CN109413676B - Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network - Google Patents

Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network Download PDF

Info

Publication number
CN109413676B
CN109413676B CN201811511596.6A CN201811511596A CN109413676B CN 109413676 B CN109413676 B CN 109413676B CN 201811511596 A CN201811511596 A CN 201811511596A CN 109413676 B CN109413676 B CN 109413676B
Authority
CN
China
Prior art keywords
task
user
migration
convex
macro
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.)
Active
Application number
CN201811511596.6A
Other languages
Chinese (zh)
Other versions
CN109413676A (en
Inventor
郑杰
刘艺
郑勇
王文涛
许鹏飞
汪霖
高岭
王海
杨旭东
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.)
Northwest University
Original Assignee
Northwest 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 Northwest University filed Critical Northwest University
Priority to CN201811511596.6A priority Critical patent/CN109413676B/en
Publication of CN109413676A publication Critical patent/CN109413676A/en
Application granted granted Critical
Publication of CN109413676B publication Critical patent/CN109413676B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W24/00—Supervisory, monitoring or testing arrangements
    • H04W24/02—Arrangements for optimising operational condition
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/1008—Server selection for load balancing based on parameters of servers, e.g. available memory or workload
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/101—Server selection for load balancing based on network conditions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W28/00—Network traffic management; Network resource management
    • H04W28/02—Traffic management, e.g. flow control or congestion control
    • H04W28/08—Load balancing or load distribution
    • 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
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Abstract

In order to meet the delay sensitivity requirements of calculation intensive type and data intensive type of mobile intelligent application, an effective method for combining uplink and downlink access and resource allocation needs to be designed to reduce intelligent application delay. Firstly, modeling a combined uplink and downlink calculation migration problem as a minimized system delay optimization model, and simultaneously considering the energy consumption of the user equipment. Because the problem of joint downlink and uplink calculation migration is a mixed integer programming problem, the problem is converted into a resource allocation sub-problem and a calculation migration sub-problem, and an effective method for joint downlink and uplink migration and resource allocation is provided. The experimental result verifies the effectiveness of the method in the aspects of system time delay and energy consumption.

Description

Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network
Technical Field
The invention belongs to the technical field of mobile communication cellular networks, and relates to a method for realizing joint downlink and uplink computing resource allocation and joint uplink and downlink access and downlink and uplink communication resource allocation of a user, in particular to a joint downlink and uplink edge computing migration method in an ultra-dense heterogeneous network.
Background
The intelligent user terminal is widely used for various applications with harsh resource requirements, such as augmented reality, multimedia applications, automatic driving and the like. Running smart applications on mobile devices can result in long response delays, and energy consumption is also a limiting factor for mobile applications for the effective battery capacity of mobile users.
To handle compute-intensive task migration in real-time, the proposed edge computing may provide lower latency and higher performance computing services to users, while using network edge servers close to the users, improving the energy efficiency of the mobile device. Unlike mobile cloud computing, where a user device migrates a computing task to a cloud platform over a wireless network and the results sent back to the user device do not provide a low enough response time for many smart applications, emerging edge computing may provide enough computing proximity to the user to meet the low latency requirements of smart applications.
Ultra-dense heterogeneous networks play an important role in fifth generation communication system (5G) technology, improving the wireless access capability of cellular networks by deploying low-power micro-cells under high-power macro-cells. For mobile cellular networks, user access depends on the received downlink Signal Reference Power (RSRP) from the base station, but can result in macro base station overload and micro base station underrun. Furthermore, if we use joint load balancing of uplink and downlink to allocate users, the downlink will be allocated more resources because the power of the downlink is greater than the power of the uplink and the traffic of the downlink is also greater. Therefore, there is a need to design a wireless access that combines uplink and downlink computation and communication resource allocation to meet the low latency requirements of both computation-intensive and data-intensive tasks.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a combined Downlink and Uplink edge computing migration method in an ultra-dense heterogeneous network, and a combined Downlink (DL, Downlink) and Uplink (UL, Uplink) resource allocation method, which is used for load migration in the ultra-dense heterogeneous network, and not only considers computing-intensive and data-intensive tasks, but also considers computing resources and communication resources of an edge macro cell and a scene of an ultra-dense micro cell aiming at the problem of combined Downlink and Uplink computing migration of the ultra-dense heterogeneous network. And establishing a system model and constraint conditions by taking the total delay of the minimized task as a target and jointly considering the resource constraint of the user and the base station and the wireless channel conditions of the uplink and the downlink. To solve the mixed integer optimization problem in ultra-dense heterogeneous networks, the problem is transformed into two sub-problems: allocation of communication and computing resources and 0-1 integer programming of computational migration decisions. By adopting an alternating iterative optimization technique, an effective scheme combining a downlink and an uplink can be obtained for resource allocation and computation migration in the ultra-dense heterogeneous network. The experimental result shows that the method has excellent performance and can be well adapted to the demand change of calculation and data.
In order to achieve the purpose, the invention adopts the technical scheme that:
a joint lower and upper edge calculation migration method in an ultra-dense heterogeneous network comprises the following steps:
1) a double-layer heterogeneous network formed by macro cells and micro cells is constructed, and a user can only select to access a single macro cell or micro cell but cannot access two macro cells or micro cells simultaneously;
2) for each macro cell, measuring uplink and downlink channel states and interference states of the micro cells and users in an area covered by the macro cell, reporting the result to the macro cell, and determining the access of the users and the allocation of computing resources after the macro cell is computed;
3) for each user, calculating the received downlink signal strength in the whole bandwidth, respectively selecting a macro cell and a micro cell as a set of candidate access cells, and determining that a task is executed locally or in the macro cell or the micro cell according to a joint downlink and uplink edge calculation migration method;
4) combining the initialization of the edge calculation migration method of the lower row and the upper row, the problem is divided into two sub-problems: communication and computing resource allocation and migration decision problems are solved by adopting an alternative iteration optimization technology;
the preliminary preparation and the process of the edge calculation migration method used in the step 4) for combining the lower row and the upper row are as follows:
firstly, modeling interference types of a downlink and an uplink, establishing a signal to interference plus noise ratio (SINR) model, using time division multiplexing by the same base station, measuring a channel state and an interference state through a micro-cell and a user in an area covered by a macro-cell to obtain the SINR model, and obtaining uplink and downlink transmission rates of the user by adopting a Shannon capacity formula;
② modeling T the calculation task of the useri＝(di,si,ci)，
Figure BDA0001900926540000021
Wherein s isiIs the size of the uploaded data, eiIs the size of the returned data after the task is completed, ciIs to complete TiA required CPU period is adopted, and then a local calculation model and an edge calculation model are established;
establishing a joint uplink and downlink edge calculation migration model:
A. task migration decision scheme: the user's task should be handled in a local or edge macro cell or in an edge pico cell;
B. resource allocation algorithm: the computational and communication resources of the edge macro and pico cells should be allocated to each task, thus modeled as an optimization problem (OP 1):
OP1:
Figure BDA0001900926540000031
C1:
Figure BDA0001900926540000032
C2:
Figure BDA0001900926540000033
C3:
Figure BDA0001900926540000034
C4：
Figure BDA0001900926540000035
C5:
Figure BDA0001900926540000036
c1 shows that the decision of user task migration is represented by xi,jIs represented by (j) 0, x i,j1 indicates that the task of user i is processed locally, j ═ 1,2, … M + P, x i,j1 indicates that the task of the user i is processed at the macro base station or the micro base station edge server; c2 indicates that the energy consumption of the user's local or marginal computing task is less than EiRepresents the remaining battery energy of user i; c3 denotes
Figure BDA0001900926540000037
Is the processing power allocated on the macro base station server, fMLimit for total processing capacity of macro base station; c4 denotes
Figure BDA0001900926540000038
Is the allocation of processing capacity, f, on the macrocell serverPA micro base station total processing capacity limit; c5 denotes that the macro and micro cell downlinks and uplinks are constrained in time resources for less than one frame; wherein f isiAnd ρiRespectively representing the computing power and energy cost per CPU cycle of user i, and therefore, by calculating TiThe processing delay and energy cost of (a) are:
Figure BDA0001900926540000039
and
Figure BDA00019009265400000310
wherein c isiIs the energy consumption per CPU cycle, and furthermore, the user's local or marginal computing task T can be obtainediEnergy consumption of (2):
Figure BDA00019009265400000311
wherein EiIndicates the remaining battery energy of user i, the downlink rate of user i is
Figure BDA00019009265400000312
The uplink rate of the user is
Figure BDA00019009265400000313
Wherein
Figure BDA00019009265400000314
And
Figure BDA00019009265400000315
time or bandwidth of downlink and uplink, task TiThe execution time in the macro cell is:
Figure BDA0001900926540000041
task TiThe processing time and energy consumption at the microcell can be calculated as:
Figure BDA0001900926540000042
it is easy to find that OP1 is a mixed integer programming problem, converting the OP1 objective function into OP 2:
Figure BDA0001900926540000043
the problem belongs to the mixed integer programming problem, the problem is divided into two sub-problems, and an alternative iteration optimization technology is adoptedSolving the operation;
communication and computing resource allocation problems: if the decision variable xi,jAccessing a set of microcells p
Figure BDA0001900926540000044
And access to a set of macrocells m
Figure BDA0001900926540000045
Is positive, then the original question OP1 is about fi,jAnd bi,jThe convex planning of (a) can then be solved using a convex optimization tool, given a decision variable xi,jAccessing a set of microcells p
Figure BDA00019009265400000414
And access to a set of macrocells m
Figure BDA0001900926540000046
Then OP1 on fi,jAnd bi,jThe original optimization problem of (1) is a convex planning problem, if the task has migrated to a macro or micro cell, then only communication and computing resource allocation problems remain, that is, xi,j，
Figure BDA0001900926540000047
And
Figure BDA0001900926540000048
being fixed, the objective function OP1 may be converted to OP2 as follows:
Figure BDA0001900926540000049
since f (x) 1/x is convex and the sum of convex functions is convex with a constrained convex set (C3-C5), it is clear that OP2 is convex and therefore an efficient solution can be done using a convex optimization (CVX) tool;
the 0-1 integer smoothing decision method in the step F is as follows:
solving the transformed optimization problem OP2 by a load migration decision algorithm as described in algorithm 1 below, comprising the steps of:
A. initialization: given a
Figure BDA00019009265400000410
And an error threshold δ;
B. if k is 0, then
Figure BDA00019009265400000411
And k ═ k + 1;
C. if the obtained iteration result is
Figure BDA00019009265400000412
Then calculate
Figure BDA00019009265400000413
i＝1,2,…N,j＝1,2…,M+P+1；
D. By using
Figure BDA0001900926540000051
Instead of X, solving the transformed convex problem OP2 using a convex tool
Figure BDA0001900926540000052
E. If it is not
Figure BDA0001900926540000053
The optimum result is obtained
Figure BDA0001900926540000054
Otherwise, returning to the step 3 when k is k + 1;
function for optimization problem OP2
Figure BDA0001900926540000055
Replacing variable xi,jIn combination with zi,jIn place of xi,jThe transformed problem is convex optimization, and the constraint conditions form a convex space, so that the transformed problem can be effectively solved through a convex tool;
sixthly, migrationAnd (3) decision-making problem: for the solution
Figure BDA0001900926540000056
And
Figure BDA0001900926540000057
the sub-problem f (φ) is a 0-1 integer programming problem with respect to φ, since xi,j，
Figure BDA0001900926540000058
And
Figure BDA0001900926540000059
is a 0-1 variable to indicate whether the task for user i is performed locally, macro or micro, with optimal utilization
Figure BDA00019009265400000510
And
Figure BDA00019009265400000511
can be calculated to obtain
Figure BDA00019009265400000512
Then the problem OP2 is solved by an integer program from 0 to 1 as shown below
Figure BDA00019009265400000513
And (3) constraint:
Figure BDA00019009265400000514
xi,j∈{0,1}
using a smoothed logarithmic function log (| x)i,j| + σ) instead of 0-1 variable θ (x)i,j) Then the linear function log (| x) is minimized by iterationi,j| + σ), given a continuous variable zi,jIs not less than 0 and
Figure BDA00019009265400000515
where N is the number of users, a linear function is definedThe following were used:
Figure BDA00019009265400000516
where ε is a very small regularization constant, k denotes the number of iterations, and then χ (z)i,j) Replacement of xi,jAfter the original function is transformed, OP2 is a convex function, and the following algorithm 1 is adopted for solving;
5) the computing task for each user is determined to be local or migrated to a macro or micro cell, and then only the communication and computing resource allocation problem remains, and then solved by employing a convex optimization tool.
6) A binary decision algorithm for task migration for given communication and computing resource allocation, using a smooth logarithmic function log (| x)i,jAnd | plus σ) replaces the decision variable of 0-1, is converted into a convex function, and is further solved by adopting a convex optimization tool.
7) And (4) carrying out repeated iteration on the communication and computing resource allocation problem and the task migration decision algorithm until the change of the obtained computing task migration decision and the communication and computing resource allocation result is less than a given error threshold, and terminating the iteration.
8) After iteration is finished, each user carries out migration of calculation tasks, the macro cells and the micro cells configure calculation and communication resources for uplink and downlink, and finally, migration results of the tasks are returned.
The invention has the beneficial effects that:
in the invention, aiming at joint downlink and uplink edge computation migration in an ultra-dense heterogeneous network, a novel framework is provided to minimize the total delay of all tasks, and simultaneously, different computing capacities of macro cells and micro cells and different data rates of a downlink and an uplink in the ultra-dense heterogeneous network are considered, so that a joint downlink and uplink computation migration and resource allocation scheme is designed. Simulation results show that the performance of the designed method is superior to that of the existing scheme, and the necessity of calculating and migrating the joint downlink and the uplink is verified.
Drawings
FIG. 1 is a diagram illustrating the variation of the total delay with the task computation;
FIG. 2 is a diagram illustrating the variation of energy consumption with task computation according to the present invention;
FIG. 3 is a graph of the total delay variation with packet size according to the present invention;
fig. 4 shows the variation of power consumption with packet size according to the present invention.
Detailed Description
The present invention will be further described with reference to the accompanying drawings, but the present invention is not limited to the following embodiments.
A joint lower and upper edge calculation migration method in an ultra-dense heterogeneous network comprises the following steps:
in a Time Division Duplex (TDD) system of an ultra-dense heterogeneous network, multiple users are covered by macrocells and microcells. Let N ═ {1,2,3 … N }, M ═ {1,2,3 … M } and P ═ 1,2,3 … P ] denote the set of all users, the set of all macrocells and all microcells. Each user has a task that requires computation, and can access either the macro cell or the micro cell, but cannot access both the macro cell and the pico cell, and each task is assumed to be minimal and not resolvable.
In order to provide edge computing services to users, macrocells or microcells each have an edge computing server. For each user, there are a maximum of three migration decision choices to complete their computational tasks: local computation at the user, migration to the macrocell edge computation server and migration to the microcell edge computation server.
Communication model
The interference types for the downlink and uplink in a time division multiplexed heterogeneous network are first modeled. Assuming that the downlink and uplink are deployed for the same channel in the entire bandwidth, the downlink and uplink interference can be classified into three types: interference between base stations, interference between users and interference between base stations and interference between users and interference between base stations. Interference between the base station and the base station means that the base station receives uplink signals that are interfered by base station downlink transmissions, which severely impairs the uplink rate. The interference between users means that the users receive the interference from the uplink transmission to the downlink transmission of the users. Interference between a base station and a user includes interference by uplink transmissions of the user to uplink reception of the base station, and interference by downlink transmissions of other base stations to downlink reception of the base station.
On the basis of the interference model, a signal-to-dryness ratio model can be obtained. The transmission rate of the user can then be calculated by the shannon equation as follows: uplink rate:
Figure BDA0001900926540000071
wherein p isiRepresenting the transmission power of the user. Downlink rate:
Figure BDA0001900926540000072
wherein p isjRepresenting the power of the base station.
The interference relationship of the base stations is determined by a predetermined threshold of Reference Signal Received Power (RSRP) between the base stations, and similarly, the interference relationship may also be decided by a physical distance between the base stations. The downlink rate of the user is
Figure BDA0001900926540000073
The uplink rate of the user is
Figure BDA0001900926540000074
Wherein
Figure BDA0001900926540000075
And
Figure BDA0001900926540000076
is the time or bandwidth of the downlink and uplink.
Task model
For the computation task of any user device i, described as Ti：Ti＝(di,si,ci)，
Figure BDA0001900926540000077
Wherein s isiIs the size of the uploaded data, eiIs the size of the returned data after the task is completed, ciIs to complete TiThe required CPU (Central Processing Unit) cycle.
Micro cells are densely deployed in the ultra-dense heterogeneous network, but one user only accesses macro cells or micro cells and cannot access the macro cells or the micro cells at the same time. In the present invention, we study the calculation of local task migration to edge macro cell base station or edge small cell base station. Assume that the user is covered by the overlap of macro and micro cells during task migration. The decision of user task migration is made by xi,jIs represented by (j) 0, xi,j1 indicates that the task of user i is processed locally, j ═ 1,2, … M + P, xi,jThat is, 1 indicates that the task of the user i is handled by the macro base station or the micro base station edge server
C1:
Figure BDA0001900926540000081
Note that the user's local and edge task migration decisions xi,jOnly one may be 1. In addition, task TiThe transmission energy overhead for migrating to edge base stations can be obtained by:
Figure BDA0001900926540000082
local computation model
Each user executes the user's task on the local CPU, calculates the user's task TiEnergy costs and processing delays. Let fiAnd ρiRepresenting the computation power and energy cost per CPU cycle of user i, respectively. Thus, by calculating TiThe processing delay and energy cost of (a) are:
Figure BDA0001900926540000083
and
Figure BDA0001900926540000084
wherein c isiIs every CPU cycleEnergy consumption of the phase.
In addition, the computing task T of the user at the local or edge can be obtainediEnergy consumption of (2):
Figure BDA0001900926540000085
C2:
Figure BDA0001900926540000086
wherein EiRepresenting the remaining battery energy of user i.
Edge calculation model
For the calculation task of the user in the ultra-dense heterogeneous network, the user has two migration edge servers, except for local calculation in the CPU of the user, one is to perform the task TiComputation migration to macrocells, another is to migrate TiMigrating to a micro-cellular server, determining task TiWhere to perform, the lowest delay in the performance of the user's energy constrained task needs to be considered.
1) Migrating to a macrocell server: the user decides to migrate his task to the macro server that has access to the macro cell, so as to complete the task T in the macro serveri. Considering the communication model, the delay of the task execution mainly comprises two aspects, namely the task T from the user i to the macro cell by wireless transmissioniTransmission time of and task T on the macrocell serveriThe execution time of. Thus, task TiThe execution time of (c) is:
Figure BDA0001900926540000091
wherein the content of the first and second substances,
Figure BDA0001900926540000092
is the processing power allocated on the macro base station server, fMIs the limitation of the total processing capacity of the macro base station.
C3:
Figure BDA0001900926540000093
2) Migrating to a micro-cellular server: the user decides to migrate his task to the micro-cell server, i.e. the user accesses the micro-cell, and the task TiIs done in the microcellular server. As described above, TiThe processing time and energy consumption of (c) can be calculated as follows:
Figure BDA0001900926540000094
wherein the content of the first and second substances,
Figure BDA0001900926540000095
is the allocation of processing capacity, f, on the macrocell serverPIs a micro base station total processing capacity limit.
C4：
Figure BDA0001900926540000096
The macro cell and the micro cell have the same bandwidth, and the downlink and uplink may be multiplexed on time resources. Thus, the time resource constraints are as follows:
C5:
Figure BDA0001900926540000097
optimization problem modeling
Due to the limited computational power of heterogeneous different edge servers and the different transmission powers of macro and pico cells, we tailor the problem to two optimal choices: 1) task migration decision scheme: the user's task should be handled in the local or edge macro cell or in the edge pico cell, 2) resource allocation algorithm: the computation and communication resources of the edge macro and pico cells should be allocated to each task. Therefore, the problem OP1 is expressed as follows:
Figure BDA0001900926540000098
and (3) constraint: (C1) - (C5)
It is easy to find that the task migration problem OP1 that minimizes latency is an NP-hard problem even without consideration of computation and communication resource allocation.
Combining an uplink load migration algorithm and a downlink load migration algorithm;
in this section, the OP1 objective function is converted to:
Figure BDA0001900926540000101
the above problem pertains to mixed integer programming. In the invention, the problem is divided into two sub-problems, and the solution is carried out by adopting an alternative iteration optimization technology:
1) communication and computing resource allocation issues: if the decision variable xi,jAccessing a set of microcells p
Figure BDA0001900926540000102
And access to a set of macrocells m
Figure BDA0001900926540000103
Is positive, then the original question OP1 is about fi,jAnd bi,jThe convex planning of (a) can then be solved using a convex optimization tool.
2) Migration decision problem: for the solution
Figure BDA0001900926540000104
And
Figure BDA0001900926540000105
the sub-problem f (φ) is a 0-1 integer programming problem with respect to φ.
Communication and computing resource allocation
Given xi,j，
Figure BDA0001900926540000106
And
Figure BDA0001900926540000107
then OP1 on fi,jAnd bi,jThe original optimization problem of (2) is a convex programming problem. If the task has migrated to a macro or micro cell, then only communication and computing resource allocation issues remain. That is, xi,j，
Figure BDA0001900926540000108
And
Figure BDA0001900926540000109
being fixed, the objective function OP1 may be converted to OP2 as follows:
Figure BDA00019009265400001010
since f (x) 1/x is a convex function and the sum of the convex functions is convex with a constrained convex set (C3-C5), it is clear that OP2 is convex. Therefore, a convex optimization (CVX) tool can be used for efficient solution.
Load migration decision
Due to xi,j，
Figure BDA00019009265400001011
And
Figure BDA00019009265400001012
is a variable 0-1 to indicate whether the task for user i is performed at a local, macro or micro base station. Using optimal
Figure BDA00019009265400001013
And
Figure BDA00019009265400001014
can be calculated to obtain
Figure BDA00019009265400001015
Then the problem OP2 is solved by an integer program from 0 to 1 as shown below
Figure BDA00019009265400001016
And (3) constraint:
Figure BDA00019009265400001017
xi,j∈{0,1}
using a smoothed logarithmic function log (| x)i,j| + σ) instead of 0-1 variable θ (x)i,j) Then the linear function log (| x) is minimized by iterationi,j| + σ). Given a continuous variable zi,jIs not less than 0 and
Figure BDA0001900926540000111
where N is the number of users, a linear function is defined as follows:
Figure BDA0001900926540000112
where ε is a very small regularization constant and k represents the number of iterations. Then, with chi (z)i,j) Replacement of xi,j. OP2 is a convex function after the primitive function transformation, and is solved using algorithm 1 as follows.
Algorithm 1: load migration decision algorithm
1. Initialization: given a
Figure BDA0001900926540000113
Sum error threshold delta
2. If k is 0, then
Figure BDA0001900926540000114
And k is k +1
3. If the obtained iteration result is
Figure BDA0001900926540000115
Then calculate
Figure BDA0001900926540000116
i＝1,2,…N,j＝1,2…,M+P+1
4. By using
Figure BDA0001900926540000117
Instead of X, solving the transformed convex problem OP2 using a convex tool
Figure BDA0001900926540000118
5. If it is not
Figure BDA0001900926540000119
The optimum result is obtained
Figure BDA00019009265400001110
Otherwise k +1 returns to step 3.
Function for optimization problem OP2
Figure BDA00019009265400001111
Replacing variable xi,jIn combination with zi,jIn place of xi,j. The post-transform problem is convex optimization and the constraints form a convex space. Thus, the transformed problem can be solved efficiently by the convex tool.
Performance evaluation
This section verifies the performance of the proposed joint downlink and uplink edge computation migration algorithm in ultra dense heterogeneous networks and provides a comparative analysis against these existing solutions. In the simulation, a dense urban multi-user edge computing scenario of an ultra-dense heterogeneous network is considered, where the dense urban user density is typically 350 users per square kilometer. The users are uniformly distributed in the coverage area of the two layers of heterogeneous base stations, and the density of macro cells is 5 macro base stations/km2The density of microcells is 300 micro base stations/km2. The positions of the macro base station and the micro base station are subject to uniform distribution. The system parameters are set as in table I, and the simulation results are averaged by averaging 500 trials. The parameter settings of the simulation are shown in Table 1
TABLE 1
Figure BDA0001900926540000121
The proposed Method (ouradvanced Method) was compared with the following different migration schemes:
1) migration by calculation based on uplink signal strength (Offloading with UL RSRP): when a user decides not to process a task locally, the calculation task is migrated to an edge macro cell or a micro cell according to the Reference Signal Received Power (RSRP) of an uplink, and the communication and calculation resource allocation is executed by adopting the algorithm in the invention.
2) Migration by computation (Offloading with DL RSRP) based on downlink signal strength: when the user decides not to process the task locally, the calculation task is migrated to the edge macro cell or the micro cell according to the Reference Signal Received Power (RSRP) of the downlink, and the communication and calculation resource allocation are executed by adopting the algorithm in the invention.
First, a case where migration performance varies with the amount of calculation of a task whose data size follows a uniform distribution of 6MB average is evaluated.
In fig. 1 and 2, as the amount of calculation of the task increases, the delay and power consumption of the task also increase. Furthermore, the proposed method shows shorter task delays and lower energy consumption compared to the Offloading with UL RSRP migration method and Offloading with DL RSRP migration method. This is because the proposed method takes into account the different computing power of the macro and micro base stations and the different data rates of the downlink and uplink, with minimal task delay and energy consumption.
In fig. 1, the proposed method can reduce task delay by 53.3% and task delay by 19.3% compared to Offloading with DL RSRP scheme migration and Offloading with UL RSRP scheme migration. In fig. 2, the proposed method can reduce the energy consumption by 37.7% and 16.9% compared to migration with the Offloading with DL RSRP scheme and migration with the Offloading with UL RSRP scheme. This is because migration of computations using the Offloading with DL RSRP scheme results in macro overload and micro cell underload, and migration of computations using the Offloading with UL RSRP scheme may result in load balancing between macro cells and micro cells, but does not take into account the differences in communication and computation resources between macro cells, micro cells and users.
In addition, the migration performance was further evaluated as a function of data size, where the size of the packets obeyed a uniform distribution based on a 1Gigacycle mean. As can be seen in fig. 3 and 4, the larger the amount of task data, the greater the energy consumption, and the greater the task delay. The proposed method may result in shorter delays and lower energy consumption compared to migration using Offloading with UL RSRP scheme and migration using Offloading with DL RSRP scheme. In fig. 3, the proposed method reduces the task delay by 47.3% and 28.9% compared to DL RSRP scheme migration and migration using UL RSRP scheme. In fig. 4, the proposed method reduces the energy cost by 47.2% and 33.7% compared to migration with DL RSRP scheme and migration with UL RSRP scheme.

Claims (1)

1. A joint lower and upper edge calculation migration method in an ultra-dense heterogeneous network is characterized by comprising the following steps:
1) a double-layer heterogeneous network formed by macro cells and micro cells is constructed, and a user can only select to access a single macro cell or micro cell but cannot access two macro cells or micro cells simultaneously;
2) for each macro cell, measuring uplink and downlink channel states and interference states of the micro cells and users in an area covered by the macro cell, reporting the result to the macro cell, and determining the access of the users and the allocation of computing resources after the macro cell is computed;
3) for each user, calculating the received downlink signal strength in the whole bandwidth, respectively selecting a macro cell and a micro cell as a set of candidate access cells, and determining that a task is executed locally or in the macro cell or the micro cell according to a joint downlink and uplink edge calculation migration method;
4) combining the initialization of the edge calculation migration method of the lower row and the upper row, the problem is divided into two sub-problems: communication and computing resource allocation and migration decision problems are solved by adopting an alternative iteration optimization technology;
the preliminary preparation and the process of the edge calculation migration method used in the step 4) for combining the lower row and the upper row are as follows:
firstly, modeling interference types of a downlink and an uplink, establishing a signal to interference plus noise ratio (SINR) model, using time division multiplexing by the same base station, measuring a channel state and an interference state through a micro-cell and a user in an area covered by a macro-cell to obtain the SINR model, and obtaining uplink and downlink transmission rates of the user by adopting a Shannon capacity formula;
② modeling T the calculation task of the useri＝(di,si,ci)，
Figure FDA0003265040470000011
Wherein s isiIs the size of the uploaded data, diIs the size of the returned data after the task is completed, ciIs to complete TiA required CPU period is adopted, and then a local calculation model and an edge calculation model are established;
establishing a joint uplink and downlink edge calculation migration model:
A. task migration decision scheme: the user's task should be handled in a local or edge macro cell or in an edge pico cell;
B. resource allocation algorithm: the computational and communication resources of the edge macro and pico cells should be allocated to each task, thus modeled as an optimization problem (OP 1):
OP1:
Figure FDA0003265040470000012
C1：
Figure FDA0003265040470000021
C2：
Figure FDA0003265040470000022
C3：
Figure FDA0003265040470000023
C4：
Figure FDA0003265040470000024
C5：
Figure FDA0003265040470000025
c1 shows that the decision of user task migration is represented by xi,jIs represented by (j) 0, xi,j1 indicates that the task of user i is processed locally, j ═ 1,2, … M + P, xi,j1 indicates that the task of the user i is processed at the macro base station or the micro base station edge server; c2 indicates that the energy consumption of the user's local or marginal computing task is less than EiRepresents the remaining battery energy of user i; c3 denotes fi mIs the processing power allocated on the macro base station server, fMLimit for total processing capacity of macro base station; c4 denotes fi pIs the allocation of processing capacity, f, on the macrocell serverPA micro base station total processing capacity limit; c5 denotes that the macro and micro cell downlinks and uplinks are constrained in time resources for less than one frame; wherein f isiAnd ρiRespectively representing the computing power and energy cost per CPU cycle of user i, and therefore, by calculating TiThe processing delay and energy cost of (a) are:
Figure FDA0003265040470000026
and
Figure FDA0003265040470000027
wherein c isiIs the energy consumption per CPU cycle, and furthermore, the user's local or marginal computing task T can be obtainediEnergy consumption of (2):
Figure FDA0003265040470000028
wherein EiIndicates the remaining battery energy of user i, the downlink rate of user i is
Figure FDA0003265040470000029
The uplink rate of the user is
Figure FDA00032650404700000210
Wherein
Figure FDA00032650404700000211
And
Figure FDA00032650404700000212
time or bandwidth of downlink and uplink, task TiThe execution time in the macro cell is:
Figure FDA00032650404700000213
task TiThe processing time and energy consumption at the microcell can be calculated as:
Figure FDA00032650404700000214
it is easy to find that OP1 is a mixed integer programming problem, converting the OP1 objective function into OP 2:
Figure FDA0003265040470000031
the problem belongs to a mixed integer programming problem, the problem is divided into two sub-problems, and an alternative iteration optimization technology is adopted for solving the problem;
communication and computing resource allocation problems: decision variable x if task is migratedi,jAccessing a set of microcells p
Figure FDA0003265040470000032
And access to a set of macrocells m
Figure FDA0003265040470000033
Is positive, then the original question OP1 is about fi,jAnd bi,jThe convex planning of (a) can then be solved using a convex optimization tool, given a decision variable xi,jAccessing a set of microcells p
Figure FDA0003265040470000034
And access to a set of macrocells m
Figure FDA0003265040470000035
Then OP1 on fi,jAnd bi,jThe original optimization problem of (1) is a convex planning problem, if the task has migrated to a macro or micro cell, then only communication and computing resource allocation problems remain, that is, xi,j，
Figure FDA0003265040470000036
And
Figure FDA0003265040470000037
being fixed, the objective function OP1 may be converted to OP2 as follows:
Figure FDA0003265040470000038
since f (x) 1/x is convex and the sum of convex functions is convex with a constrained convex set (C3-C5), it is clear that OP2 is convex and therefore an efficient solution can be done using a convex optimization (CVX) tool;
the 0-1 integer smoothing decision method is as follows:
solving the transformed optimization problem OP2 through a load migration decision algorithm in the algorithm 1;
sixthly, migration decision problem: for the solution
Figure FDA0003265040470000039
And
Figure FDA00032650404700000310
the sub-problem f (φ) is a 0-1 integer programming problem with respect to φ, since xi,j，
Figure FDA00032650404700000311
And
Figure FDA00032650404700000312
is a 0-1 variable to indicate whether the task for user i is performed locally, macro or micro, with optimal utilization
Figure FDA00032650404700000313
And
Figure FDA00032650404700000314
can be calculated to obtain
Figure FDA00032650404700000315
Then the problem OP2 is solved by an integer program from 0 to 1 as shown below
Figure FDA00032650404700000316
And (3) constraint:
Figure FDA00032650404700000317
xi,j∈{0,1}
using a smoothed logarithmic function log (| x)i,j| + σ) instead of 0-1 variable θ (x)i,j) Then the linear function log (| x) is minimized by iterationi,j| + σ), given a continuous variable zi,jIs not less than 0 and
Figure FDA0003265040470000041
where N is the number of users, a linear function is defined as follows:
Figure FDA0003265040470000042
where ε is a very small regularization constant, k denotes the number of iterations, and then χ (z)i,j) Replacement of xi,jAfter the original function is converted, OP2 is a convex function, and an algorithm 1 is adopted for solving;
5) determining that the computing task of each user is local or is migrated to a macro cell or a micro cell, and solving by adopting a convex optimization tool if only the communication and computing resource allocation problem is left;
6) a binary decision algorithm for task migration for given communication and computing resource allocation, using a smooth logarithmic function log (| x)i,j| + σ) replaces the decision variable of 0-1, is converted into a convex function, and is further solved by adopting a convex optimization tool;
7) the communication and calculation resource allocation problem and task migration decision algorithm are iterated repeatedly until the change of the obtained calculation task migration decision and communication and calculation resource allocation result is less than a given error threshold, and then iteration is terminated;
8) after iteration is finished, each user carries out migration of calculation tasks, the macro cell and the micro cell configure calculation and communication resources for the uplink and the downlink, and finally returns a migration result of the tasks;
the algorithm 1 is a load migration decision algorithm and comprises the following steps:
1. initialization: given a
Figure FDA0003265040470000043
Sum error threshold delta
2. If k is 0, then
Figure FDA0003265040470000044
And k is k +1
3. If the obtained iteration result is
Figure FDA0003265040470000045
Then calculate
Figure FDA0003265040470000046
Figure FDA0003265040470000047
4. By using
Figure FDA0003265040470000048
Instead of X, solving the transformed convex problem OP2 using a convex tool
Figure FDA0003265040470000049
5. If it is not
Figure FDA00032650404700000410
The optimum result is obtained
Figure FDA00032650404700000411
Otherwise, returning to the step 3 when k is k + 1;
function for optimization problem OP2
Figure FDA00032650404700000412
Replacing variable xi,jIn combination with zi,jIn place of xi,j(ii) a The transformed problem is convex optimization, and the constraint condition forms a convex space; thus, the transformed problem can be solved efficiently by the convex tool.
CN201811511596.6A 2018-12-11 2018-12-11 Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network Active CN109413676B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201811511596.6A CN109413676B (en) 2018-12-11 2018-12-11 Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201811511596.6A CN109413676B (en) 2018-12-11 2018-12-11 Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network

Publications (2)

Publication Number Publication Date
CN109413676A CN109413676A (en) 2019-03-01
CN109413676B true CN109413676B (en) 2021-11-02

Family

ID=65458509

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201811511596.6A Active CN109413676B (en) 2018-12-11 2018-12-11 Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network

Country Status (1)

Country Link
CN (1) CN109413676B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110311979B (en) * 2019-07-03 2022-05-17 广东工业大学 Task migration method of MEC server and related device
CN110602103B (en) * 2019-09-17 2021-09-10 中国联合网络通信集团有限公司 Electronic lock protocol conversion optimization method and electronic lock protocol conversion optimizer
CN110505644B (en) * 2019-09-26 2021-09-10 江南大学 User task unloading and resource allocation joint optimization method
CN110677858B (en) * 2019-10-25 2022-05-17 国家电网有限公司 Transmission power and computing resource allocation method based on task migration period of Internet of things
CN111309460B (en) * 2020-01-15 2023-04-07 南京大学 Task processing method of intelligent mobile equipment in mobile edge computing scene
CN111338760B (en) * 2020-02-27 2023-04-25 长沙市源本信息科技有限公司 Service instance cross-node telescoping method and device for edge computing
CN111491332B (en) * 2020-04-20 2021-08-27 中国石油大学(北京) Dynamic service migration method and device
CN111857976B (en) * 2020-08-07 2023-06-02 华侨大学 Multi-objective optimization computing migration method based on decomposition
CN112104737B (en) * 2020-09-17 2022-08-30 南方电网科学研究院有限责任公司 Calculation migration method, mobile computing equipment and edge computing equipment
CN112492265B (en) * 2020-10-30 2023-05-02 广东电网有限责任公司电力调度控制中心 Uplink and downlink resource joint allocation method applied to smart grid
CN112486674A (en) * 2020-11-18 2021-03-12 北京邮电大学 Artificial intelligence service-oriented resource allocation method and device and electronic equipment
CN113050482A (en) * 2021-03-11 2021-06-29 合肥福彻工业科技有限公司 Reactor control system and method based on edge calculation technology and reactor system
CN113596901B (en) * 2021-07-29 2024-03-08 西北大学 Online learning parallel edge computing method

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1728621A (en) * 2004-07-29 2006-02-01 大唐移动通信设备有限公司 Method for estimating capability in sector of TD-SCDMA system
CN102264084A (en) * 2010-05-26 2011-11-30 中兴通讯股份有限公司 Methods and device for optimizing coverage by combining multiple base stations
CN103957563A (en) * 2014-05-22 2014-07-30 西安电子科技大学 Joint uplink and downlink load distribution method based on reinforced inter-cell interference coordination
WO2015069983A1 (en) * 2013-11-08 2015-05-14 Spidercloud Wireless, Inc. Fractional frequency reuse schemes assigned to radio nodes in an lte network
WO2017148535A1 (en) * 2016-03-04 2017-09-08 Telefonaktiebolaget Lm Ericsson (Publ) Inter-frequency load balancing
CN107306169A (en) * 2016-04-18 2017-10-31 中兴通讯股份有限公司 A kind of uplink control information transmission method, apparatus and system

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1728621A (en) * 2004-07-29 2006-02-01 大唐移动通信设备有限公司 Method for estimating capability in sector of TD-SCDMA system
CN102264084A (en) * 2010-05-26 2011-11-30 中兴通讯股份有限公司 Methods and device for optimizing coverage by combining multiple base stations
WO2015069983A1 (en) * 2013-11-08 2015-05-14 Spidercloud Wireless, Inc. Fractional frequency reuse schemes assigned to radio nodes in an lte network
CN103957563A (en) * 2014-05-22 2014-07-30 西安电子科技大学 Joint uplink and downlink load distribution method based on reinforced inter-cell interference coordination
WO2017148535A1 (en) * 2016-03-04 2017-09-08 Telefonaktiebolaget Lm Ericsson (Publ) Inter-frequency load balancing
CN107306169A (en) * 2016-04-18 2017-10-31 中兴通讯股份有限公司 A kind of uplink control information transmission method, apparatus and system

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
《A novel resource allocation algorithm based on downlink and uplink decouple access scheme in heterogeneous networks》;Laiwei Jiang，Xuejun Sha，Xuanli Wu，Naitong Zhang;《China Communications》;20160714;全文 *
《Efficient subchannel allocation based on clustered interference alignment in ultra-dense femtocell networks》;Hao Zhang，Hongyan Li，Jung Hoon Lee;《China Communications》;20170518;全文 *
《超密集网络中基于移动边缘计算的卸载策略研究》;郭俊;《中国优秀硕士学位论文全文数据库信息科技辑》;20181115;全文 *
《超密集网络资源管理关键技术研究》;张琪;《中国优秀硕士学位论文全文数据库信息科技辑》;20180415;全文 *

Also Published As

Publication number Publication date
CN109413676A (en) 2019-03-01

Similar Documents

Publication Publication Date Title
CN109413676B (en) Combined downstream and upstream edge calculation migration method in ultra-dense heterogeneous network
CN108809695B (en) Distributed uplink unloading strategy facing mobile edge calculation
Hasan et al. Distributed resource allocation in 5G cellular networks
CN104770004B (en) A kind of communication system and method
CN108965009B (en) Load known user association method based on potential game
CN112512065B (en) Method for unloading and migrating under mobile awareness in small cell network supporting MEC
Su et al. 5G multi-tier radio access network planning based on voronoi diagram
Ye et al. Towards an optimal user association in heterogeneous cellular networks
CN107454601B (en) Wireless virtual mapping method considering inter-cell interference under ultra-dense environment
Sial et al. Analysis of K-tier 5G heterogeneous cellular network with dual-connectivity and uplink–downlink decoupled access
Mollahasani et al. Density-aware, energy-and spectrum-efficient small cell scheduling
Yu et al. Interference coordination strategy based on Nash bargaining for small‐cell networks
Zhou et al. Multi-server federated edge learning for low power consumption wireless resource allocation based on user QoE
CN107396372B (en) Distributed energy efficiency optimization method for maximum and minimum fairness under time domain interference coordination
CN110602718B (en) Heterogeneous cellular network power distribution method and system based on alternative direction multiplier method
Alexandris et al. Load-aware handover decision algorithm in next-generation HetNets
CN108834158B (en) Interference management method for ultra-dense networking
Sidiq et al. A new method of hybrid optimization of small cell range development and density for energy efficient ultra‐dense networks
Li et al. Energy-aware interference management for ultra-dense multi-tier HetNets: Architecture and technologies
CN110012483B (en) Interference coordination method combining asymmetric access and wireless energy-carrying communication
Lee et al. Theoretical-analysis-based distributed load balancing over dynamic overlay clustering
Liu et al. Distributed self-optimizing interference management in ultra-dense networks with non-orthogonal multiple access
Xu et al. Energy efficiency optimization-oriented control plane and user plane adaptation with a frameless network architecture for 5G
Ziaeddini et al. An optimized multi-layer resource management in mobile edge computing networks: a joint computation offloading and caching solution
Chiumento et al. Scalable LTE interference mitigation solution for HetNet deployment

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
GR01 Patent grant
GR01 Patent grant