CN105955834A - Method for optimizing system available time through considering system life and soft errors - Google Patents

Abstract

The present invention discloses a method for optimizing system available time through considering system life and soft errors. The method includes the following steps of modeling for a task set Tn and a processor; and in a case in which the task set Tn and a processor task execution frequency set Fn are given, calculating mean time between temporary failures MTTF<T> by using a provided analysis method and calculating mean time between permanent failures MTTF<P> by using a system-level life reliability modeling tool, wherein overall available time of a system depends on a smaller value of the MTTF<T> and the MTTF<P>. The relationship between the [gamma]MTTF<T> and the MTTF<P>, shown in the description, is classified into four types and then corresponding processing is carried out, so the overall available time of the system is maximized. For a case (ii), a partial backup and accelerated heuristic algorithm is adopted. Traditional reliability optimization usually focuses on only one of system life reliability and soft error reliability. Under the premise of fully considering a permanent failure, a temporary failure and a throughput constraint, the present invention provides the method, thereby solving the problem of maximization of the available time of the system.

Description

A kind of method optimizing system uptime weighing lifetime of system and soft error

Technical field

The present invention relates to the knowledge of real-time system, particularly relate to one and consider lifetime of system and soft error reliability two Plant different reliabilities, by the method weighing two kinds of reliability optimization system entirety pot lifes；Specifically a kind of balance Lifetime of system and the method optimizing system uptime of soft error.

Background technology

In 30 years of past, along with the raising of transistor integrated level on chip, the performance of processor becomes the most superior. But the raising of power density causes operating steeply rising and frequently changing of temperature, can accelerate due to electron transfer (EM), time The chip that the factors such as the dielectric breakdown (TDDB) of decision, stress migration (SM), thermal cycle (TC) cause is aging.Along with aging adds Speed, permanent fault is more easy to occur, causes the minimizing of lifetime of system.The reduction of operation voltage makes integrated circuit be more vulnerable to wink Time fault puzzlement, reduce soft error reliability.Expert has carried out in-depth study with regard to these problems both at home and abroad.Amrouch Et al. have studied influencing each other between the aging impact for lifetime of system of one-sided mechanism and each side mechanism.Chantem etc. People proposes task distribution and the dispatching algorithm of a kind of online reliability perception, and this method can slow down the aging speed of processor Degree.Huang et al. develops a kind of software/hardware Restoration Mechanism based on dispatching algorithm and tolerates permanent and instantaneity two side The fault in face.

The one of two kinds of reliabilities is simply paid close attention in existing great majority work, and the optimization for one of which would generally be weak Change alternative reliability.Give overall consideration to two kinds of reliabilities, utilize unified quantification of targets to weigh and seem critically important.For mostly For number system, user only focuses on and maximizes system uptime in raising MTTF (mean free error time), and does not exist The kind (instantaneous or permanent) of meaning fault.The reduction of operation voltage makes integrated circuit be more vulnerable to the puzzlement of transient fault, Reduce the reliability of soft error.Reduce repair, replace whole system cost and maintain service quality, improve lifetime of system and Soft error reliability becomes the most valuable.

Summary of the invention

The purpose of the present invention proposes for system entirety pot life optimization problem, when being optimized, and Quan Miankao Consider reliability of both lifetime of system and soft error reliability, according to the relation of two kinds of reliabilities, use different process sides Formula to realize the optimization of system entirety pot life, a kind of system uptime that optimizes weighing lifetime of system and soft error Method.For situation (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T<<MTTF_P, have employed opening of a kind of incremental backup and acceleration Hairdo algorithm (PRS).The present invention, considering permanent fault, transient fault and throughput constraints on the premise of comprehensively, formulates and conciliates The certainly maximized problem of system uptime.

The object of the present invention is achieved like this:

A kind of method optimizing system uptime weighing lifetime of system and soft error, it is characterised in that the method includes Following steps:

Step one: task-set and processor model are set up；

Step 2: calculate γ MTTF_TAnd MTTF_P, wherein MTTF_TFor averagely without the time of transient fault, MTTF_PFor averagely Without the time of permanent fault, γ isMinima in Liang Zhe is i.e. defined as the pot life that system is overall；

Step 3: according to γ MTTF_TAnd MTTF_PRelation, carry out correspondence process.Situation (i) γ MTTF_T<<MTTF_P, Go to step four, situation (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T<<MTTF_P, go to step five, situation (iii) γ MTTF_T> MTTF_PRather than γ MTTF_T＞ ＞ MTTF_P, go to step six, situation (iv) γ MTTF_T＞ ＞ MTTF_PAnd go to step seven；

Step 4: at (i) γ MTTF_T<<MTTF_PIn the case of, each ancestral task in task-set is backed up, for Ancestral task and backup tasks, processor is all with peak frequency f_maxExecution task, and go to step eight；

Step 5: at (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T＜ ＜ MTTF_PIn the case of, have employed a kind of part standby Part and the heuritic approach (PRS) accelerated select partial task to carry out backup operation or with peak frequency f_maxPerform task operating, And go to step eight；

Step 6: at (iii) γ MTTF_T>MTTF_PRather than γ MTTF_T>>MTTF_PIn the case of, use DVS method, by appointing The scaling fall low operating temperature of business frequency, it is achieved MTTF_PRaising, go to step eight；

Step 7: at (iv) γ MTTF_T>>MTTF_PIn the case of, use a kind of method optimizing reliability of service life to adjust Whole, it is achieved MTTF_PRaising, go to step eight；

Step 8: optimize and terminate.

Described step 1 specifically includes:

Step A1: the foundation of task-set model:

Wherein: n is task-setThe number of middle independent task；

Step A2: the foundation of task execution time model:

t_i=c_i/f_i

Wherein: t_iFor task τ_iIt is f in frequency_iTime the execution time, f_i(f_min≤f_i≤f_max) be processor operation frequency (frequency is with f for rate_maxBeing normalized for standard, frequency values scope is [0,1]), f_minFor processor minimum operation frequency, f_maxFor processor maximum operation frequency, c_iFor task τ_iAt peak frequency f_maxThe lower execution time；

Described step 2 specifically includes:

Step B1: the foundation of failure rate model:

$\mathrm{\&lambda;}\left(f\right)={\mathrm{\&lambda;}}_0{10}^{\frac{d(1-f)}{1-f_{\min }}}$

Wherein: λ₀For at processor maximum operation frequency f_maxTime failure rate, d (> 0) be hardware specific constant, f For processor frequencies, f_minFor processor minimum operation frequency；

Step B2: the foundation of mission reliability model:

$R_i=e^{-\mathrm{\&lambda;}\left(f_i\right)\frac{c_i}{f_i}}$

Wherein: λ (f_i) it is operation frequency f_iTime failure rate, c_iFor task τ_iAt peak frequency f_maxDuring lower execution Between, f_iOperation frequency for processor；

Step B3: the foundation of the mission reliability model after employing backup method:

$R_i^{rep}=1-{(1-e^{-\mathrm{\&lambda;}\left(f_i\right)\frac{c_i}{f_i}})}^2$

Wherein: λ (f_i) it is operation frequency f_iTime failure rate, c_iFor task τ_iAt peak frequency f_maxDuring lower execution Between, f_iOperation frequency for processor；

Step B4: task-setThe foundation of failure rate model:

Wherein: R_iFor task τ_iIt is f performing frequency_iTime reliability；

Step B5: averagely without the time MTTF of transient fault_TThe foundation of model:

Wherein:For the execution time of task lump, The first round executory expected time in task-set is there is for first-time fault；

Step B6: system reliability model is set up:

$\{\frac{{\mathrm{MTTF}}_T}{{\mathrm{MTTF}}_T+{\mathrm{MTTR}}_T},\frac{{\mathrm{MTTF}}_P}{{\mathrm{MTTF}}_P+{\mathrm{MTTR}}_P}\}$

Wherein: MTTF_TFor averagely without the time of transient fault, MTTF_PFor averagely without the time of permanent fault, MTTR_TFor Average transient fault repair time, MTTR_PFor average permanent fault repair time；

Step B7: optimization of system reliability:

max:{γMTTF_T,MTTR_P}

Wherein: MTTF_TFor averagely without the time of transient fault, MTTF_PFor averagely without the time of permanent fault, γ is

Described step 4 specifically includes:

Step C1: at (i) γ MTTF_T<<MTTF_PIn the case of, system uptime optimization:

To task-set τ_nIn each ancestral task carry out backup operation, for ancestral task and backup tasks, processor All with peak frequency f_maxExecution task

Wherein: f_maxFor processor peak frequency；

Described step 5 specifically includes:

Step D1: at (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T<<MTTF_PIn the case of, system uptime optimization:

The algorithm (PRS) using incremental backup and acceleration selects partial task carry out backup operation or accelerate (with maximum frequency Rate f_maxPerform task operating)；

Wherein: f_maxFor processor peak frequency；

Step D2: the backup of individual task or the foundation of acceleration mark:

Step D3: backup operation soft error reliability increment and the difference model accelerating operation soft error reliability increment are built Vertical:

${\mathrm{\&Delta;R}}_i^{r-s}={\mathrm{\&Delta;R}}_i^r-{\mathrm{\&Delta;R}}_i^s=2R_i-R_i^2-R_{i|f_i=f_{max}}$

Wherein: For task τ after employing backup_iReliability, R_iFor without task τ during backup_i Reliability,It is f for operation frequency_maxTime task τ_iReliability；

Step D4:RS state determines algorithm:

1)

2)

3)

4)

5)

6)

7)

8)

9)return RS_i

Wherein:For task τ after employing backup_iReliability, R_iFor without task τ during backup_i Reliability,It is f for operation frequency_maxTime task τ_iReliability, f^*For working as Time corresponding processor frequencies, work as RS_iBackup operation is carried out when=1；Work as RS_iOperation it is accelerated, to optimize soft error when=0 Reliability.

Step D5:PRS heuritic approach:

Wherein:For the task-set of input, mission frequency set F_n={ f₁,f₂,…f_i,…f_n, f_minMinimum for processor Operation frequency, f_maxFor processor maximum operation frequency, γ is MTTF_PAnd MTTR_TRatio,For pending backup or add The task-set of speed operation,For the most not backing up the system current task collection the most not accelerating operation, first system preserves one IndividualBackup, update after decision when every step is finishedThen iteration comparesWith's Value, whenTime, first calculate each task τ to be operated_iThe MTTF increased_T, determine that it is standby Part or acceleration operation, the MTTF of system_TMaximal increment is the most found, updates currentWhen Condition when being unsatisfactory for, algorithm returns maximum system uptime；

Described step 6 specifically includes:

Step E1: at (iii) γ MTTF_T>MTTF_PRather than γ MTTF_T>>MTTF_PIn the case of, system uptime optimization:

Use DVS method, by mission frequency f_iScaling fall low operating temperature

Wherein: f_i(f_min≤f_i≤f_max) it is the operation frequency of processor；

Described step 7 specifically includes:

Step F1: at (iv) γ MTTF_T＞ ＞ MTTF_PIn the case of, system uptime optimization:

A kind of method optimizing reliability of service life is utilized to be adjusted.

The present invention is by calculating the average time MTTF without transient fault_TWork is modeled with utilizing system level reliability of service life Tool calculates the average time MTTF without permanent fault_P, according to γ MTTF_TAnd MTTF_PRelation, be divided into 4 kinds of situations to carry out correspondence Process, the pot life overall to maximize system, whereinIt is set to definite value.For situation (ii), have employed one Plant incremental backup and the heuritic approach (PRS) of acceleration.Traditional reliability optimization often only focuses on lifetime of system and soft error The one of reliability, the optimization for one would generally weaken another kind.The present invention is considering permanent fault, transient fault comprehensively On the premise of throughput constraints, formulate and solve the maximized problem of system uptime.

Accompanying drawing explanation

Fig. 1 is flow chart of the present invention；

Fig. 2 is γ MTTF in 15 different benchmark task collection tests_TAnd MTTF_PResult figure；

Fig. 3 is under 4 kinds of different conditionsThe explanatory diagram of value；

Fig. 4 is the PRS algorithm in the present invention and the contrast signal in synthesis benchmark task collection test of other 4 kinds of algorithms Figure；

Fig. 5 is the PRS algorithm in the present invention and the contrast signal in true benchmark task collection is tested of other 4 kinds of algorithms Figure.

Detailed description of the invention

Below in conjunction with drawings and the specific embodiments, the present invention is described in further detail.

Task-set in the present inventionIt is made up of n independent task, given meeting In the case of handling capacity, task-set continuously performs.Dynamic voltage/frequency scaling is used to perform task-set, frequency sets F can be expressed as_i(f_min≤f_i≤f_max), f_minFor processor minimum operation frequency, f_maxFor processor maximum operation frequency, appoint Business is with frequency f_iTime required during execution is t_i=c_i/f_i, c_iFor task τ_iAt peak frequency f_maxThe lower execution time.

The present invention uses λ₀Represent at processor maximum operation frequency f_maxTime failure rate, mean failure rate is about λ₀ Exponential function, be represented byWith frequency f_iDuring execution, mission reliability can be expressed asUsing backup tolerance transient fault in the present invention, the mission reliability after backing up can be expressed as Task-setFailure rate model is i.e. represented byThis In brightFor the execution time of task lump,For event first There is the first round executory expected time in task-set in barrier.UtilizeWithThree parameters, MTTF_T(averagely without the time of transient fault) is i.e. expressed as

The present invention is meeting given throughput constraints condition, and considers lifetime of system and soft error reliability failure simultaneously In the case of, to formulate a single-object problem realization of goal system uptime and maximized, system uptime optimizes mould Type is established asMTTF_TFor averagely without the time of transient fault, MTTF_PFor average without permanent The time of property fault, MTTR_TThe time repaired for average transient fault, MTTR_PThe time repaired for average permanent fault.WillIt is set to γ, γ and can be seen as constant, then consider under lifetime of system and soft error reliability situation comprehensively, system entirety Pot life is decided by γ MTTF_TAnd MTTR_PIn smaller value, for making system uptime maximize, problem can be converted into max:{γMTTF_T,MTTR_P}。

As it is shown in figure 1, in the present invention, according to γ MTTF_TAnd MTTR_PRelation, be divided into 4 kinds of situations to carry out classification process.? Before classification processes, first verify that 4 kinds of situations are implicitly present in.Consider different benchmark task collection test program, core status, Under the factor such as temperature curve and frequency, the present invention has carried out some simulation experiments and has compared γ MTTF_TAnd MTTR_PValue.Given identical Processor, after change benchmark task collection test program and parameter are set for experiment, result is as in figure 2 it is shown, four kinds of situations are certain Exist.

For situation (i) γ MTTF_T<<MTTF_P, task-setIn each ancestral task carry out backup operation, for former Beginning task and backup tasks, processor is all with peak frequency f_maxExecution task；For situation (iii) γ MTTF_T>MTTF_PRather than γMTTF_T>>MTTF_P, use DVS method, by mission frequency f_iScaling fall low operating temperature, to realize system uptime Maximize；For situation (iv) γ MTTF_T>>MTTF_P, use the strategy of reliability of service life perception to be adjusted, it is achieved system can With the maximization of time.The solution of above 3 kinds of situations has had a variety of, therefore, only makes a brief description in the present invention.Right In situation (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T<<MTTF_P, have employed the heuritic approach of a kind of incremental backup and acceleration (PRS)。

In the heuritic approach of incremental backup and acceleration, first basis InValue determineWithBetween relation, wherein backup operation soft error reliability increment is For task τ after employing backup_iReliability, R_iFor without task τ during backup_iReliability；Accelerate Operation soft error reliability incrementIt is f for operation frequency_maxTime task τ_iReliably Property, R_iFor frequency f_iTime mission reliability.Fig. 3 is under 4 kinds of different conditionsThe explanation of value.

Determine that algorithm determines RS according to RS state in step E4_iValue, RS_iWhen being 1, task τ_iCarry out backup operation；RS_i When being 0, task τ_iIt is accelerated operation.Temporary duty collectionFor treating the task-set of decision-making (back up or accelerate), first preserve OneCopy, be updated after each decision-making.System current task collectionIt is initialised toIncremental backup and The workload iteration that system is current is compared γ MTTF by the heuritic approach (PRS) accelerated_TAnd MTTF_PValue.When system meetsTime, algorithm first calculates each task to be operatedThe MTTF increased_T, certainly Its operation (back up or accelerate) fixed, the MTTF of system_TMaximal increment is the most found.After carrying out the operation of complete part or acceleration, Update currentBy τ_iFromIn remove.Finally, whenCondition be discontented with During foot, algorithm returns maximum system uptime.

In the present invention, after 4 kinds of different situations are carried out the operation of correspondence, consider lifetime of system and soft error failure comprehensively In the case of, it is achieved system uptime is maximum, optimizes complete.

In order to illustrate that the heuritic approach (PRS) of incremental backup and the acceleration proposed in the present invention is to system uptime Optimizing, experimental section will be carried out according to following steps:

1. in synthesis task-set benchmark test, the PRS algorithm in the present invention and the contrast situation of other 4 kinds of algorithms.

2. in true benchmark task collection is tested, the PRS algorithm in the present invention and the contrast situation of other 4 kinds of algorithms.

Embodiment

Step 1: utilize Alpha 21264 microprocessor to be respectively as follows: as hardware platform, processor frequencies/voltage level 1.0GHz/0.7V, 1.25GHz/0.8V, 1.5GHz/0.9V, 1.75GHz/1.0V and 2.0GHz/1.1V, utilize HotSpot5.0 Obtain temperature curve, initial temperature and ambient temperature and be set to 330 °F and 318.5 °F.

Step 2: utilize system-level modeling for life instrument to obtain MTTF_P, for estimating MTTF_T, parameter is set to λ₀=10^-7、 D=3, in simulation experiment, arranges γ=1.

Step 3:5 group verifies PRS algorithm and other 4 kinds of algorithms based on the benchmark test that synthesis task-set is the most integrated, Each task-set comprises 20 tasks, and experimental result is as shown in Figure 4.

4 kinds of comparison algorithms are respectively as follows: random algorithm (RA), and task backs up at random or accelerates；Accelerating algorithm completely (FSA), each task performs with peak frequency；Algorithm (FRA) completely, each task has a backup operation；Energy Amount is saved and reliability perception algorithm (ERA), and each task performs in the frequency of an energy efficient, and each task is altogether Enjoy a recovery tasks to guarantee system reliability.

PRS algorithm and other 4 kinds of algorithms, application are verified in benchmark task collection based on the true application test of step 4:4 group It is respectively automobile industry (automotive-industrial), consumption network (consumer-networking), telecommunications And mpeg standard (telecom), each task-set comprises 16,20,17,15 tasks respectively, experimental result such as Fig. 5 institute Show.4 kinds of comparison algorithms are ibid.

System entirety pot life is decided by γ MTTF_TAnd MTTF_PIn smaller value.The result of Fig. 4 and Fig. 5 shows invention In PRS algorithm be better than RA algorithm, FSA algorithm, FRA algorithm and ERA algorithm, average raising ratio is respectively 36.6%, 19.8%, 24.3% and 11.5%.Further improving PRS algorithm, relative 4 kinds of comparison algorithms, the raising of performance is respectively 85%, 33.5%, 51% and 45.1% can be reached.

The present invention proposes a new analytic method and calculates the MTTF determined due to soft error, and elaborates one Balance lifetime of system and the method optimizing system uptime of soft error.For situation (ii), have employed a kind of incremental backup With the heuritic approach (PRS) accelerated.Experiment shows, the algorithm representative relative to 4 kinds, our heuritic approach (PRS) there is well optimization for system uptime.

Claims (7)

1. the method optimizing system uptime weighing lifetime of system and soft error, it is characterised in that the method include with Lower step:

Step 1: task-set and processor model are set up；

Step 2: calculate γ MTTF_TAnd MTTF_P, wherein MTTF_TFor averagely without the time of transient fault, MTTF_PFor average without permanent The time of property fault, γ isMinima in Liang Zhe is i.e. defined as the pot life that system is overall；

Step 3: according to γ MTTF_TAnd MTTF_PRelation, point 4 kinds of situations process；Situation (i) γ MTTF_T＜ ＜ MTTF_P, turn Step 4；Situation (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T＜ ＜ MTTF_P, go to step 5；Situation (iii) γ MTTF_T>MTTF_P Rather than γ MTTF_T＞ ＞ MTTF_P, go to step 6；Situation (iv) γ MTTF_T＞ ＞ MTTF_P, go to step 7；

Step 4: at (i) γ MTTF_T＜ ＜ MTTF_PIn the case of, each ancestral task in task-set is carried out backup operation, right In ancestral task and backup tasks, processor is all with peak frequency f_maxExecution task, and go to step 8；

Step 5: at (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T＜ ＜ MTTF_PIn the case of, have employed a kind of incremental backup and add The heuritic approach (PRS) of speed selects partial task to carry out backup operation or with peak frequency f_maxPerform task operating, and turn step Rapid 8；

Step 6: at (iii) γ MTTF_T>MTTF_PRather than γ MTTF_T＞ ＞ MTTF_PIn the case of, use DVS method, by task frequency The scaling fall low operating temperature of rate, it is achieved MTTF_PRaising, go to step 8；

Step 7: at (iv) γ MTTF_T＞ ＞ MTTF_PIn the case of, use a kind of method optimizing reliability of service life to be adjusted, real Existing MTTF_PRaising, go to step 8；

Step 8: optimize and terminate.

2. optimization system uptime method as claimed in claim 1, it is characterised in that described step 1 specifically includes:

Step A1: the foundation of task-set model:

Wherein: n is task-setThe number of middle independent task；

Step A2: the foundation of task execution time model:

t_i=c_i/f_i

Wherein: t_iFor task τ_iIt is f in frequency_iTime the execution time, f_i(f_min≤f_i≤f_max) it is the operation frequency of processor, frequently Rate is with f_maxBeing normalized for standard, frequency values scope is [0,1], f_minFor processor minimum operation frequency, f_maxFor place Reason device maximum operation frequency, c_iFor task τ_iAt peak frequency f_maxThe lower execution time.

3. optimization system uptime method as claimed in claim 1, it is characterised in that described step 2 specifically includes:

Step B1: the foundation of failure rate model:

$\mathrm{\&lambda;}\left(f\right)={\mathrm{\&lambda;}}_0{10}^{\frac{d(1-f)}{1-f_{\min }}}$

Wherein: λ₀For at processor maximum operation frequency f_maxTime failure rate, d (> 0) be hardware specific constant, f for process Device frequency, f_minFor processor minimum operation frequency；

Step B2: the foundation of mission reliability model:

$R_i=e^{-\mathrm{\&lambda;}\left(f_i\right)\frac{c_i}{f_i}}$

Wherein: λ (f_i) it is operation frequency f_iTime failure rate, c_iFor task τ_iAt peak frequency f_maxThe lower execution time, f_iFor The operation frequency of processor；

Step B3: the foundation of the mission reliability model after employing backup method:

$R_i^{rep}=1-{(1-e^{-\mathrm{\&lambda;}\left(f_i\right)\frac{c_i}{f_i}})}^2$

Wherein: λ (f_i) it is operation frequency f_iTime failure rate, c_iFor task τ_iAt peak frequency f_maxThe lower execution time, f_iFor The operation frequency of processor；

Step B4: task-setThe foundation of failure rate model:

Wherein: R_iFor task τ_iIt is f performing frequency_iTime reliability；

Step B5: averagely without the time MTTF of transient fault_TThe foundation of model:

Wherein:For the execution time of task lump,Headed by There is the first round executory expected time in task-set in secondary fault；

Step B6: system entirety pot life model is set up:

$\{\frac{{\mathrm{MTTF}}_T}{{\mathrm{MTTF}}_T+{\mathrm{MTTR}}_T},\frac{{\mathrm{MTTF}}_P}{{\mathrm{MTTF}}_P+{\mathrm{MTTR}}_P}\}$

Wherein: MTTF_TFor averagely without the time of transient fault, MTTF_PFor averagely without the time of permanent fault, MTTR_TFor averagely Transient fault repair time, MTTR_PFor average permanent fault repair time；

Step B7: system uptime optimization:

max:{γMTTF_T,MTTR_P}

Wherein: MTTF_TFor averagely without the time of transient fault, MTTF_PFor averagely without the time of permanent fault, γ is

4. optimization system uptime method as claimed in claim 1, it is characterised in that described step 4 specifically includes:

Step C1: at (i) γ MTTF_T＜ ＜ MTTF_PIn the case of, system uptime optimization:

To task-setIn each ancestral task carry out backup operation, for ancestral task and backup tasks, processor is all with Big frequency f_maxExecution task

Wherein: f_maxFor processor peak frequency.

5. optimization system uptime method as claimed in claim 1, it is characterised in that described step 5 specifically includes:

Step D1: at (ii) γ MTTF_T<MTTF_PRather than γ MTTF_T＜ ＜ MTTF_PIn the case of, system uptime optimization:

The algorithm (PRS) using incremental backup and acceleration selects partial task carry out backup operation or accelerate (with peak frequency f_max Perform task operating)；

Wherein: f_maxFor processor peak frequency；

Step D2: the backup of individual task or the foundation of acceleration mark:

Step D3: backup operation soft error reliability increment and the difference model accelerating operation soft error reliability increment are set up:

${\mathrm{\&Delta;R}}_i^{r-s}={\mathrm{\&Delta;R}}_i^r-{\mathrm{\&Delta;R}}_i^s=2R_i-R_i^2-R_{i|f_i=f_{max}}$

Wherein: For task τ after employing backup_iReliability, R_iFor without task τ during backup_iCan By property,It is f for operation frequency_maxTime task τ_iReliability；

Step D4:RS state determines algorithm:

$1)---{\mathrm{if\&Delta;R}}_i^{r-s}\left(f_{min}\right)>0$

$2)---then{\mathrm{RS}}_i=1;//{\mathrm{\&Delta;R}}_i^{r-s}\left(f_i\right)>{\mathrm{\&Delta;R}}_i^{r-s}\left(f_{min}\right)\&DoubleRightArrow;{\mathrm{\&Delta;R}}_i^r>{\mathrm{\&Delta;R}}_i^s$

$3)---{\mathrm{if\&Delta;R}}_i^{r-s}\left(f_{max}\right)<0$

$4)---then{\mathrm{RS}}_i=0;//{\mathrm{\&Delta;R}}_i^{r-s}\left(f_i\right)<{\mathrm{\&Delta;R}}_i^{r-s}\left(f_{max}\right)\&DoubleRightArrow;{\mathrm{\&Delta;R}}_i^r<{\mathrm{\&Delta;R}}_i^s$

$6)---then{\mathrm{RS}}_i=1;//{\mathrm{\&Delta;R}}_i^{r-s}\left(f_i\right)>{\mathrm{\&Delta;R}}_i^{r-s}\left(f^{*}\right)\&DoubleRightArrow;{\mathrm{\&Delta;R}}_i^r>{\mathrm{\&Delta;R}}_i^s$

$8)---then{\mathrm{RS}}_i=0;//{\mathrm{\&Delta;R}}_i^{r-s}\left(f_i\right)<{\mathrm{\&Delta;R}}_i^{r-s}\left(f^{*}\right)\&DoubleRightArrow;{\mathrm{\&Delta;R}}_i^r<{\mathrm{\&Delta;R}}_i^s$

9)return RS_i

Wherein: For task τ after employing backup_iReliability, R_iFor without task τ during backup_iCan By property,It is f for operation frequency_maxTime task τ_iReliability, f^*For working asTime Corresponding processor frequencies, works as RS_iBackup operation is carried out when=1；Work as RS_iBe accelerated when=0 operation, with optimize soft error can By property；Step D5:PRS heuritic approach:

Wherein:For the task-set of input, mission frequency set F_n={ f₁,f₂,…f_i,…f_n, f_minFor processor minimum operation Frequency, f_maxFor processor maximum operation frequency, γ is MTTF_PAnd MTTR_TRatio,For pending backup or acceleration behaviour The task-set made,For the most not backing up the system current task collection the most not accelerating operation, first system preserves one Backup, update after decision when every step is finishedThen iteration comparesWithValue, whenTime, first calculate each task to be operatedThe MTTF increased_T, determine standby Part or acceleration operation, the MTTF of system_TMaximal increment is the most found；After carrying out the operation of complete part or acceleration, update current 'sBy τ_iFromIn remove；Finally, whenCondition when being unsatisfactory for, calculate Method returns maximum system uptime.

6. optimization system uptime method as claimed in claim 1, it is characterised in that described step 6 specifically includes:

Step E1: at (iii) γ MTTF_T>MTTF_PRather than γ MTTF_T＞ ＞ MTTF_PIn the case of, system uptime optimization:

Use DVS method, by mission frequency f_iScaling fall low operating temperature

Wherein: f_i(f_min≤f_i≤f_max) it is the operation frequency of processor.

7. optimization system uptime method as claimed in claim 1, it is characterised in that described step 7 specifically includes:

Step F1: at (iv) γ MTTF_T＞ ＞ MTTF_PIn the case of, system uptime optimization:

A kind of method optimizing reliability of service life is utilized to be adjusted.