CN109582461A - A kind of calculation resource disposition method and system for linux container - Google Patents
A kind of calculation resource disposition method and system for linux container Download PDFInfo
- Publication number
- CN109582461A CN109582461A CN201811352854.0A CN201811352854A CN109582461A CN 109582461 A CN109582461 A CN 109582461A CN 201811352854 A CN201811352854 A CN 201811352854A CN 109582461 A CN109582461 A CN 109582461A
- Authority
- CN
- China
- Prior art keywords
- resource
- container
- dimension
- physical host
- host
- 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.)
- Granted
Links
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
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/505—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering the load
Landscapes
- Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention provides a kind of calculation resource disposition method and system for linux container.The method, it include: 1) for each physical host, the case where N-dimensional resource can be provided to the conditions of demand of N-dimensional resource and the physical host according to the container to be built, calculates matching degree of the physical host in the resource of every dimension with the container, N > 1；2) in N-dimensional space, based on the feasible zone range determined by the feasible zone boundary value m being arranged, the one or more physical hosts being within the scope of the feasible zone with the matching degree of the container, m > 0 are filtered out in the resource of every dimension.The present invention can be effectively reduced resource fragmentation, improve host close to the resource utilization under full load situation, avoid the wasting of resources.Also, the present invention improves about 100% in terms of being deployed in resource balancing degree compared to traditional Greedy strategy after tested.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of resource deployment schemes for linux container.
Background technique
As an important technology for realizing linux kernel function, linux container technology (LXC) is away from the present existing more than ten years
Developing history.The technology to be isolated from each other between different processes by the build tool packet, and such isolation is not needed by hard
Part virtualization.Be explained herein with reference to principle of the Fig. 1 to linux container technology, as shown, by container be used as every
The system of disembarking, is different zones by the division of resources of host (i.e. physical machine entity), and multiple containers are shared in the same Linux
Core accommodates an application in each container, so that the execution for the program of different application is independent of one another.It is believed that container
It is the virtualization for having carried out one layer of software in existing operating system, kernel is shared between container, virtualization level is seldom, makes
It has smaller expense and consumed resource for the virtualization of virtual machine.Since container technique is with light, simple
Clean advantage receives the favor of major cloud service provider, is virtualization technology burning hot at present.
It include: Docker, openVZ, FreeBSD jails, Solaris than more typical linux container technology
Containers.Most of open source systems based on these technologies are all very simple to the deployment strategy of resource, often only consider
For the matching of CPU and memory.However, the resource in cluster is usually all multidimensional, such as CPU, memory, network bandwidth,
GPU, block I/O bandwidth, storage etc., if then holding carrying out to guarantee that the balanced of multi dimensional resource uses when resource deployment
It easily causes and occurs resource fragmentation in cluster.Assuming that only consider the three dimensional resource in cluster, CPU, memory and storage, based on existing
Strategy it is possible that CPU is finished and memory there remains many situations, will cause the waste of resource in this way.It is then desired to one
Kind is capable of the deployment strategy of balanced all kinds of resources, to make full use of the multi dimensional resource in cluster.
Summary of the invention
Therefore, it is an object of the invention to overcome the defect of the above-mentioned prior art, a kind of money for linux container is provided
Source dispositions method, comprising:
1) each physical host, conditions of demand and the physics according to the container to be built to N-dimensional resource are directed to
Host can provide the case where N-dimensional resource, calculate of the physical host in the resource of every dimension with the container
With degree, N > 1；
2) it in N-dimensional space, based on the feasible zone range determined by the feasible zone boundary value m being arranged, filters out every
The one or more physical hosts being in the resource of dimension with the matching degree of the container within the scope of the feasible zone, m >
0。
Preferably, according to the method for the present invention, wherein step 1) include: by the physical host each dimension resource
The upper matching degree with the container is transformed into unified numerical intervals.
Preferably, according to the method for the present invention, wherein by following formula by the physical host in the resource of each dimension
Unified numerical intervals are transformed into the matching degree of the container:
ScoreD(p, c)=(useD+γD)/TD*100
Wherein, ScoreDThe matching degree in resource with the container c, use are tieed up in D for physical host pDIt is physical host
Usage amount of the p for D dimension resource, TDIt is total amount of the physical host p for D dimension resource, γDIt is container c to institute
State the demand of D dimension resource.
Preferably, according to the method for the present invention, wherein feasible zone range described in step 2) be in N dimension space with most
The distance of excellent solution is less than the region of m, and the optimal solution is the solution that the value of each dimension is equal in N-dimensional space.
Preferably, according to the method for the present invention, wherein judge whether a physical host meets using following formula in step 2)
It is within the scope of the feasible zone in the resource of every dimension with the matching degree of the container:
Wherein, xiFor the resource matched degree of i-th dimension, N is the number of dimensions of resource, and m is the feasible zone boundary value of setting.
Preferably, according to the method for the present invention, wherein feasible zone range described in step 2) be in N dimension space with original
The distance of point is less than the region of m, and the origin is the point that the value of each dimension is 0 in N-dimensional space.
Preferably, according to the method for the present invention, wherein further include:
3) it for each physical host filtered out, calculates for indicating all to lead after assigning them to the container
The resources balance factor of the resources balance situation of machine cluster；
4) according to the resources balance factor, selection is so that the optimal physics master of the resources balance situation of whole mainframe clusters
Machine distributes to the container.
Preferably, according to the method for the present invention, wherein step 3) is calculated as follows current container being deployed in alternative host k
Resources balance factor when upper:
Wherein, Hostk_useDIt is that alternative host k has used resource situation, Host on dimension Di_useDIt is that other are standby
Host i is selected to use resource situation on dimension D.γDIt is demand of the current container to D dimension resource, N is money
The number of dimensions in source, M are the quantity of physical host in cluster.
Also, step 4) includes: to select the smallest physical host of the resources balance factor as distributing to the container
Physical host.
Preferably, according to the method for the present invention, wherein step 4) includes: the resources balance factor in multiple physical hosts
When identical, according to the resource that the application to be run in the container is stressed, adjust each in the resources balance factor of physical host
Ratio shared by a dimension resource.
Preferably, according to the method for the present invention, before step 1) further include:
0-1) determine the multiple containers for needing to build in system；
0-2) according to the dependence between the multiple container, the size of stock number needed for each container, select into
The priority of each container when row resource deployment, with according to the priority be each container allocation physical host.
Preferably, according to the method for the present invention, wherein step 0-2) include: container to be relied on by other containers, and/or
Need the container of hardware resource and/or the higher priority of container setting that required stock number is big.
A kind of computer readable storage medium, wherein being stored with computer program, the computer program is when executed
For realizing method described in above-mentioned any one.
A kind of resource deployment system for for linux container, comprising:
Storage device and processor；
Wherein, for storing computer program, the computer program executes the storage device by the processor
When for realizing method described in above-mentioned any one.
Compared with the prior art, the advantages of the present invention are as follows:
Screened using the feasible zone of hyperspace in the resource of every dimension be in the matching degree of container it is feasible
Physical host within the scope of domain, so that will not to occur resource in any one dimension excessive or very few for the physical host filtered out
The case where, to avoid the case where resource fragmentation occur during container runs program.Physical host is being calculated every
When matching degree in the resource of dimension with container, the matching degree of each dimension is transformed into unified numerical intervals, so that
Screening in N-dimensional space becomes easier.Feasible zone range is defined as being less than at a distance from optimal solution in N-dimensional space
The region of the m value of setting, the solution that the value that optimal solution is defined as each dimension in N-dimensional space is equal, can accurately by
The physical host screened is limited in so that near the optimal solution of the equilibrium situation of the resource of each dimension.For used
In the biggish situation of m value, may filter out more than one can distribute to the physical host of current container, and further judgement is every at this time
Can a physical host improve the resources balance situation of entire physical host cluster, and allowing to find improves entire physical host
The Resource Allocation Formula of the resources balance situation of cluster.The solution of the present invention improves traditional greedy algorithm, calculation amount
It is relatively small, and resource fragmentation can be effectively reduced, improve host close to the resource utilization under full load situation, it avoids
The wasting of resources.This deployment strategy improves about 100% in terms of being deployed in resource balancing degree compared to Greedy strategy after tested.
Detailed description of the invention
Embodiments of the present invention is further illustrated referring to the drawings, in which:
Fig. 1 is the schematic illustration of linux container technology；
Fig. 2 is the flow chart of the method for the resource deployment according to an embodiment of the invention for linux container；
Fig. 3 a is the feasible zone range that memory and cpu resource are directed to according to one embodiment of present invention in two-dimensional space
Schematic diagram；
Fig. 3 b is according to one embodiment of present invention in three-dimensional space for memory, CPU and network bandwidth resources
The schematic diagram of feasible zone range；
Fig. 4 shows using the solution of the present invention and carries out using the scheme of traditional greedy algorithm money when resource deployment
The simulation result of source balance factor；
Fig. 5 shows using the solution of the present invention and carries out using the scheme of traditional greedy algorithm object when resource deployment
Manage the simulation result of the comprehensive resources utilization rate of mainframe cluster.
Specific embodiment
Since the resource deployment for linux container needs to consider to filter out from a large amount of physical host respectively for more
The deployment strategy of a container, and the resource class considered needed for each physical host is also multidimensional, by traversing each physics
Host and each container match the resource of every dimension of each container, and calculation amount and time-consuming are often difficult to
Meet the requirement for system of building.Thus, the prior art generallys use greedy calculation when executing the resource deployment for being directed to linux container
Method only executes the search of limited calculation amount, preferred resource deployment scheme is determined in search range.
Inventor after research by having found, based on traditional resource deployment scheme using greedy algorithm, in host resource
Close to there are many unserviceable fragment resources in the case where saturation, i.e., when most host resource occupied situation
Under be also difficult to further increase the average utilization of resource by continuing searching suitable resource deployment scheme.Also, this feelings
Condition is also embodied in existing resource deployment scheme and is difficult to balancedly dispose the resource of multiple dimensions, i.e., as described in background technique
Ground, it is understood that there may be there are also remaining situations for the resource of another dimension when the resource of certain dimension is used up.And the application in container
Program is typically necessary while meeting the resources of multiple dimensions and can just run, and the case where above-mentioned resources left occurs and can then waste object
Resource, the reduction system of reason host can concurrently deployed number of containers.
In this regard, the invention proposes a kind of calculation resource disposition methods for linux container, it is with reference to the accompanying drawing and specific real
Mode is applied to elaborate to the present invention.
With reference to Fig. 2, according to one embodiment of present invention, which comprises
Step 1. determined according to the specific requirements of system building environment need the container disposed and its mirror image address, container it
Between dependence, all kinds of software and hardware resources needed for container.Here the dependence between container refers to the fortune of container A
Row need to be using container B as condition, such as container A needs to call container B or need to trigger container B etc. at runtime.Table 1 is shown
The information of container determined needed for according to one embodiment of present invention.
Table 1
Wherein, Container Name is for identifying each container；Mirror image address is for creating container；Hardware is bound to hold for recording
Device needs the hardware resource (such as GPU, network interface card, vCPU etc.) bound, and the container for being bundled with hardware resource then needs preferentially
Consideration meets its requirement；The resource of N number of dimension can be container using required all kinds of resources, such as CPU, GPU, memory,
Storage, network bandwidth, block I/O bandwidth etc..
Step 2. is carried out according to the dependence between each container, the size of stock number needed for each container, selection
The priority of each container when resource deployment, with according to the priority be each container allocation physical host.According to this
One embodiment of invention determines that the principle of the priority of each container is in this step, is preferably by other containers
The container of dependence and the big container allocation physical host of required stock number.Reason for doing so is that in order to prevent
After the complete physical host of the relatively small container allocation of resource occupation amount, remaining inadequate resource is big for resource occupation amount
Container used.Carrying out deployment based on the dependence between container is also for similar consideration, to prevent deployment
The case where the application on container cannot be executed smoothly.
According to other embodiments of the invention, in this step, it can artificially select to carry out each appearance when resource deployment
The priority of device.According to another embodiment of the present invention, the priority of each container can also be determined using following manner:
Step 2-1: compare with the presence or absence of dependence between two containers, if then continuing step 2-2, if otherwise continuing
Step 2-3；
Step 2-2: according to the dependence between container, if the application in container A needs to rely on answering in container B
With the priority of B being then set above to the priority of A, and continue step 2-4；
Step 2-3: judging whether it has bound hardware resource for container, has the container of hardware resource to be set as binding
With higher priority, and continue step 2-4；
Step 2-4: for there are multiple containers to have the case where equal priority, by the stock number of certain required dimension
Big container is set as having higher priority.
It determines as a result, and distributes physical host in what order for multiple containers.
Step 3. is directed to the container of current physical host to be allocated, all alternative physical hosts is traversed, for every
One physical host, according to the container to be built to the demand of whole multi dimensional resources and the physical host for described
The total amount and usage amount of multi dimensional resource calculate matching degree of each physical host in every dimension resource with the container.
According to one embodiment of present invention, can be according to the actual conditions of container to be allocated, it will obvious undesirable physics
Host excludes alternative physical host.
Here can using it is any it is appropriate by the way of calculate a physical host in the resource of any one dimension with institute
The matching degree of container is stated, such as whether the physical host meets the needs of container for the total amount of the resource of the dimension
Whether amount or the physical host meet the demand, etc. of the container for the unused amount of the resource of the dimension.
Due to whether will judge a physical host using the feasible zone of multiple dimensions in later step of the invention
For container use, processing, reduction calculation amount preferably walk herein according to one embodiment of present invention for convenience
Matching degree of the physical host in every dimension resource with the container is calculated by normalized mode in rapid, it will be different
The numerical value of the matching degree of dimension is unified to the identical order of magnitude.For example, calculating can be normalized using following formula:
ScoreD(p, c)=(useD+γD)/TD*100
Wherein, ScoreDThe matching degree in resource with container c, use are tieed up in D for physical host pDIt is that physical host p is directed to
The usage amount of the D dimension resource, TDIt is total amount of the physical host p for D dimension resource, γDIt is container c to described
The demand of D dimension resource.
Calculating formula in this way can calculate obtain a physical host in every dimension resource with the container
Normalized matching degree.
Step 4. is directed to the feasible zone range of N-dimensional resource according to determined by the feasible zone boundary value m of setting, filters out every
Matching degree in dimension resource is in one or more physical hosts within the scope of the feasible zone.Here feasible zone side
Dividing value m is the parameter for determining feasible zone range, and m is greater than 0, can be determined based on experience value or by user by test.
Biggish m corresponds to biggish feasible zone range, is conducive to filter out more satisfactory physical hosts, so that search result
Closer to traversal search, bigger calculation amount is also corresponded to.Lesser m corresponds to lesser feasible zone range, can subtract
The quantity for the physical host that container uses that is filtered out less distribute to, search process is closer to greedy algorithm, calculation amount
It is smaller.
According to one embodiment of present invention, for by the physical host in the resource of each dimension with the appearance
The matching degree of device is transformed into unified numerical intervals, such as based on aforementioned normalized calculating formula by the matching degree of each dimension
The case where being converted into percentage can according to need and feasible zone range is determined as in N-dimensional space at a distance from origin
Region less than m.
According to still another embodiment of the invention, feasible zone range can also be determined as in N-dimensional space with it is optimal
The distance of solution is less than the region of m, and optimal solution here refers to the comparable solution of the balance degree of each dimension in N-dimensional space.
The purpose for the arrangement is that physical host is each in the ideal case for the considerations of filtering out resource more balanced allocation plan
When essentially equal with the matching degree of container in dimension, the equilibrium degree of resource is optimal.However, for most situations, physics master
Matching degree of the machine in the resource of each dimension is simultaneously unequal.Such as standard is used in some technologies for solving resources balance
The mode that difference, variance calculate filters out the best scheme of equilibrium degree, however inventor thinks that such mode can not be complete
Avoid the occurrence of resource fragmentation.This is because, although standard deviation and variance can provide the degree for deviateing mean value on the whole,
It does not embody and deviates from ideal range, the object filtered out in this way with the presence or absence of the resource of some or several dimensions
Reason host is likely to appear in the situation that resource is excessive or very few in some or several dimensions.In contrast, using feasible zone
The generation of this case then can be effectively avoided in range.
Below by the feasible zone boundary value m for being illustrated how by two specific examples based on setting determine N-dimensional can
Row domain range.
Fig. 3 a is shown according to one embodiment of present invention, for the feasible zone range of 2 dimension resources, wherein 2 dimension
Resource is respectively memory and CPU.With reference to Fig. 3 a, when memory matching degree is equal with CPU matching degree, the equilibrium degree of resource is optimal,
Memory matching degree and the CPU matching degree for meeting the physical host of this condition are true by point (0,0) and point (100,100) institute in figure
On fixed straight line.It is referred to by the range that feasible zone boundary value m is determined true by memory matching degree and CPU the matching degree institute of host
The shortest distance between fixed point and the straight line of optimal solution is less than m.By taking Fig. 3 a as an example, point (0, d) indicate CPU matching degree be 0 and
Memory matching degree is d, it is assumed that the distance that the point reaches optimal solution straight line is m, thenIt is possible thereby to determine feasible zone model
It encloses (Feasible Region), i.e. the region between two dotted lines of Fig. 3 a, wherein a dotted line is by point (0, d) and point
(100-d, 100) is determined, another dotted line is determined by point (d, 0) and point (100,100-d), or is denoted as |
CPU matching degree-memory matching degree |Remaining region is fringe region (Edge Region).
It, can matching by a host in all N number of dimension resources with container after feasible zone range has been determined
Degree is compared with the feasible zone range, if entirely falling within the scope of the feasible zone, then it is assumed that the physical host meets resource
Balanced screening criteria.
By taking above-mentioned 2 dimension resource as an example, it is assumed that feasible zone boundary value m is 0.3, if the CPU matching degree of host and interior
Depositing matching degree is respectively 0.65 and 0.7, and the difference of the two is less thanThen the host meets screening criteria, if one
The CPU matching degree and memory matching degree of host are respectively 0.4 and 0.9, and the difference of the two is greater thanThen the host is not inconsistent
Close screening criteria.
Fig. 3 b is shown according to one embodiment of present invention, for the feasible zone range of 3 dimension resources, wherein 3 dimension
Resource is respectively memory, CPU and network bandwidth.The memory by host is referred to by the range that feasible zone boundary value m is determined
With most short between the point and the straight line of optimal solution on three-dimensional space determined by degree, CPU matching degree and network bandwidth matching degree
Distance is less than m.As shown in Fig. 3 b, feasible zone range is homogeneous with memory matching degree, CPU matching degree and network bandwidth matching degree
Deng straight line as central axes, using m as radius cylindrical body, the area of space other than the cylindrical body belongs to edge
Region.Above-mentioned feasible zone range can be indicated by following calculating formula: (Scorecpu-Scoremem)2+(Scorecpu-
Scorenet)2+ (Scorenet-Scoremem)2< 3m2
Wherein, Scorecpu、Scoremem、ScorenetRespectively indicate CPU matching degree, memory matching degree and network bandwidth
With degree.The host for meeting above formula meets screening criteria.
With 2 peacekeepings 3 tie up it is resources-type as, can also be directed to N-dimensional resource feasible zone range, and be based on the feasible zone model
It encloses and determines whether the physical host meets screening with the matching degree of container in the resource of each dimension with a physical host
Standard.In this regard, inventor summarizes the feasible zone range in N-dimensional space, meet following formula:
Wherein, xiFor the resource matched degree of i-th dimension, N is the number of dimensions of resource, and m is the feasible zone boundary value of setting.If
It is that a physical host meets the expression formula, then it is assumed that it meets screening criteria.
One or more physical hosts for meeting current container demand, and physics master can be filtered out by step 4
There is excessive or very few situation in the resource that machine does not have any one dimension, can meet the needs of container and money simultaneously
The standard of source equilibrium.If the physical host for meeting container demand can be filtered out just by feasible zone range, then may be used
It, then can be by all if the quantity of the physical host filtered out is greater than 1 the physical host is directly distributed to the container
As subsequent step is further selected.
Step 5. is directed to each physical host filtered out, calculates for indicating complete after assigning them to the container
The resources balance factor of the resources balance situation of portion's mainframe cluster, with according to the resources balance factor, selection is distributed to described
The physical host of container.The reason of so implementing this step is that a physical host can often be supported to dispose multiple simultaneously
Container is then likely to occur if only considering the matching degree between physical host and container and is deployed in a large amount of container together
Situation on one physical host, this may impact the resources balance situation of whole physical host clusters.Thus,
In this step, it may be considered that if current container is deployed on different alternative physical hosts to whole physical host clusters
The influence of resources balance situation, to filter out so that the optimal allocation result of the whole equilibrium situation of cluster.
According to one embodiment of present invention, when current container can be deployed on alternative host k based on following formula calculating
The resources balance factor of whole physical host clusters:
Wherein, Hostk_useDIt is that alternative host k has used resource situation, Host on dimension Di_useDIt is that other are standby
Host i is selected to use resource situation on dimension D.γDIt is demand of the container c to D dimension resource, N is resource
Number of dimensions, M are the quantity of physical host in cluster.
It can be seen that the resources balance factor by above-mentioned expression formula and be defined as the side that each multi dimensional resource of cluster uses
The mean value of difference, the value the big, indicates that the use of resource is more unbalanced.Therefore, the money for each alternative host can be compared
Source balance factor selects so that the above-mentioned the smallest physical host of the resources balance factor distributes to the container.
If it was found that acceptable basis is current when there is a situation where that the resources balance factor of more than one physical host is equal
Different coefficients is arranged for every dimension resource in the demand for the application to be run in container.For example, it is assumed that being run in current container
Application be the application of video code conversion class, such application to CPU, memory it is more demanding, therefore alternative host can be turned up in CPU
With the ratio in the two dimensions of memory with the matching degree of the container, such as the ratio of CPU is adjusted to
The ratio of memory is adjusted toThe ratio for keeping the resource of other dimensions isIn another example false
If the application run in current container is the application of machine learning class, such application is more demanding to GPU, can be by the ratio of GPU
It is adjusted toThe ratio for keeping the resource of other dimensions is
The priority of step 6. each container according to determined by through step 2, determines next physics master to be allocated
The container of machine, and the 3-5 that repeats the above steps, until completing resource deployment.
It can achieve the effect that verify deployment strategy according to the present invention, it is real that inventor has carried out following emulation
It tests, compares the effect using balanced deployment strategy and traditional Greedy strategy of the invention.Test when, be respectively adopted by
The mainframe cluster that 5-12 physical host is constituted carries out ten groups of experiments for each cluster and seeks average result.
Fig. 4 shows the mean value of the resources balance factor for whole physical hosts, wherein the broken line with point is traditional
Greedy strategy, the broken line without point are result of the invention.It can be seen that no matter containing how much physics master in mainframe cluster
Machine, the resources balance factor of the invention are respectively less than traditional prior art, and value is about the half of the prior art.This explanation
It can make entire mainframe cluster that there is very good resources balance situation using method of the invention.
Fig. 5 shows entire mainframe cluster to the mean value of the utilization rate of multi dimensional resource.In Fig. 5, dark column indicates this
Invention, light column indicate traditional Greedy strategy.As can be seen that in the utilization to the host in cluster close to the case where saturation
Under, better resource utilization can be obtained using the solution of the present invention, improve about 5%-10% compared to the prior art.
Inventor thinks that the reason of generating this effect is that balanced deployment strategy according to the present invention considers the equilibrium of multi dimensional resource
It uses, host can be improved close to the resource utilization under full load situation efficiently against resource fragmentation.It is contemplated that
It arrives, the solution of the present invention is particularly suitable for the extensive large data sets environment such as cloud computing.
It should be noted that each step introduced in above-described embodiment is all not necessary, those skilled in the art
Can carry out according to actual needs it is appropriate accept or reject, replacement, modification etc..
It should be noted last that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting.On although
The invention is described in detail with reference to an embodiment for text, those skilled in the art should understand that, to skill of the invention
Art scheme is modified or replaced equivalently, and without departure from the spirit and scope of technical solution of the present invention, should all be covered at this
In the scope of the claims of invention.
Claims (13)
1. a kind of calculation resource disposition method for linux container, comprising:
1) each physical host, conditions of demand and the physical host according to the container to be built to N-dimensional resource are directed to
The case where N-dimensional resource can be provided, matching degree of the physical host in the resource of every dimension with the container is calculated,
N > 1；
2) it in N-dimensional space, based on the feasible zone range determined by the feasible zone boundary value m being arranged, filters out every one-dimensional
The one or more physical hosts being in the resource of degree with the matching degree of the container within the scope of the feasible zone, m > 0.
2. according to the method described in claim 1, wherein step 1) include: by the physical host in the resource of each dimension
Unified numerical intervals are transformed into the matching degree of the container.
3. according to the method described in claim 2, wherein by following formula by the physical host in the resource of each dimension with
The matching degree of the container is transformed into unified numerical intervals:
ScoreD(p, c)=(uSeD+γD)/TD*100
Wherein, ScoreDThe matching degree in resource with the container c, use are tieed up in D for physical host pDIt is that physical host p is directed to
The usage amount of the D dimension resource, TD are total amount of the physical host p for D dimension resource, γDIt is container c to described
The demand of D dimension resource.
4. method described in any one of -3 according to claim 1, wherein feasible zone range described in step 2) is in N-dimensional sky
The interior region for being less than m at a distance from optimal solution, the optimal solution are the solution that the value of each dimension is equal in N-dimensional space.
5. according to the method described in claim 4, wherein judging whether a physical host meets using following formula in step 2)
It is within the scope of the feasible zone in the resource of every dimension with the matching degree of the container:
Wherein, xiFor the resource matched degree of i-th dimension, N is the number of dimensions of resource, and m is the feasible zone boundary value of setting.
6. according to the method in claim 2 or 3, wherein feasible zone range described in step 2) be in N-dimensional space with original
The distance of point is less than the region of m, and the origin is the point that the value of each dimension is 0 in N-dimensional space.
7. according to the method described in claim 1, wherein, further includes:
3) it for each physical host filtered out, calculates for indicating whole host sets after assigning them to the container
The resources balance factor of the resources balance situation of group；
4) according to the resources balance factor, selection is so that the optimal physical host point of the resources balance situation of whole mainframe clusters
Container described in dispensing.
8. according to the method described in claim 7, wherein step 3) is calculated as follows current container is deployed on alternative host k
When the resources balance factor:
Wherein, Hostk_useDIt is that alternative host k has used resource situation, Host on dimension Di_useDIt is that other are alternative main
Machine i has used resource situation on dimension D.γDIt is demand of the current container to D dimension resource, N is the dimension of resource
Degree amount, M are the quantity of physical host in cluster.
Also, step 4) includes: to select the smallest physical host of the resources balance factor as the object for distributing to the container
Manage host.
9. according to the method described in claim 8, wherein step 4) include: multiple physical hosts the resources balance factor it is identical
When, according to the resource that the application to be run in the container is stressed, adjust each dimension in the resources balance factor of physical host
Spend ratio shared by resource.
10. method described in any one of -3 according to claim 1, before step 1) further include:
0-1) determine the multiple containers for needing to build in system；
0-2) according to the dependence between the multiple container, the size of stock number needed for each container, selection is provided
The priority of each container when source is disposed, with according to the priority for each container allocation physical host.
11. according to the method described in claim 10, wherein step 0-2) include: container to be relied on by other containers, and/or
Need the container of hardware resource and/or the higher priority of container setting that required stock number is big.
12. a kind of computer readable storage medium, wherein being stored with computer program, the computer program is used when executed
In method of the realization as described in any one of claim 1-11.
13. a kind of for being directed to the resource deployment system of linux container, comprising:
Storage device and processor；
Wherein, the storage device is used for storing computer program, the computer program when being executed by the processor
In method of the realization as described in any one of claim 1-11.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811352854.0A CN109582461B (en) | 2018-11-14 | 2018-11-14 | Resource deployment method and system for Linux container |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811352854.0A CN109582461B (en) | 2018-11-14 | 2018-11-14 | Resource deployment method and system for Linux container |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109582461A true CN109582461A (en) | 2019-04-05 |
| CN109582461B CN109582461B (en) | 2021-03-09 |
Family
ID=65922272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811352854.0A Active CN109582461B (en) | 2018-11-14 | 2018-11-14 | Resource deployment method and system for Linux container |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109582461B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046102A (en) * | 2019-11-27 | 2020-04-21 | 复旦大学 | High-performance block chain service system based on ether house |
| CN112256430A (en) * | 2020-10-23 | 2021-01-22 | 北京三快在线科技有限公司 | Container deployment method, device, equipment and storage medium |
| CN112559129A (en) * | 2020-12-16 | 2021-03-26 | 西安电子科技大学 | Device and method for testing load balancing function and performance of virtualization platform |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102215168A (en) * | 2011-06-03 | 2011-10-12 | 黄东 | Method for optimizing and scheduling service resources based on laminated network |
| US8082549B2 (en) * | 2006-11-08 | 2011-12-20 | Board Of Regents, The University Of Texas System | System, method and apparatus for allocating resources by constraint selection |
| CN104834569A (en) * | 2015-05-11 | 2015-08-12 | 北京京东尚科信息技术有限公司 | Cluster resource scheduling method and cluster resource scheduling system based on application types |
| CN104881322A (en) * | 2015-05-18 | 2015-09-02 | 中国科学院计算技术研究所 | Method and device for dispatching cluster resource based on packing model |
| CN106502761A (en) * | 2016-10-18 | 2017-03-15 | 华南师范大学 | A kind of virtual machine deployment method of resources effective utilization |
| CN107562545A (en) * | 2017-09-11 | 2018-01-09 | 南京奥之云信息技术有限公司 | A kind of container dispatching method based on Docker technologies |
| US9886307B2 (en) * | 2014-11-21 | 2018-02-06 | International Business Machines Corporation | Cross-platform scheduling with long-term fairness and platform-specific optimization |
| CN107734052A (en) * | 2017-11-02 | 2018-02-23 | 华南理工大学 | The load balancing container dispatching method that facing assembly relies on |
| US20180081722A1 (en) * | 2016-09-20 | 2018-03-22 | International Business Machines Corporation | Multi-platform scheduler for permanent and transient applications |
| US20180276044A1 (en) * | 2017-03-27 | 2018-09-27 | International Business Machines Corporation | Coordinated, topology-aware cpu-gpu-memory scheduling for containerized workloads |
-
2018
- 2018-11-14 CN CN201811352854.0A patent/CN109582461B/en active Active
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8082549B2 (en) * | 2006-11-08 | 2011-12-20 | Board Of Regents, The University Of Texas System | System, method and apparatus for allocating resources by constraint selection |
| CN102215168A (en) * | 2011-06-03 | 2011-10-12 | 黄东 | Method for optimizing and scheduling service resources based on laminated network |
| US9886307B2 (en) * | 2014-11-21 | 2018-02-06 | International Business Machines Corporation | Cross-platform scheduling with long-term fairness and platform-specific optimization |
| CN104834569A (en) * | 2015-05-11 | 2015-08-12 | 北京京东尚科信息技术有限公司 | Cluster resource scheduling method and cluster resource scheduling system based on application types |
| CN104881322A (en) * | 2015-05-18 | 2015-09-02 | 中国科学院计算技术研究所 | Method and device for dispatching cluster resource based on packing model |
| US20180081722A1 (en) * | 2016-09-20 | 2018-03-22 | International Business Machines Corporation | Multi-platform scheduler for permanent and transient applications |
| CN106502761A (en) * | 2016-10-18 | 2017-03-15 | 华南师范大学 | A kind of virtual machine deployment method of resources effective utilization |
| US20180276044A1 (en) * | 2017-03-27 | 2018-09-27 | International Business Machines Corporation | Coordinated, topology-aware cpu-gpu-memory scheduling for containerized workloads |
| CN107562545A (en) * | 2017-09-11 | 2018-01-09 | 南京奥之云信息技术有限公司 | A kind of container dispatching method based on Docker technologies |
| CN107734052A (en) * | 2017-11-02 | 2018-02-23 | 华南理工大学 | The load balancing container dispatching method that facing assembly relies on |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111046102A (en) * | 2019-11-27 | 2020-04-21 | 复旦大学 | High-performance block chain service system based on ether house |
| CN111046102B (en) * | 2019-11-27 | 2023-10-31 | 复旦大学 | High performance blockchain service system |
| CN112256430A (en) * | 2020-10-23 | 2021-01-22 | 北京三快在线科技有限公司 | Container deployment method, device, equipment and storage medium |
| CN112559129A (en) * | 2020-12-16 | 2021-03-26 | 西安电子科技大学 | Device and method for testing load balancing function and performance of virtualization platform |
| CN112559129B (en) * | 2020-12-16 | 2023-03-10 | 西安电子科技大学 | Device and method for testing load balancing function and performance of virtualization platform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109582461B (en) | 2021-03-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022262167A1 (en) | Cluster resource scheduling method and apparatus, electronic device and storage medium | |
| US10129333B2 (en) | Optimization of computer system logical partition migrations in a multiple computer system environment | |
| TWI488111B (en) | System and method for translating program functions for correct handling of local-scope variables and computing system incorporating the same | |
| JP5967672B2 (en) | Database management system, computer, database management method | |
| US20150309842A1 (en) | Core Resource Allocation Method and Apparatus, and Many-Core System | |
| CN109684074A (en) | Physical machine resource allocation methods and terminal device | |
| CN109582461A (en) | A kind of calculation resource disposition method and system for linux container | |
| EP2472398A1 (en) | Memory-aware scheduling for NUMA architectures | |
| TWI786564B (en) | Task scheduling method and apparatus, storage media and computer equipment | |
| CN114356543A (en) | Kubernetes-based multi-tenant machine learning task resource scheduling method | |
| CN105893010A (en) | Host scheduling method and system used for container deployment | |
| WO2021141875A1 (en) | Automated local scaling of compute instances | |
| CN108241531A (en) | A kind of method and apparatus for distributing resource for virtual machine in the cluster | |
| US20240061712A1 (en) | Method, apparatus, and system for creating training task on ai training platform, and medium | |
| US9218198B2 (en) | Method and system for specifying the layout of computer system resources | |
| US20210390405A1 (en) | Microservice-based training systems in heterogeneous graphic processor unit (gpu) cluster and operating method thereof | |
| Ma et al. | vLocality: Revisiting data locality for MapReduce in virtualized clouds | |
| CN114386560A (en) | Data processing method and device | |
| CN113391913A (en) | Distributed scheduling method and device based on prediction | |
| CN114625500A (en) | Method and application for scheduling micro-service application based on topology perception in cloud environment | |
| CN113535087B (en) | Data processing method, server and storage system in data migration process | |
| CN110958192B (en) | Virtual data center resource allocation system and method based on virtual switch | |
| CN116661979A (en) | Heterogeneous job scheduling system and method | |
| CN114048040B (en) | Task scheduling method based on time delay relation between memory and image classification model | |
| CN116149843A (en) | Dynamic programming-based resource allocation method, network, storage medium and processor |
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 |