CN104035865B - Method for deployment optimization of monitoring points in operating process of programs - Google Patents

Abstract

The invention relates to a method for deployment optimization of monitoring points in the operating process of programs. The method includes the steps that firstly, monitored and executed program codes are represented through a hierarchical structure model; secondly, module information is read, and a monitoring point deployment scheme is set as an initial solution; thirdly, an objective function E(M) for evaluating the monitoring points is constructed, and an objective function value of the initial solution is calculated and stored; fourthly, the module information, the initial solution and the objective function value are read, a simulated annealing algorithm is conducted, and then an optimized monitoring point deployment scheme is obtained. Compared with the prior art, the method has the advantages of being simple and flexible in process, high in practicality and efficiency and the like and solves the problem that the monitoring points are not reasonable deployed in the operating process of the programs.

Description

A kind of monitoring point optimization dispositions method when program is run

Technical field

The present invention relates to a kind of monitoring point dispositions method, especially relate to monitoring point optimization deployment when a kind of program is run Method.

Background technology

In checking when program is run, produce monitor generally from system program demand, monitor is in program structure The position of setting is monitoring point, by the execution of monitoring program, monitor checks whether program operation process meets system program Demand, one side monitor needs to detect mistake when program is run, and on the other hand needs to find mistake in time, repaiies for mistake The multiple time providing abundance and space, these are all with the deployment scheme close relation of monitoring point, the deployment scheme of monitoring point Both needed all to cover each several part in program, and needed to consider monitoring point, monitored area, the optimum power between the monitoring time again Weighing apparatus, if the method for monitoring point is also increasingly paid attention to by technical staff in therefore reasonable deployment program, and proposes Advocate:

1) havelund et al. carries in " synthesizing monitors for safety properties " literary composition Go out a kind of algorithm, it is read the expression formula into a past tense linear time temporal logic and then dynamically generates generation corresponding with expression formula Code, this Codabar code will analyze the Perform sequence of finite length in linear session, and whether judgement sequence of obtaining a result meets use The stipulations that family is formulated.This algorithm is referred to as monitor composition algorithm in article, has realization in pathexplorer project, this It is the instrument of a run time verification also under development, this article is patrolled it is also mentioned that algorithm also can be developed in the future describing others Collect preferably to be applied to real-time system, or find the latent fault of multi-threaded system.

2) bonakdarpour et al. propose in " sampling based runtime verification " " when Between triggering sampling " run time verification, every the time internal trigger sampling action of a segment length, this sampling interval is by designer Depend on the circumstances.This sampling interval of this article is set to the time interval the shortest of all variable change on the other hand, this Sample can once sample in can all correlated variabless of pull-in range change, thus reducing the state of system under test (SUT).This article solves How to reduce space expense (variable change log buffer) simultaneously, maximizing the optimization problem of acquisition interval.This article Optimization aspect based on making system stipulations, the variable in the range of article algorithm discussion is all to participate in ltl stipulations meter The variable calculated, thus preferably go back the state change of original system with relatively small number of record space.

3) stoller et al. is based in " runtime verification with state estimation " literary composition The run time verification scheme of " Time Triggered sampling ", finds to have logout in status switch during sampling system execution sometimes Loss, propose a challenge to accurately carrying out run time verification.This article is held to system using HMM Row model is modeled, and adjusts the correlation of HMM from the example of system Perform sequence by baum-welch algorithm Transition probability between parameter, such as hidden state, thus the probability of the generation for estimating some sequence is prepared.This article assumes system Loss partial-length in Perform sequence is l, is then estimated in a certain bar Perform sequence according to the fsm that ltl stipulations generate What kind of event node is vacancy should fill out thus meeting the fsm of correlation, then using the Hidden Markov mould that precondition is good Type is estimating the probability sum of be likely to occur sequence.

Content of the invention

The purpose of the present invention is exactly to overcome the defect of above-mentioned prior art presence to provide a kind of reliability high journey Monitoring point optimization dispositions method during sort run.

The purpose of the present invention can be achieved through the following technical solutions:

A kind of monitoring point optimization dispositions method when program is run, comprising:

Step s1: monitored program code hierarchy Model is represented, that is, is divided into disparate modules, including 1 Top-level module m₀With multiple submodule m, obtain the oriented hierarchy chart of tree, tree is preserved as module information；

Step s2: read module information, if a monitoring point deployment scheme is as initial solution, the mould of stationary monitoring wanted in record Block；

Step s3: construction evaluates object function e (m) of monitoring point, and calculates the target function value of initial solution and preserve:

E (m)=g (m₀)*∑_c(m_x)/(1+∑r(m_x)),m_x∈m (1)

Wherein, m_xIt is x-th submodule in module collection m, c (m_x) it is submodule m_xMonitoring overhead, r (m_x) it is submodule Block m_xMonitoring redundancy, g (m₀)^-1=p (m₀), g (m₀) it is top-level module m₀Monitoring granularity, p (m₀) represent is current The lower top-level module m of observation strategy₀It is divided into many small parts；

Step s4: read module information, initial solution and target function value, it is simulated annealing algorithm, obtain an optimization Monitoring point deployment scheme.

Top-level module m under Current observation strategy₀It is divided into many small parts and adopt p (m₀) represent, p (m₀) refer to a journey The degree that sequence code is divided in monitoring, is defined as follows:

$p\left(m_0\right)=\left\{\begin{array}{c}1,d\left(m\right)=0,m\&element;o\left(m_0\right)\\ 1,d\left(m\right)=0,m\&notelement;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),s\left(m\right)=next\left(m\right),m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),s\left(m\right)=0,o\left(m\right)\&notequal;\mathrm{\&phi;},m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),m\&notelement;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),0<s\left(m\right)<next\left(m\right),m\&element;o\left(m_0\right)\end{array}\right.---\left(2\right)$

Wherein, d (m) is the depth of submodule m, o (m₀) it is top-level module m₀In monitored submodule set, h (m) It is the set of next straton module of submodule m, s (m), next (m) represent respectively and determine quilt in next straton module of submodule m The submodule of monitoring and submodule set of blocks.

Described monitoring redundancy r (m_x) refer to that submodules at different levels remark additionally the degree of higher level's modules, are defined as follows:

$r\left(m_x\right)=f\left(m_x\right),\&exists;m^{\&prime;},path\left(m^{\&prime;}\right)\&element;paths\left(m^{\&prime;}\right),m_x\&element;path\left(m^{\&prime;}\right),m^{\&prime;},m_x\&element;o\left(m_0\right)---\left(3\right)$

Wherein, path (m') is the node set with regard to module m', and paths (m') represents all roads that slave module m' is set out The set in footpath, it is assumed that there are module m₁, and there is Module nodes set path (m₁)={ m₁,m₂,...,m_d+1, each of which Individual m_k∈h(m_k-1), d is module m₁Depth, k=1,2,3 ..., d+1, slave module m₁The collection in all paths set out is combined into paths(m₁), path (m₁)∈paths(m₁)；Assume there is module m_xAnd m_y, wherein m_x∈h(m_y), f (m_x) it is in module m_xSend out The probability of raw fault, is defined as f (m_x)=f (m_y)/next(m_y), next (m_y) representation module m_yNext level submodule total Number.

Simulated annealing in described step s4 concretely comprises the following steps:

401: given initial temperature t₀, primary iteration number of times i₀, initial solution is the current solution of first time iteration, present energy e₀Target function value for initial solution；

402: the module that can refine monitoring to each of current solution calculates the neighborhood that candidate is currently solved, and remembers Record is in buffering；

403: if neighborhood is not empty, execution step 404, if neighborhood is sky, simulated annealing terminates, and output is current Solution；

404: the monitoring overhead of each of neighborhood according to current solution member's scheme, select the minimum conduct of expense new Solution；

405: calculate monitoring redundancy, monitoring granularity and the monitoring overhead of new explanation, and then the target function value of calculating new explanation, The i.e. energy e of new explanation₁；

406: work as e₁＜ e₀Or work as e₁≥e₀, calculate Probability p r [accept]=exp (- (e₁-e₀)/t), pr [accept] ＜ R, r are an equally distributed random number between 0 to 1, then accepting new explanation is current solution, even e₀=e₁, adjust stationary monitoring Module collection, deletes about being refined the related choosing solution of the module of monitoring, current iteration number of times i=i-1 in neighborhood, and executes step Rapid 408；

407: work as e₁≥e₀, pr [accept] >=r, current iteration number of times i=i-1；

408: if i ＞ 0, return to step 402, if i=0, make Current Temperatures t=0.95t, iterationses i=i₀；

409: if t=0, simulated annealing terminates, output is current to be solved, if t ≠ 0, return to step 402.

Compared with prior art, the invention has the advantages that

1) present invention carries out, to the execution of system under test (SUT), the structural model that abstract modeling draws a program, and proposes The measure of supporting evaluation monitoring scheme, makes program structure become apparent from, simultaneously can be according to evaluating tolerance to monitoring Point deployment scheme is verified the reliability it is ensured that deployment scheme, and then ensures that monitor can when program is run and is monitored Quickly and accurately to find mistake.

2) algorithm frame of the structural model based on program for the present invention and simulated annealing, obtains monitoring point optimization deployment Scheme, decreases deployment time, improves the efficiency for complicated process Optimization deployment monitoring point, may be used on practical programs execution It is achieved that the prototype experiment instrument of a run time verification in system, carry out the execution of dynamic authentication structured program.

Brief description

Fig. 1 is the structural representation of the present invention；

Fig. 2 is the comparison diagram of the initial solution of modular structure and observation station deployment and last solution in example；

Fig. 3 is simulated annealing each iteration energy and temperature change in example.

Specific embodiment

The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention Premised on implemented, give detailed embodiment and specific operating process, but protection scope of the present invention be not limited to Following embodiments.

As shown in figure 1, taking the deployment of the observation station of the execution of lwip protocol stack as a example, test platform: intel (r) core (tm) 2quad 2.66ghz, internal memory 2gb, stm32f107 72mhz, internal memory 64kb；Operating system: windows xp, need to monitor One section of execution from listen state to syn_rcvd state of lwip protocol stack, monitoring point optimization portion when a kind of program is run Arranging method, comprising:

Step s1: monitored program code hierarchy Model is represented, that is, is divided into disparate modules, obtain tree-shaped The oriented hierarchy chart of structure, including 1 top-level module m₀With multiple submodule m, top-level module m₀Represent whole program execution code, It is a module black box treating deep subdivision, from top-level module m₀Triggering can reach any one submodule m, tree simultaneously Preserved as module information；

Step s2: read module information, if a monitoring point deployment scheme is as initial solution, the mould of stationary monitoring wanted in record Block；

Step s3: construction evaluates object function e (m) of monitoring point, and calculates the target function value of initial solution and preserve:

E (m)=g (m₀)*∑c(m_x)/(1+∑r(m_x)),m_x∈m (1)

Wherein, m_xIt is x-th submodule, c (m_x) it is submodule m_xMonitoring overhead, r (m_x) it is submodule m_xMonitoring superfluous Remaining, g (m₀)^-1=p (m₀), g (m₀) it is top-level module m₀Monitoring granularity, p (m₀) represent is under Current observation strategy, Module m₀It is divided into many small parts, be defined as follows:

$p\left(m_0\right)=\left\{\begin{array}{c}1,d\left(m\right)=0,m\&element;o\left(m_0\right)\\ 1,d\left(m\right)=0,m\&notelement;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),s\left(m\right)=next\left(m\right),m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),s\left(m\right)=0,o\left(m\right)\&notequal;\mathrm{\&phi;},m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),m\&notelement;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),0<s\left(m\right)<next\left(m\right),m\&element;o\left(m_0\right)\end{array}\right.---\left(2\right)$

Wherein, d (m) is the depth of the comprised submodule of module m, o (m₀) it is top-level module m₀In monitored submodule Set, h (m) is the set of next straton module of module m, and s (m), next (m) are respectively in next straton module of representation module m Determine monitored submodule and submodule set of blocks；

Monitoring redundancy r (m) refers to the degree of submodules supplementary notes higher level's modules at different levels, is defined as follows:

$r\left(m\right)=f\left(m\right),\&exists;m^{\&prime;},path\left(m^{\&prime;}\right)\&element;paths\left(m^{\&prime;}\right),m\&element;path\left(m^{\&prime;}\right),m^{\&prime;},m\&element;o\left(m_0\right)---\left(3\right)$

Wherein, path (m') is the node set with regard to module m', and paths (m') represents all roads that slave module m' is set out The set in footpath, it is assumed that there are module m₁, and there is Module nodes set path (m₁)={ m₁,m₂,...,m_d+1, each of which Individual m_k∈h(m_k-1), d is module m₁Depth, k=1,2,3 ..., d+1, slave module m₁The collection in all paths set out is combined into paths(m₁), path (m₁)∈paths(m₁)；Assume there is module m_xAnd m_y, wherein m_x∈h(m_y), f (m_x) it is in module m_xSend out The probability of raw fault, is defined as f (m_x)=f (m_y)/next(m_y), next (m_y) representation module m_yNext level submodule total Number.

Step s4: read module information, initial solution and target function value, it is simulated annealing algorithm, obtain an optimization Monitoring point deployment scheme, concretely comprise the following steps:

401: given initial temperature t₀, primary iteration number of times i₀, initial solution is the current solution of first time iteration, present energy e₀Target function value for initial solution；

402: the module that can refine monitoring to each of current solution calculates the neighborhood that candidate is currently solved, and remembers Record is in buffering；

403: if neighborhood is not empty, execution step 404, if neighborhood is sky, simulated annealing terminates, and output is current Solution；

404: the monitoring overhead of each of neighborhood according to current solution member's scheme, select the minimum conduct of expense new Solution；

405: calculate monitoring redundancy, monitoring granularity and the monitoring overhead of new explanation, and then the target function value of calculating new explanation, The i.e. energy e of new explanation₁；

406: work as e₁＜ e₀Or work as e₁≥e₀, calculate Probability p r [accept]=exp (- (e₁-e₀)/t), pr [accept] ＜ R, r are an equally distributed random number between 0 to 1, then accepting new explanation is current solution, even e₀=e₁, adjust stationary monitoring Module collection, deletes about being refined the related choosing solution of the module of monitoring, current iteration number of times i=i-1 in neighborhood, and executes step Rapid 408；

407: work as e₁≥e₀, pr [accept] >=r, current iteration number of times i=i-1；

408: if i ＞ 0, return to step 402, if i=0, make Current Temperatures t=0.95t, iterationses i=i₀；

409: if t=0, simulated annealing terminates, output is current to be solved, if t ≠ 0, return to step 402.

Finally can be according to the monitoring granularity of prioritization scheme and monitoring redundancy come proof scheme.

As shown in Fig. 2 the contrast in front and back that in modular structure, monitoring point is disposed, the initial solution that left side is disposed for monitoring point, right The last solution that side is disposed for monitoring point.

As shown in figure 3, the simulated annealing that have recorded in example has carried out 10 iteration, along with energy value and temperature Value decline, wherein the 6th time iteration, algorithm logic receives once poor solution with 88% probability, thus open new Excellent solution search path.System under test (SUT) has been divided into 13 parts by final scheme, and observation station generally covers the model at place Node, observation redundancy is 0.30873 simultaneously, maintains relatively low expense.From the point of view of the iterative process of optimized algorithm, some moulds Block is subdivided into its all submodules it is achieved that refinement observation, and have then provides module and part or all submodules simultaneously Information it is achieved that the redundancy of observation information.Consider observation granularity, redundancy and expense from algorithm logic simultaneously comprehensively, with Determine which module to be applied which type of observation program in, wherein expense limits and is placed in first place.

Claims (4)

1. monitoring point optimization dispositions method when a kind of program is run is it is characterised in that include:

Step s1: monitored program code hierarchy Model is represented, that is, is divided into disparate modules, including 1 top layer Module m₀With multiple submodule m, obtain the oriented hierarchy chart of tree, tree is preserved as module information；

Step s2: read module information, if a monitoring point deployment scheme is as initial solution, the module of stationary monitoring wanted in record；

Step s3: construction evaluates object function e (m) of monitoring point, and calculates the target function value of initial solution and preserve:

E (m)=g (m₀)*σc(m_x)/(1+∑r(m_x)),m_x∈m (1)

Wherein, m_xIt is x-th submodule in module collection m, c (m_x) it is submodule m_xMonitoring overhead, r (m_x) it is submodule m_x's Monitoring redundancy, g (m₀)^-1=p (m₀), g (m₀) it is top-level module m₀Monitoring granularity, p (m₀) represent is in Current observation plan Slightly descend top-level module m₀It is divided into many small parts；

Step s4: read module information, initial solution and target function value, it is simulated annealing algorithm, obtain the prison of an optimization Measuring point deployment scheme.

2. monitoring point optimization dispositions method when a kind of program according to claim 1 is run is it is characterised in that current The lower top-level module m of observation strategy₀It is divided into many small parts and adopt p (m₀) represent, p (m₀) refer to a program code in monitoring The degree being divided, is defined as follows:

$p\left(m_0\right)=\left\{\begin{array}{c}1,d\left(m\right)=0,m\&element;o\left(m_0\right)\\ 0,d\left(m\right)=0,m\&notelement;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),s\left(m\right)=next\left(m\right),m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),s\left(m\right)=0,o\left(m\right)\&notequal;\mathrm{\&phi;},m\&element;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right),m_x\&element;h\left(m\right),m\&notequal;o\left(m_0\right)\\ \mathrm{\&sigma;}p\left(m_x\right)+1,m_x\&element;h\left(m\right),0<s\left(m\right)<next\left(m\right),m\&element;o\left(m_0\right)\end{array}\right.---\left(2\right)$

Wherein, d (m) is the depth of submodule m, o (m₀) it is top-level module m₀In monitored submodule set, h (m) is son The set of next straton module of module m, s (m), next (m) represent respectively determine in next straton module of submodule m monitored Submodule and submodule set of blocks.

3. monitoring point optimization dispositions method when a kind of program according to claim 2 is run is it is characterised in that described Monitoring redundancy r (m_x) refer to that submodules at different levels remark additionally the degree of higher level's modules, are defined as follows:

$r\left(m_x\right)=f\left(m_x\right),{\&exists;m}^{\&prime;},\mathrm{path}\left(m^{\&prime;}\right)\&element;\mathrm{paths}\left(m^{\&prime;}\right),m_x\&element;\mathrm{path}\left(m^{\&prime;}\right),m^{\&prime;},m_x\&element;o\left(m_0\right)---\left(3\right)$

Wherein, path (m') is the node set with regard to module m', and paths (m') represents all paths that slave module m' is set out Set, it is assumed that there are module m₁, and there is Module nodes set path (m₁)={ m₁,m₂,...,m_d+1, each of which m_k ∈h(m_k-1), d is module m₁Depth, k=1,2,3 ..., d+1, slave module m₁The collection in all paths set out is combined into paths (m₁), path (m₁)∈paths(m₁)；Assume there is module m_xAnd m_y, wherein m_x∈h(m_y), f (m_x) it is in module m_xThere is event The probability of barrier, is defined as f (m_x)=f (m_y)/next(m_y), next (m_y) representation module m_yNext level submodule sum.

4. monitoring point optimization dispositions method when a kind of program according to claim 1 is run is it is characterised in that described In step s4, simulated annealing concretely comprises the following steps:

401: given initial temperature t₀, primary iteration number of times i₀, initial solution is the current solution of first time iteration, present energy e₀For The target function value of initial solution；

402: the module that can refine monitoring to each of current solution calculates the neighborhood that candidate is currently solved, and records In buffering；

403: if neighborhood is not empty, execution step 404, if neighborhood is sky, simulated annealing terminates, and output is current to be solved；

404: the monitoring overhead of each of neighborhood according to current solution member's scheme, select expense minimum as new explanation；

405: calculate monitoring redundancy, monitoring granularity and the monitoring overhead of new explanation, and then the target function value of calculating new explanation, that is, newly The energy e of solution₁；

406: work as e₁＜ e₀Or work as e₁≥e₀, calculate Probability p r [accept]=exp (- (e₁-e₀)/t), pr [accept] ＜ r, r For an equally distributed random number between 0 to 1, then accepting new explanation is current solution, even e₀=e₁, adjust stationary monitoring mould Set of blocks, deletes in neighborhood about being refined the related choosing solution of the module of monitoring, current iteration number of times i=i-1, and execution step 408；

407: work as e₁≥e₀, pr [accept] >=r, current iteration number of times i=i-1；

408: if i ＞ 0, return to step 402, if i=0, make Current Temperatures t=0.95t, iterationses i=i₀；

409: if t=0, simulated annealing terminates, output is current to be solved, if t ≠ 0, return to step 402.