDI1.ET AL: "Improved multiobjective salp swarm optimization for virtual machine placement in cloud computing", pages 1 - 24 *
卓金武等: "MATLAB数学建模方法与实践", 30 June 2018, 北京航空航天大学出版社, pages: 121 - 126 *
卓金武等: "高职高专MATLAB数据建模", 31 August 2019, 北京航空航天大学出版社, pages: 124 - 129 *
张炳达: "智能信息处理技术基础", 31 October 2008, 天津大学出版社, pages: 50 - 53 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113537833A (en) * 2021-08-12 2021-10-22 北方工业大学 Emergency vehicle dispatching method, device and equipment for traffic accidents
CN113537833B (en) * 2021-08-12 2023-08-22 北方工业大学 Emergency vehicle dispatching method, device and equipment for traffic accidents

Similar Documents

Publication Publication Date Title
Song et al. Drq: dynamic region-based quantization for deep neural network acceleration
US20210304066A1 (en) Partitioning for an execution pipeline
CN113485826B (en) Load balancing method and system for edge server
CN111966495B (en) Data processing method and device
CN113469354B (en) Memory-constrained neural network training
CN113449861A (en) Speculative training using partial gradient update
US20230252353A1 (en) On-device training method to train an artificial intelligent model and a system therefor
CN109189552A (en) Virtual network function dilatation and capacity reduction method and system
CN115016938A (en) Calculation graph automatic partitioning method based on reinforcement learning
CN111338757A (en) Energy-optimized virtual machine deployment method and system
CN110175172B (en) Extremely-large binary cluster parallel enumeration method based on sparse bipartite graph
CN114761920A (en) Hardware accelerator with reconfigurable instruction set
Gerogiannis et al. Deep reinforcement learning acceleration for real-time edge computing mixed integer programming problems
CN115794354A (en) Cloud video transcoding task scheduling based on K-Means clustering and improved AOA algorithm
CN110781003B (en) Load balancing method for particle swarm fusion variation control
CN116033026A (en) Resource scheduling method
CN113407336B (en) Full-comparison data distribution method based on tabu search optimization algorithm
US11797280B1 (en) Balanced partitioning of neural network based on execution latencies
CN115016889A (en) Virtual machine optimization scheduling method for cloud computing
CN114064235A (en) Multitask teaching and learning optimization method, system and equipment
JP2023531538A (en) Neural network generation method, device and computer readable storage medium
de Freitas Cunha et al. On the impact of MDP design for reinforcement learning agents in resource management
CN116932164B (en) Multi-task scheduling method and system based on cloud platform
Lu et al. Towards Dynamic Request Updating with Elastic Scheduling for Multi-tenant Cloud-based Data Center Network
CN116089070A (en) Task processing method, device, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination