CN110018832B - Radar software component deployment strategy based on improved dynamic programming - Google Patents

Radar software component deployment strategy based on improved dynamic programming Download PDF

Info

Publication number
CN110018832B
CN110018832B CN201910124405.9A CN201910124405A CN110018832B CN 110018832 B CN110018832 B CN 110018832B CN 201910124405 A CN201910124405 A CN 201910124405A CN 110018832 B CN110018832 B CN 110018832B
Authority
CN
China
Prior art keywords
platform
resource
communication
processing
processor
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.)
Expired - Fee Related
Application number
CN201910124405.9A
Other languages
Chinese (zh)
Other versions
CN110018832A (en
Inventor
郑文荣
张云雷
席泽敏
颜礼彬
黄凯
陈冰
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.)
Naval University of Engineering PLA
Original Assignee
Naval University of Engineering PLA
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 Naval University of Engineering PLA filed Critical Naval University of Engineering PLA
Priority to CN201910124405.9A priority Critical patent/CN110018832B/en
Publication of CN110018832A publication Critical patent/CN110018832A/en
Application granted granted Critical
Publication of CN110018832B publication Critical patent/CN110018832B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F8/00—Arrangements for software engineering
    • G06F8/60—Software deployment
    • G06F8/61—Installation
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multi Processors (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a radar software component deployment strategy based on improved dynamic programming, which comprises a radar computing resource management and deployment strategy; the software requirement consists of a plurality of computing resource requirements, and a computing method of a standard dynamic planning cost function is improved on the basis of using the existing software and hardware resource model for reference; simulation results show that the improved dynamic planning strategy is adopted, so that the deployment time can be shortened, and the deployment success rate can be greatly improved.

Description

Radar software component deployment strategy based on improved dynamic programming
Technical Field
The invention relates to a radar platform resource deployment strategy, in particular to a radar software component deployment strategy based on improved dynamic programming.
Background
Software and hardware systems of the radar system are separated and mutually independent, the development of a software radar hardware platform can be separated from a specific application, various requirements are met, a platform with strong adaptability can be developed, various application programs can be effectively compatible, and the open system structure platform becomes the mainstream. However, different applications are developed on an open architecture, and then the application software is mapped to a general hardware platform, so that a series of problems such as the following need to be solved: how resources owned in a hardware platform can be efficiently described; how the resource requirements of an application can be efficiently described and how it can be ensured that the application is mapped onto the hardware platform stably and efficiently, etc.
Disclosure of Invention
The invention aims to provide a radar software component deployment strategy based on improved dynamic programming, which has the advantages of simulating a corresponding resource mapping optimization result and being applied to waveform resource configuration optimization of a software radar so as to solve the problems in the background art.
In order to achieve the purpose, the invention provides the following technical scheme: a radar software component deployment strategy based on improved dynamic programming comprises radar computing resource management;
the radar computing resource management comprises hardware computing storage resources and a communication resource model;
the hardware computing storage resource is as follows:
the processing equipment is responsible for running specific components, signal processing tasks and executing actual data processing, N platform processing equipment forms a certain interconnection topological structure on the application hardware platform processing resources, and the processing resources of the platform can be modeled into a processing equipment sequence with the length of N, namely:
C＝(C1，C2，...，CN)
the vector represents the device process model and ranks the processing power of the processors from P1 to PN to calculate the total processing power of the platform as CT＝C1+C2+...+CN；
The hardware communication resources are as follows:
the communication resources of the topology platform devices of different communications may be represented by a matrix B:
Figure GSB0000198795890000021
wherein the element BijRepresents a slave platform processor PiTo PjThe sum of all terms in the above equation is the total processing bandwidth B of the platformT；
The method comprises the following steps that T matching characteristics and T' distribution characteristics are described, the matching characteristics of processing equipment resources comprise processor types, processor models and development software attributes, and the matching characteristics are used for judging matching characteristics of slave applications to platforms in the deployment process; the allocation characteristics describe the capacity of processing resources, processors of different classes respectively adopt independent resource measurement standards to measure, the matching characteristics and the allocation characteristics are respectively modeled into one-dimensional vectors with lengths of T and T', each element describes one matching characteristic or allocation characteristic, therefore, a resource matching model and an allocation model of the platform processing equipment are finally modeled into a two-dimensional matrix, the abscissa of the matrix is identified as each processing equipment resource, and the ordinate of the matrix is identified as the specific matching characteristic and allocation characteristic;
the hardware platform communication resource model is as follows: the communication resources are responsible for data transmission between different processing devices. According to the topological structure of the platform interconnection technology, the method can be divided into point-to-point, shared bus and exchange architecture modes, and the communication and communication bandwidth information between processors need to be obtained in the modeling of the communication equipment;
the platform communication resource: the platform communication resource model is divided into a platform communication resource matching characteristic and a platform communication resource allocation characteristic, and the platform communication resource matching characteristic and the platform communication resource allocation characteristic are both a two-dimensional matrix of N multiplied by N;
platform communication resource matching model:
Figure GSB0000198795890000031
wherein, the element LijDescribing connectivity between a processing device i and a processing device j in the platform, wherein the value is 0 or 1, and 1 represents that the two are connected, and the former is not connected;
platform communication resource allocation model:
Figure GSB0000198795890000032
wherein, the element BijDescribing the communication bandwidth between the platform processing device i and the processing device j, the unit is MBPS, namely the transmission capability of megabytes per second, the diagonal line is infinity to represent that the communication bandwidth inside a single processor is infinite, if the communication bandwidth is 0, the communication bandwidth represents that no connection exists between the two processors, for the switching structure, the communication bandwidth between any two platform processors is limited by the communication bandwidth of a source end, a destination end and a switching node, and the communication bandwidth capability is the minimum.
Further, the software hardware calculates the storage resources: the application platform processes the resource model correspondences, assuming that one application is composed of M responsesComposed of components, therefore, the application processing resource requirement model needs to be modeled as a sequence of application components of length M, namely: c ═ c1，c2，...，cM) (4)
The application components are respectively modeled into one-dimensional vectors with the lengths of T and T 'by the T matching characteristic and the T' distribution characteristic for the resource requirements of the processing equipment, so that an applied processing equipment resource requirement matching model and a distribution model are finally modeled into a two-dimensional matrix, the abscissa of the two-dimensional matrix is marked as each application component, and the ordinate of the two-dimensional matrix is marked as the matching characteristic and the distribution characteristic of a specific component for the resource requirements of the target processing equipment; as with the platform processing resource model, the resource requirement matching characteristics of a component to a target processing device also include the type of processor required by the component, the type of processor, the operating system, and the development software attributes, and there may be multiple resource requirement matching and allocation characteristics for the same component. The allocation characteristics mainly describe the capacity of the components for the requirement of target processing resources, and the application components of different classes can be measured by adopting respective independent resource measurement references;
the software communication resources are as follows:
the communication resource requirements of the application components are used for describing the connectivity between any two application components and the communication bandwidth requirements between the two application components, wherein the former is a matching characteristic and the latter is an allocation characteristic;
the communication resource requirement matching model is as follows:
Figure GSB0000198795890000041
wherein, the element lijDescribing the connectivity between the component i and the component j in the application, wherein the value is 0 or 1, 1 represents that the two are connected, and 0 represents that the two are not connected;
the software communication resource demand distribution model comprises the following steps:
Figure GSB0000198795890000042
wherein, the element bijThe communication bandwidth between the application from the component i to the component j is described, and the unit is MBPS, namely the transmission capacity of megabyte per second, and the diagonal line 0 indicates that no communication needs exist between the single components;
furthermore, the matching requirement describes the real characteristic requirement of an application on platform configuration, the allocation requirement describes the requirement of application software on resources required to be allocated to the platform, and in the process of deploying from the software to the hardware platform, the matching characteristic determines whether the application and the platform are matched, and the subsequent deployment process can be carried out only after the matching attribute is met; and the allocation characteristics will determine whether the platform has the capability of applying the resource requirements for some aspect.
Further, the hardware computation storage resource model comprises available hardware resources, topology information of the platform processor, and the communication resource model is information describing connectivity and communication bandwidth between the processors.
The invention provides another technical scheme, a method for managing radar platform resources, which comprises the following steps;
s1: the cost function can guide the specific deployment process of the dynamic planning algorithm, and related resources can be reasonably distributed by limiting the cost function under the constraint of given system parameters. In particular, the cost function may organize the computational resources available and needed to manage;
s2: the cost is selected as the proportion of the calculation (storage and communication) of the soft price module needing to be deployed occupying the current hardware resource, and the cost is larger when the proportion is larger, as follows:
Figure GSB0000198795890000051
wherein
Figure GSB0000198795890000052
Figure GSB0000198795890000053
The meaning of each expression in the above formula is as follows:
Figure GSB0000198795890000054
a processing cost value of a processor computing power;
Figure GSB0000198795890000055
a cost value of the communication resource;
ch: component fh(ii) self-processing requirements;
Figure GSB0000198795890000056
handle assembly fhDeployment to processor Pk(l)After, processor Pk(l)The remaining processing capacity;
Figure GSB0000198795890000057
processing bandwidth requirements between the two components;
Figure GSB0000198795890000058
component deployment to processor Pk(l)After, processor Pk(l)And a processor P (f)u) The remaining bandwidth in between;
wherein q and 1-q are normalized weight factors, and the larger weight indicates that the resource is more important, possibly because the resource is more limited. In practical application, a reasonable weight value can be selected according to system requirements, and the default value is q-1-q-0.5;
s3: and sequentially selecting the optimal input path of each node in the path to backtrack, wherein the finally obtained path is the inverse process of the optimal deployment decision, thereby completing the deployment decision.
Compared with the prior art, the invention has the beneficial effects that:
according to the radar software component deployment strategy based on improved dynamic planning, software requirements consist of multiple computing resource requirements, and a computing method of a standard dynamic planning cost function is improved on the basis of the existing software and hardware resource model; simulation results show that the improved dynamic planning strategy is adopted, so that the deployment time can be shortened, and the deployment success rate can be greatly improved.
Drawings
FIG. 1 is a schematic diagram of the application deployment process of the present invention;
FIG. 2 is a first step in the deployment of a custom algorithm of the present invention;
FIG. 3 is a second step in the custom algorithm deployment of the present invention;
FIG. 4 is a third step in the custom algorithm deployment of the present invention;
FIG. 5 is a fourth step of custom algorithm deployment of the present invention;
FIG. 6 is a fifth step of custom algorithm deployment of the present invention;
FIG. 7 is a sixth step in the custom algorithm deployment of the present invention;
FIG. 8 is a seventh step in the custom algorithm deployment of the present invention;
FIG. 9 is a diagram of a custom algorithm deployment result of the present invention;
FIG. 10 is a comparison of the run times of the present invention;
FIG. 11 is a comparison of algorithm failure rates of the present invention;
FIG. 12 is a comparison of algorithm failure rates of the present invention;
FIG. 13 is a comparison graph of the run time of the present invention.
Detailed Description
The technical scheme in the embodiment of the invention will be made clear by combining the embodiment of the invention; fully described, it is to be understood that the described embodiments are merely exemplary of some, but not all, embodiments of the invention and that all other embodiments, which can be derived by one of ordinary skill in the art based on the described embodiments without inventive faculty, are within the scope of the invention.
Referring to fig. 1, a radar software component deployment strategy based on improved dynamic programming includes radar computing resource management;
the radar computing resource management comprises hardware computing storage resources and a communication resource model;
hardware computing storage resources:
the processing equipment is responsible for running specific components, signal processing tasks and executing actual data processing, N platform processing equipment forms a certain interconnection topological structure on the application hardware platform processing resources, and the processing resources of the platform can be modeled into a processing equipment sequence with the length of N, namely:
C＝(C1，C2，...，CN)
the vector represents the device process model and ranks the processing power of the processors from P1 to PN to calculate the total processing power of the platform as CT＝C1+C2+...+CN。
Hardware communication resources:
the communication resources of the topology platform devices of different communications may be represented by a matrix B:
Figure GSB0000198795890000071
wherein the element BijRepresents a slave platform processor PiTo PjThe available bandwidth in between. The internal bandwidth of the processor is high enough that the sum of all the terms in the above equation is the total processing bandwidth B of the platformT；
The method comprises the following steps that T matching characteristics and T' distribution characteristics are described, the matching characteristics of processing equipment resources comprise attributes such as processor types, processor models and development software, and the attributes are used for judging the matching characteristics of the slave applications to the platform in the deployment process; the allocation characteristics describe the capacity of processing resources, processors of different classes respectively adopt independent resource measurement references to measure, the matching characteristics and the allocation characteristics are respectively modeled into one-dimensional vectors with the lengths of T and T', and each element describes one matching characteristic or allocation characteristic. Therefore, the resource matching model and the allocation model of the platform processing equipment are finally modeled into a two-dimensional matrix, the abscissa of the matrix is marked as each processing equipment resource, and the ordinate of the matrix is marked as the specific matching characteristic and allocation characteristic;
hardware platform communication resource model: the communication resources are responsible for data transmission between different processing devices. According to the topological structure of the platform interconnection technology, the method can be divided into a point-to-point mode, a shared bus mode, an exchange architecture mode and the like, and the modeling of the communication equipment needs to obtain the information such as the connectivity between the processors, the communication bandwidth and the like;
platform communication resources: is the ability to describe the connectivity between any two processors, the former being a matching characteristic, and the latter being an allocation characteristic, and its communication bandwidth. Therefore, the platform communication resource model is divided into a platform communication resource matching characteristic and a platform communication resource allocation characteristic, and both the platform communication resource matching characteristic and the platform communication resource allocation characteristic are an NxN two-dimensional matrix;
platform communication resource matching model:
Figure GSB0000198795890000081
wherein, the element LijAnd describing the connectivity between the processing equipment i and the processing equipment j in the platform, wherein the value is 0 or 1, and 1 represents that the processing equipment i and the processing equipment j are connected, and the other is not.
Platform communication resource allocation model:
Figure GSB0000198795890000082
wherein, the element BijDescribing the communication bandwidth between the platform processing device i and the processing device j, the unit is MBPS, i.e. the transmission capability of megabytes per second, the diagonal line is infinity to indicate that the communication bandwidth inside a single processor is infinite, if 0, then no connection exists between the two processors, and for the switching fabric, the communication between any two platform processorsThe bandwidth is limited by the communication bandwidth of the source end, the destination end and the exchange node, and the communication bandwidth capability is the minimum.
Software hardware computing storage resources: the application platform processing resource models correspond to each other, and an application is assumed to be composed of M application components, so that the application processing resource demand model needs to be modeled as an application component sequence with a length of M, that is: c ═ c1，c2，...，cM) (4)
Further, the resource requirements of the application components for the processing equipment are obtained, and the resource requirements are respectively modeled into one-dimensional vectors with the lengths of T and T 'by the T matching characteristic and the T' distribution characteristic, so that the resource requirement matching model and the distribution model of the applied processing equipment are finally modeled into a two-dimensional matrix, the abscissa of the two-dimensional matrix is marked as each application component, and the ordinate of the two-dimensional matrix is marked as the matching characteristic and the distribution characteristic of the specific component for the resource requirements of the target processing equipment;
similar to the platform processing resource model, the resource requirement matching characteristics of the component for the target processing device also include attributes such as the type of processor required by the component, the type of processor, the operating system, and the development software, and for the same component, there may be multiple resource requirement matching and allocation characteristics. The allocation characteristics mainly describe the capacity of the components for the requirement of target processing resources, and the application components of different classes can be measured by adopting respective independent resource measurement benchmarks;
software communication resources:
the communication resource requirements of the application components are used for describing the connectivity between any two application components and the communication bandwidth requirements between the two application components, wherein the former is a matching characteristic, and the latter is an allocation characteristic;
communication resource requirement matching model:
Figure GSB0000198795890000091
wherein, the element lijDescribing the connectivity between the component i and the component j in the application, the value is 0 or 1, and 1 represents twoThe two are connected, and 0 means that the two are not connected.
Software communication resource demand allocation model:
Figure GSB0000198795890000101
wherein, the element bijThe communication bandwidth from component i to component j in application d is depicted in units of MBPS, i.e. megabytes per second of transmission capacity, and the diagonal line 0 indicates no communication demand between the individual components.
The matching requirement describes the real characteristic requirement of an application on platform configuration, the distribution requirement describes the requirement of application software on the resource required to be distributed by the platform, in the process of deploying from the software to the hardware platform, the matching characteristic determines whether the application and the platform are matched, and the subsequent deploying process can be carried out only after the matching attribute is met; and the allocation characteristics will determine whether the platform has the capability of applying the resource requirements for some aspect.
The hardware computing storage resource model comprises available hardware resources, topological structure of the platform processor and other information. The communication resource model is information describing connectivity between processors and communication bandwidth.
The invention provides another technical scheme, a radar software component deployment strategy based on improved dynamic programming comprises the following steps;
the method comprises the following steps: the cost function can guide the specific deployment process of the dynamic planning algorithm, and related resources can be reasonably distributed by limiting the cost function under the constraint of given system parameters. In particular, the cost function may organize the computational resources available and needed to manage;
step two: the cost is selected as the proportion of the calculation (storage and communication) of the soft price module needing to be deployed occupying the current hardware resource, and the cost is larger when the proportion is larger, as follows:
Figure GSB0000198795890000102
wherein
Figure GSB0000198795890000103
Figure GSB0000198795890000111
The meaning of each expression in the above formula is as follows:
Figure GSB0000198795890000112
a processing cost value of a processor computing power;
Figure GSB0000198795890000113
a cost value of the communication resource;
ch: component fh(ii) self-processing requirements;
Figure GSB0000198795890000114
handle assembly fhDeployment to processor Pk(l)After, processor Pk(l)The remaining processing capacity;
Figure GSB0000198795890000115
processing bandwidth requirements between the two components;
Figure GSB0000198795890000116
component deployment to processor Pk(l)After, processor Pk(l)And a processor P (f)u) The remaining bandwidth in between;
wherein q and 1-q are normalized weight factors, and the larger weight indicates that the resource is more important, possibly because the resource is more limited. In practical application, a reasonable weight value can be selected according to system requirements, and the default value is q-1-q-0.5;
step three: and sequentially selecting the optimal input path of each node in the path to backtrack, wherein the finally obtained path is the inverse process of the optimal deployment decision, thereby completing the deployment decision.
Comparative example:
the t algorithm (classical dynamic programming algorithm) is exemplified by:
first part applies software f1Is pre-assigned to each of the N processors. It checks N-1 t nodes { P ═ P1，f1And { P }2，f1H, if f1Is assigned to P1，
CT{P1，f1}＝p(f1)/P(P1，f1)＝2/4＝0.5
Occupies 50% of the processor resources, and 0.5;
if f1Is assigned to P2，CT{P2，f1}＝p(f1)/P(P2，f1)＝2/3＝0.66
Occupies 66% of the processor resources, and 0.66:
the computational and communication resources of the processor node are updated and the second part of the t-map sequentially processes the t-nodes between step 2 and step 4. In the subsequent deployment, the calculation cost of each component deployed in the processor and the communication cost generated by the leading component deployed in different processors need to be considered.
If f2The components being distributed at P1On the processor, consider f1Position of component, if f1The components being distributed at P1On the processor, at P1The deployment cost on the platform is:
WT(P1)＝p(f2)/P(P1，f2)＝1.5/2＝0.75，
at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)+b(f2)/B(P2，f2)＝1.5/4+0.75/1.5＝0.875。
CT(P1，P1，f2)＝CT{P1，f1}+WT(P1)＝1.25；
CT(P2，P1，f2)＝CT{P1，f1}+WT(P2)＝1.54。
the less expensive of each is f2The path of the component. Last path is P1、P1The computing resources and communication resources of the processor node are updated.
If f is2The components being distributed at P2On the processor, consider f1Position of component, if f1The components being distributed at P1On the processor, at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)+b(f2)/B(P2，f2)＝1.5/3+0.75/1.5＝1，
if f1The components are distributed at P2On the processor, at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)＝1.5/1＞1，
platform P2The residual calculation cost is less than the cost required by the software, the platform cannot process the residual calculation cost, and the result is expressed by infinity. CT (P)1，P2，f2)＝CT{P1，f1}+WT(P1)＝1.5；
So the last path is P1、P2The computing resources and communication resources of the processor node are updated.
When f is3The components are distributed at P2On the processor, consider the deployment position of the previous step algorithm, if f2The components being distributed at P1The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P1，f3)+b(f3)/B(P2，f3)＝0.5/3+0.5/1.5＝0.5，
if f2Component dispensingAt P2The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P2，f3)＝0.5/1.5＝0.33，
therefore, CT (P)1，P2，f3)＝CT{P1，f2}+WT(P3)＝1.25+0.5＝1.75；
CT(P2，P2，f3)＝CT{P2，f2}+WT(P3)＝1.5+0.33＝1.83
So the total cost of the former is small and the path is P1、P1、P2The computing resources and communication resources of the processor node are updated.
When f is3The components being distributed at P1On the processor, consider the deployment position of the previous step algorithm, if f2The components being distributed at P1The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P1，f3)＝0.5/0.5＝1
if the f2 component is allocated on the P2 processor at the cost of:
WT(P3)＝p(f3)/P(P1，f3)+b(f3)/B(P1，f3)＝0.5/2+0.5/0.75＝0.58
CT(P1，P1，f3)＝CT{P1，f2}+WT(P3)＝1.25+1＝2.25；
CT(P2，P1，f3)＝CT{P2，f2}+WT(P3)＝1.5+0.58＝2.083
the latter is less costly overall, with path P1、P2、P1The computing resources and communication resources of the processor node are updated.
When f is4The components being distributed at P1On the processor, consider the deployment position of the previous step algorithm, if f4The components being distributed at P1The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4)＝0.5/1.5+0.5/1＝0.83
if f4The components are distributed at P2The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)＝0.5/0.5＝1
CT(P1，P1，f3)＝CT{P1，f2}+WT(P4)＝2.08+0.83＝2.92；
CT(P2，P1，f3)＝CT{P2，f2}+WT(P4)＝1.75+1＝2.75
the latter is therefore less costly overall, with a path P1、P1、P2、P1The computing resources and communication resources of the processor node are updated.
When f is4The components being distributed at P2On the processor, consider the deployment position of the previous step algorithm, if f3The components being distributed at P1The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4) Because b (f)4)＞B(P1，f4) The communication resource is insufficient, the platform cannot process the communication resource, and the result is represented by infinity;
if f3The components are distributed at P2The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4) Because b (f)4)＞B(P1，f4) The communication resource is insufficient, the platform cannot process the communication resource, and the result is expressed by infinity;
both steps cannot be realized, so the final result is also infinity and the algorithm fails.
The final result is obtained, one path is successful, and the final path is P1、P1、P2、P1。
Our improved algorithm:
the algorithm t conducts the deployment cost of each stage to the next stage, which causes the effect of the first stage deployment to affect the deployment decisions of all the subsequent steps, and it is unreasonable to adopt the accumulated cost function calculation strategy from this point. In view of this, we propose an improved algorithm without accumulation. The cost function only considers the current cost value and does not accumulate with the previous term. The specific process is as follows:
software component f1Is pre-assigned to each of the N processors. Check N-1 t nodes { P ═ P1，f1And { P }2，f1H, if f1Is assigned to P1，
CT{P1，f1}＝p(f1)/P(P1，f1)＝2/4＝0.5
Occupies 50% of the processor resources, and is 0.5;
if f1 is assigned to P2, CT { P }2，f1}＝p(f1)/P(P2，f1)＝2/3＝0.66
Occupying 66% of the processor resources, and 0.66. As shown in fig. 2:
the computational and communication resources of the processor node are updated and the second part sequentially processes the nodes between step 2 and step 4. In the subsequent deployment, only the deployment cost is considered, namely the calculation cost of each component deployed in the processor and the communication cost generated by the precursor component deployed in different processors.
If f2The components being distributed at P1On the processor, consider f1Position of component, if f1The components being distributed at P1On the processor, at P1The deployment cost on the platform is:
WT(P1)＝p(f2)/P(P1，f2)＝1.5/2＝0.75，
at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)+b(f2)/B(P2，f2)＝1.5/4+0.75/1.5＝0.875。
low deployment cost, and f2The component deployment path is P1，P1As shown in fig. 3.
If f2The components being distributed at P2On the processor, consider f1Position of component, if f1The components being distributed at P1On the processor, at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)+b(f2)/B(P2，f2)＝1.5/3+0.75/1.5＝1，
if f1The components being distributed at P2On the processor, at P2The deployment cost on the platform is:
WT(P2)＝p(f2)/P(P1，f2)＝1.5/1＞1，
platform P2The residual calculation cost is less than the cost required by software, the platform cannot process the residual calculation cost, and the result is expressed by infinity. Low deployment cost, and f2The component deployment path is P1，P2As shown in fig. 4.
When f is3The components being distributed at P1On the processor, consider the deployment position of the previous step algorithm, if f2The components are distributed at P1The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P1，f3)＝0.5/0.5＝1
if f2The components being distributed at P2The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P1，f3)+b(f3)/B(P1，f3)＝0.5/2+0.5/0.75＝0.58
cost of deployment, f3The component deployment path is P1，P2，P1As shown in fig. 5.
When f is3The components being distributed at P2On the processor, consider the deployment position of the previous step algorithm, if f2The components being distributed at P1The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P1，f3)+b(f3)/B(P2，f3)＝0.5/3+0.5/1.5＝0.5，
if f2The components being distributed at P2The cost of deployment on the processor is:
WT(P3)＝p(f3)/P(P2，f3)＝0.5/1.5＝0.33。
low deployment cost, and f3The component deployment path is P1，P2，P2As shown in fig. 6.
When f is4The components being distributed at P1On the processor, consider the deployment position of the previous step algorithm, if f4The components being distributed at P1The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4)＝0.5/1.5+0.5/1＝0.83
if f4The components being distributed at P2The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4)＝0.5/2+0.5/0.75＝0.92
low deployment cost, and f4The component deployment path is P1，P2，P1，P1As shown in fig. 7.
When f is4The components being distributed at P2On the processor, consider the deployment position of the previous step algorithm, if f3The components are distributed at P1The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4) Because b (f)4)＞B(P1，f4) The communication resource is insufficient, the platform cannot process the communication resource, and the result is expressed by infinity;
if f3The components being distributed at P2The cost of deployment on the processor is:
WT(P4)＝p(f4)/P(P1，f4)+b(f4)/B(P1，f4) Because b (f)4)＞B(P1，f4) The communication resource is insufficient, the platform cannot process the communication resource, and the result is expressed by infinity;
neither step can be implemented, so the final result is also infinity and an algorithm failure, as shown in fig. 8.
The final result is obtained, one path is successful, and the final path is P1、P2、P2、P1As shown in fig. 9.
Aiming at the real-time online mapping requirement of any waveform in a software radar, on the basis of uniformly describing software and hardware resources, the utilization rate of the maximized hardware resources is taken as a deployment optimization cost function, software deployment is carried out by adopting an improved dynamic programming cost function, and finally Monte Carlo simulation is carried out on a dynamic optimization algorithm to obtain a radar resource mapping optimization result.
In the process of waveform deployment, firstly, an approximate range of a processor meeting the deployment requirement of an application component is determined according to the matching characteristics of the application component and the processor, and then the processor with the minimum deployment cost function is continuously searched in the pre-selected processors, so that the calculation of a large number of impossible deployment scheme cost functions can be reduced, the effects of greatly reducing the complexity of an algorithm and improving the waveform deployment efficiency can be achieved, and waveform random parameters are set as follows:
the number of waveform components is M-18;
connectivity con ═ 0.15;
the processor requirement is uniform distribution U12500 MOPS;
the bandwidth requirement is uniformly distributed U1500 Mbps;
the algorithm is respectively deployed on four different platforms, please refer to fig. 10-12, so that it can be seen that the deployment time and failure rate of the same algorithm of different platforms are slightly different, but the overall comparison is not very different, compared with the standard dynamic programming algorithm, the improved algorithm of the same platform has the advantages of less budget time, greatly reduced failure rate and obviously optimized algorithm quality.
To sum up, based on the radar software component deployment strategy based on improved dynamic programming, the software requirement is composed of multiple computing resource requirements, and a computing method of a standard dynamic programming cost function is improved on the basis of taking the existing software and hardware resource model as reference; simulation results show that the improved dynamic planning strategy is adopted, so that the deployment time can be shortened, and the deployment success rate can be greatly improved.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and inventive concepts of the present invention are equivalent or changed and shall be covered by the scope of the present invention.

Claims (5)

1. A radar software component deployment strategy based on improved dynamic programming is characterized by comprising radar computing resource management;
the radar computing resource management comprises hardware computing storage resources and a communication resource model;
the hardware computing storage resource is as follows:
the processing equipment is responsible for running specific signal processing tasks and actual data processing, N platform processing equipment form a certain interconnection topological structure on the application hardware platform processing resource, the processing resource of the platform can be modeled into a processing equipment sequence with the length of N, namely: vector quantity
C＝(C1，C2，…，CN)
The vectors represent plant process models and rank the processing power of the processors, from P1To PNComputing the total processing capacity of the platform as CT＝C1+C2+…+CN；
The hardware communication resources are as follows:
the communication resources of the topology platform devices of different communications may be represented by a matrix B:
Figure FSB0000198795880000011
wherein the element BijRepresents a slave platform processor PiTo PjThe internal bandwidth of the processor is regarded as infinite, and the sum of all terms in the above equation is the total processing bandwidth B of the platformT；
The method comprises the following steps that T matching characteristics and T' distribution characteristics are described, the matching characteristics of processing equipment resources comprise processor types, processor models and development software attributes, and the matching characteristics are used for judging matching characteristics of slave applications to platforms in the deployment process; the allocation characteristics describe the capacity of processing resources, processors of different classes respectively adopt respective independent resource measurement standards to measure, the matching characteristics and the allocation characteristics are respectively modeled into one-dimensional vectors with the lengths of T and T', each element describes one matching characteristic or allocation characteristic, therefore, a resource matching model and an allocation model of the platform processing equipment are finally modeled into a two-dimensional matrix, the abscissa of the matrix is identified as each processing equipment resource, and the ordinate of the matrix is identified as the specific matching characteristic and allocation characteristic;
the hardware platform communication resource model is as follows: the communication resources are responsible for data transmission among different processing devices, and can be divided into point-to-point, shared bus and exchange architecture modes according to the topological structure of the platform interconnection technology, and the modeling of the communication devices needs to obtain the connectivity and communication bandwidth information among the processors;
the platform communication resource: the platform communication resource model is divided into a platform communication resource matching characteristic and a platform communication resource allocation characteristic, and the platform communication resource matching characteristic and the platform communication resource allocation characteristic are both a two-dimensional matrix of N multiplied by N;
platform communication resource matching model:
Figure FSB0000198795880000021
wherein, the element LijDescribing connectivity between a processing device i and a processing device j in the platform, wherein the value is 0 or 1, and 1 represents that the two are connected, and the former is not connected;
platform communication resource allocation model:
Figure FSB0000198795880000022
wherein, the element BijDescribing the communication bandwidth between the platform processing device i and the processing device j, the unit is MBPS, namely the transmission capability of megabytes per second, the diagonal line is infinity to represent that the communication bandwidth inside a single processor is infinite, if the communication bandwidth is 0, the communication bandwidth represents that no connection exists between the two processors, for the switching structure, the communication bandwidth between any two platform processors is limited by the communication bandwidth of a source end, a destination end and a switching node, and the communication bandwidth capability is the minimum.
2. The radar software component deployment strategy based on improved dynamic programming according to claim 1, wherein: the software hardware calculates the storage resources: the application platform processing resource models correspond to each other, and an application is assumed to be composed of M application components, so that the application processing resource demand model needs to be modeled as an application component sequence with a length of M, that is: c ═ c1，c2，…，cM) (4)
The software communication resources are as follows:
the communication resource requirements of the application components are used for describing the connectivity between any two application components and the communication bandwidth requirements between the two application components, wherein the former is a matching characteristic and the latter is an allocation characteristic;
the communication resource requirement matching model is as follows:
Figure FSB0000198795880000031
wherein, the element lijDescribing the connectivity between the component i and the component j in the application, wherein the value is 0 or 1, 1 represents that the two are connected, and 0 represents that the two are not connected;
the software communication resource demand allocation model comprises the following steps:
Figure FSB0000198795880000032
wherein, the element bijThe communication bandwidth of an application from component i to component j is described in units of MBPS, i.e., megabytes per second of transmission capacity, and the diagonal line 0 indicates no communication demand between the individual components.
3. The radar software component deployment strategy based on improved dynamic programming according to claim 1, wherein: the matching requirement describes the real characteristic requirement of an application on platform configuration, the distribution requirement describes the requirement of application software on the resource required to be distributed by the platform, in the process of deploying from the software to the hardware platform, the matching characteristic determines whether the application and the platform are matched, and the subsequent deploying process can be carried out only after the matching attribute is met; and the allocation characteristics will determine whether the platform has the capability of applying the resource requirements for some aspect.
4. The radar software component deployment strategy based on improved dynamic programming according to claim 1, wherein: the hardware computing storage resource model comprises available hardware resources, topological structure information of the platform processors, and the communication resource model is information describing connectivity and communication bandwidth between the processors.
5. An improved dynamic programming based radar software component deployment strategy as claimed in claim 1, comprising the steps of;
s1: the cost function can guide the specific deployment process of the dynamic planning algorithm, and related resources can be reasonably distributed by limiting the cost function under the constraint of given system parameters; in particular, the cost function may organize the computational resources available and needed to manage;
s2: the cost is selected as the proportion of the calculation (storage and communication) of the soft price module needing to be deployed occupying the current hardware resource, and the cost is larger when the proportion is larger, as follows:
Figure FSB0000198795880000041
wherein
Figure FSB0000198795880000042
The meaning of each expression in the above formula is as follows:
Figure FSB0000198795880000043
a processing cost value of a processor computing power;
Figure FSB0000198795880000044
a cost value of the communication resource;
ch: component fh(ii) self-processing requirements;
Figure FSB0000198795880000045
handle assembly fhDeployment to processor Pk(l)After, processor Pk(l)The remaining processing capacity;
Figure FSB0000198795880000046
processing bandwidth requirements between the two components;
Figure FSB0000198795880000047
component deployment to processor Pk(l)After, processor Pk(l)And a processor P (f)u) The remaining bandwidth in between;
wherein q and 1-q are normalization weight factors, the larger the weight is, the more important the resource is, and the possible reason is that the resource limitation degree is larger; in practical application, a reasonable weight value can be selected according to system requirements, and the default value is q-1-q-0.5;
s3: and sequentially selecting the optimal input path of each node in the path to backtrack, wherein the finally obtained path is the inverse process of the optimal deployment decision, thereby completing the deployment decision.
CN201910124405.9A 2019-02-20 2019-02-20 Radar software component deployment strategy based on improved dynamic programming Expired - Fee Related CN110018832B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910124405.9A CN110018832B (en) 2019-02-20 2019-02-20 Radar software component deployment strategy based on improved dynamic programming

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910124405.9A CN110018832B (en) 2019-02-20 2019-02-20 Radar software component deployment strategy based on improved dynamic programming

Publications (2)

Publication Number Publication Date
CN110018832A CN110018832A (en) 2019-07-16
CN110018832B true CN110018832B (en) 2022-07-12

Family

ID=67188936

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910124405.9A Expired - Fee Related CN110018832B (en) 2019-02-20 2019-02-20 Radar software component deployment strategy based on improved dynamic programming

Country Status (1)

Country Link
CN (1) CN110018832B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103297983A (en) * 2013-05-06 2013-09-11 南京邮电大学 Wireless sensor network node dynamic deployment method based on network flow
CN107332913A (en) * 2017-07-04 2017-11-07 电子科技大学 A kind of Optimization deployment method of service function chain in 5G mobile networks
CN109245932A (en) * 2018-09-20 2019-01-18 北京计算机技术及应用研究所 A kind of security function service chaining dispositions method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8291378B2 (en) * 2008-07-29 2012-10-16 International Business Machines Corporation Simplified deployment modeling

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103297983A (en) * 2013-05-06 2013-09-11 南京邮电大学 Wireless sensor network node dynamic deployment method based on network flow
CN107332913A (en) * 2017-07-04 2017-11-07 电子科技大学 A kind of Optimization deployment method of service function chain in 5G mobile networks
CN109245932A (en) * 2018-09-20 2019-01-18 北京计算机技术及应用研究所 A kind of security function service chaining dispositions method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
基于资源适配的软件无线电应用自动部署算法;宋春磊等;《微电子学与计算机》;20171005(第10期);全文 *

Also Published As

Publication number Publication date
CN110018832A (en) 2019-07-16

Similar Documents

Publication Publication Date Title
Alresheedi et al. Improved multiobjective salp swarm optimization for virtual machine placement in cloud computing
CN110688219B (en) Adaptive weight load balancing algorithm based on reverse chaotic cuckoo search
CN111861793B (en) Distribution and utilization electric service distribution method and device based on cloud edge cooperative computing architecture
CN114340016B (en) Power grid edge calculation unloading distribution method and system
CN105094970B (en) The method of more times scheduling models of task can be divided under a kind of solution distributed system
CN109947574A (en) A kind of vehicle big data calculating discharging method based on mist network
CN108111335A (en) A kind of method and system dispatched and link virtual network function
Shojaee et al. A new cat swarm optimization based algorithm for reliability-oriented task allocation in distributed systems
CN112379985A (en) Computing task allocation method and device in cloud edge computing environment
CN112764940A (en) Multi-stage distributed data processing and deploying system and method thereof
CN111159859A (en) Deployment method and system of cloud container cluster
CN110018832B (en) Radar software component deployment strategy based on improved dynamic programming
CN105335376B (en) A kind of method for stream processing, apparatus and system
Yujie et al. An effective controller placement algorithm based on clustering in SDN
Yang et al. Trust-based scheduling strategy for cloud workflow applications
CN111049900A (en) Internet of things flow calculation scheduling method and device and electronic equipment
CN116841710A (en) Task scheduling method, task scheduling system and computer storage medium
Bouroudi et al. Robust deep reinforcement learning algorithm for VNF-FG embedding
WO2014137349A1 (en) Cloud application bandwidth modeling
Fan et al. Associated task scheduling based on dynamic finish time prediction for cloud computing
CN113992520B (en) Virtual network resource deployment method and system
Klimenko et al. The comparative estimation of workload relocation approaches in the fog-and edge-computing environments
CN112615736B (en) Delay optimal distributed NNs collaborative optimization method facing linear edge network
CN110262915B (en) Multi-state system reliability evaluation method based on edge value multi-valued decision diagram
CN107341057A (en) A kind of data processing method and device

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
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220712

CF01 Termination of patent right due to non-payment of annual fee