On the Scheduling of Mixed-Criticality Real-Time Task Sets

Size: px
Start display at page:

Download "On the Scheduling of Mixed-Criticality Real-Time Task Sets"

Transcription

1 On the Schedulng of Mxed-Crtcalty Real-Tme Task Sets Donso de Nz, Karthk Lakshmanan, and Ragunathan (Raj) Rajkumar Carnege Mellon Unversty, Pttsburgh, PA Abstract The functonal consoldaton nduced by the costreducton trends n embedded systems can force tasks of dfferent crtcalty (e.g. ABS Brakes wth DVD) to share a processor and nterfere wth each other. These systems are known as mxedcrtcalty systems. Whle tradtonal temporal solaton technques prevent all nter-task nterference, they waste utlzaton because they need to reserve for the absolute worst-case executon tme (WCET) for all tasks. In many mxed-crtcalty systems the WCET s not only rare, but at tmes dffcult to calculate, such as the tme to localze all possble objects n an obstacle avodance algorthm. In ths stuaton t s more approprate to allow the executon tme to grow by stealng cycles from lower-crtcalty tasks. Even more crucal s the fact that temporal solaton technques can stop a hgh-crtcalty task (that was overrunnng ts nommal WCET) to allow a low-crtcalty task to run, makng the former mss ts deadlne. We dentfy ths as the crtcalty nverson problem. In ths paper, we characterze the crtcalty nverson problem and present a new schedulng scheme called zero-slack schedulng that mplements an alternatve protecton scheme we refer to as asymmetrc protecton. Ths protecton only prevents nterference from lower-crtcalty to hgher-crtcalty tasks and mproves the schedulable utlzaton. We use an offlne algorthm wth two parts: a zero-slack calculaton algorthm, and a slack analyss algorthm. The zero-slack calculaton algorthm mnmzes the utlzaton needed by a task set by reducng the tme low-crtcalty tasks are preempted by hgh-crtcalty ones. Ths algorthm can be used wth prorty-based preemptve schedulers (e.g. RMS, EDF). The slack analyss algorthm s specfc for each prorty-based preemptve scheduler and we develop and evaluated the one for RMS. We prove that ths algorthm provdes the same level of protecton aganst crtcalty nverson as the best known prorty assgnment for ths purpose, crtcalty as prorty asgnment (CAPA). We also prove that zero-slack RM provdes the same level of schedulable utlzaton as RMS when all tasks have equal crtcalty levels. Fnally, we present our mplementaton of the runtme enforcement mechansms n Lnux/RK to demonstrate ts practcalty. I. INTRODUCTION In modern real-tme embedded systems, such as avoncs and automotve systems, there s ncreasng pressure to reduce cost and physcal resources such as power and heat. Ths has resulted n the consoldaton of functonalty of dfferent crtcalty nto shared hardware resources (e.g. processor and memory). Unfortunately, such sharng can lead to nterference across tasks (e.g. one task usng the processor longer than expected) of dfferent crtcalty potentally causng crtcal falures. For nstance, f we deploy an ABS brakng system task on the same processor as a navgaton system task of the car, f the latter does not release the processor on tme t could make the former mss ts deadlne. Ths problem s a concern of growng nterest. In partcular, the US Ar Force Research Laboratory has been leadng the Mxed-Crtcalty Archtecture Requrements (MCAR) [1] ntatve to nvestgate buldng blocks to safely construct these mxed-crtcalty systems. Resource parttonng s the key mechansm used to prevent nterferences due to shared resources. Ths parttonng ncurs ts own costs. In partcular, we need to provson for the worst-case resource demand (e.g. worst-case executon tme WCET) even though such a demand s only exercsed n rare occasons. Obtanng ths WCET s a challengng task due to hardware unpredctablty (e.g. caches, ppelnes, speculatve executon) and the dependency of the executon tme of some algorthms on the envronment (e.g. number of obstacles to avod). As a result, WCET numbers are often obtaned through measurements and supplemented by an added cushon factor as an educated and typcally conservatve estmate. Ths leads to a twofold problem: poor average utlzaton of the processor and occasonal enforcement (temporary stoppng) of tasks that need to run longer than ther WCET. Such enforcement prevents low-crtcalty tasks from nterferng wth hghercrtcalty ones and can make the low-crtcalty task mss ts deadlne whle preventng the mssng of the deadlne of the hgher-crtcalty task. Unfortunately, such enforcement can also be suffered by the hgher-crtcalty task to prevent ts nterference on a low-crtcalty task. As a result, the enforcement can make the hgher-crtcalty task mss ts deadlne to allow the low-crtcalty one to meet ts own. We call ths type of enforcement symmetrc enforcement because t acts the same way n both drectons: low-to-hgh and hgh-to-low crtcaltes. Ths enforcement can have the opposte effect to our orgnal ntent: a lower-crtcalty task s favored over a hgher-crtcalty tasks and crtcalty nverson s sad to occur. One way to elmnate the crtcalty nverson problem s to smply assgn the prortes of the tasks n the order of ther crtcalty levels. We call ths scheme Crtcalty As Prorty Assgnment (CAPA). However, ths scheme can lead to very poor utlzaton f the resultng order turns out to be contrary to the best prorty assgnment to maxmze utlzaton (e.g. rate-monotonc prorty, proven to be optmal for fxed-prorty schedulng). Such an approach can lead to very low levels of schedulable utlzaton due to forced prorty nverson. In ths paper, we descrbe a schedulng polcy called zeroslack schedulng to provde a new form of protecton we call asymmetrc protecton. Ths new form of protecton elmnates the mpact on deadlne due to the crtcalty nverson whle mnmzng the penalty on schedulable utlzaton. We also

2 develop new metrcs to evaluate crtcalty nverson. The rest of ths paper s organzed as follows. Secton II summarzes the related work n ths area. Secton III ntroduces the crtcalty nverson problem, and develops the blockng terms requred to characterze t. Secton IV develops the zeroslack schedulng scheme to deal wth crtcalty nverson and presents performance metrcs for mxed-crtcalty schedulng. Secton V apples the zero-slack scheme to rate-monotonc (RM) schedulng, and proves useful propertes about the ntegrated zero-slack RM scheduler. Secton VI detals our mplementaton of zero-slack RM schedulng on Lnux/RK. Fnally, we provde our concludng remarks n secton VII. II. RELATED WORK Multple papers have been publshed related to overload schedulng. [2] and [3] use a form of crtcalty together wth a value assgned to task completons. Ther approach s then to maxmze the accrued value. In our case, we do not requre a value for task completon (whch can be dffcult to obtan) but use an absolute rankng that matches the common semantcs of mxed-crtcalty task sets. Addtonal studes have also been done on onlne schedulng of overloads [4 6]. These schemes do not nclude an explct noton of crtcalty and hence cannot take advantage of ths noton. Ch-Sheng et al. present n [7] an approach to map the semantc mportance of tasks n a task set nto QoS servce classes to mprove the resource utlzaton. Ther objectve s to ensure that the allocated resources are never less for a hghcrtcalty task than for a lower-crtcalty one. In our case, we focus on deadlnes, and guarantee that whenever an overload reduces the resources avalable we ensure that a hgh-crtcalty task wll not mss ts deadlne due to the allocaton of more resources to a lower-crtcalty task. Varous operatng systems lke Lnux/RK [8], QLnux [9], Wndows Vsta [10], and Ralto [11], provde extensve mechansms for resource solaton through advance reservatons. Although such exstng technques are fundamental to provdng hard tmng guarantees, they are stll agnostc to crtcalty and enforce solaton symmetrcally. In contrast, our scheme uses tasks crtcaltes and overload budgets to create an asymmetrc enforcement wthout ncurrng crtcalty nverson. In the specfc context of overload schedulng, schedulng algorthms based on the elastc task model have been proposed [12]. Tasks wth hgher elastcty are allowed to run at hgher rates when requred, whereas tasks wth lesser elastcty are restrcted to a more steady rate. Ths elastcty does not explctly capture the crtcalty of a task, and t requres all remanng tasks to be compressed under an overload scenaro. Our soluton explctly supports overload scenaros and crtcalty levels through a crtcal mode (C) of operaton n addton to the normal mode (N) of operaton. Durng ths crtcal mode, we block the lower-crtcalty tasks, thereby transferrng utlzaton only from the lower-crtcalty tasks under overloaded condtons to beneft hgher-crtcalty tasks. Closely related to our proposal for dual-executon modes vz. normal mode (N) and crtcal mode (C) s the work on dual-prorty schedulng [13]. Dual-prorty schedulng s an effectve technque for responsvely schedulng soft tasks wthout compromsng the deadlnes of crucal hard real-tme tasks. Whle ther noton of crucal tasks s somewhat smlar to crtcalty there are two stark dfferences. Frst, non-crucal tasks are not gven any schedulng guarantees n ther work, whereas, n our scheme all crtcalty levels have a guarantee on ther executon tmes. Secondly, ther lmtaton to two levels smplfes the analyss but also lmts the utlty. In contrast, we allow an unlmted number of crtcalty levels wth an mplct graceful degradaton on overload that follows the crtcalty order. Another related work to dual-prorty schedulng s the earlest deadlne zero laxty (EDZL) algorthm [14, 15]. In EDZL, tasks are scheduled based on earlest deadlne frst (EDF) untl some task reaches zero laxty (or zero slack), n whch case ts prorty s elevated to the hghest level. Whle the noton of zero slack s used n our soluton, the exstng EDZL results do not consder the noton of task crtcaltes and are not drectly applcable to the mxed-crtcalty schedulng problem. Perod transformaton has been an mportant soluton studed n the context of mxed-crtcalty systems [16, 17]. Ths technque has been subsequently extended and generalzed n [18]. For any gven task set, we can transform the tasks by scalng both ther computaton tme and perod equally. In the context of mxed crtcalty, we can perform perod transformaton to ensure that the task prortes match the requred crtcalty. The problem n usng ths soluton for dealng wth overload scenaros s that the perod transformaton needs to be performed assumng the task wll overrun ts absolute WCET,.e., wth an overloaded computaton-tme. Performng such a perod transform usng the overloaded computaton tme results n early and pessmstc preemptons by the perod-transformed hgher-crtcalty tasks over the lowercrtcalty tasks. The proposed zero-slack schedulng algorthm swtches tasks to ther crtcal mode only when there would be zero slack remanng to meet ther correspondng overloaded computaton tme requrements. Therefore, the zeroslack schedulng algorthm does not mpose more nterference than necessary to provde schedulng guarantees n mxedcrtcalty systems. III. THE CRITICALITY INVERSION PROBLEM Prorty-based preemptve schedulng polces assgn prortes to tasks, and at run-tme, attempt to schedule the hghest-prorty ready task. The prorty assgnment can be fxed across task nstances (fxed-prorty schedulers) or t can change across task nstances (dynamc prorty schedulers). In tradtonal real-tme schedulng, prortes are assgned wth the purpose of maxmzng schedulable utlzaton whle respectng the deadlnes of all the tasks n the set. The utlzaton maxmzaton approach of tradtonal realtme schedulers makes two mportant assumptons: (1) all tasks are equally mportant, and (2) the utlzaton never goes beyond the allowable threshold(s). These two assumptons seldom hold n mxed-crtcalty systems. In partcular, tasks

3 can have dfferent crtcalty levels. Hence, f there s ever a stuaton where we can only satsfy the deadlne of one task, then we should choose to meet the one wth hgher crtcalty. The second assumpton does not hold for a more subtle reason. Specfcally, the fact that enforcement s even consdered as a protecton mechansm mples that there s a possblty that some tasks can go beyond ther specfed worstcase executon tme. Whle ths s, n general, consdered a fault, an explct protecton aganst t s needed because t can create (temporal) overload stuatons. These overload stuatons are precsely when crtcalty nverson can occur. Snce the tradtonal prorty assgnment made by the scheduler was talored to ncrease schedulable utlzaton, t s agnostc to crtcalty. In partcular, under prorty-based preemptve schedulng, a low crtcalty task can have a hgher schedulng prorty than a hgher-crtcalty task. Such prorty assgnment would schedule the low crtcalty task earler than the hghercrtcalty task, potentally makng the latter mss ts deadlne. On the other hand, f we assgn prortes based on crtcalty (as ntroduced n Secton I), then we elmnate crtcalty nverson. However, ths assgnment can potentally create sgnfcant prorty nverson [19] from the perspectve of prorty-based preemptve schedulers. To consder both schedulng prortes and task crtcaltes, and explctly capture the overload executon requrements, let us defne a task τ as: where: τ = (C, C o, T, D, ζ ) C s ts worst-case executon tme under non-overloaded condtons, C o s the overload executon budget, T s the perod of the task, D s the deadlne of the task (wth D T ), ζ s the crtcalty of the task. We follow the same conventon as wth prortes: lower ts value, hgher the crtcalty. It s worth notng that the overload budget C o s used to create a precse defnton our guarantee,.e., how much overload s guaranteed. As we wll show later, t s possble to reserve a conservatve (large) C o snce t does not add up across crtcalty levels. The prorty blockng densty P B for task τ s then defned as: where: P B = τ j ζ j<ζ pb j D (1) pb j s the maxmum tme that a hgher-crtcalty task τ j can block the executon of an nstance of τ durng whch the schedulng prorty of τ s hgher than that of τ j. Smlarly, the crtcalty blockng densty CB for task τ s calculated as: CB = τ j ζ <ζ j cb j D (2) where: cb j s the maxmum tme that a lower-crtcalty task τ j can block the executon of an nstance of τ durng whch the schedulng prorty of τ j s hgher than that of τ. P B and CB become blockng utlzaton when D = T quantfyng the exact schedulable utlzaton loss. However, we present our general dscusson on utlzaton loss usng the blockng densty fgures to allow the more general case where D T. Notwthstandng we wll use the blockng utlzaton to evaluate our Zero-Slack Rate-Monotonc algorthm n Secton V-E snce D = T n ths case. Both pb j and cb j vary dependng on how and when the scheduler assgns prortes and the strateges used to tradeoff prorty and crtcalty blockng. In the next secton we dscuss the strateges taken n our new schedulng scheme. IV. ZERO-SLACK SCHEDULING Our schedulng polcy works on top of tradtonal prortybased preemptve real-tme schedulers. It s based on the observaton that crtcalty nverson only matters under overload condtons. We use ths observaton to create two executon zones for each task τ. In the frst zone, every task s ncluded whle n the second zone, every task τ j ζ j > ζ s suspended. Ths suspenson effectvely blocks the nterference of lowercrtcalty tasks n the case of an overload condton, up to the completon of the task actvaton. It must be noted that τ tself can also be suspended by a task τ c ζ c < ζ n τ c s second zone. The executon zones partton the executon of each task nto two modes: the normal mode (N mode) and the crtcal mode (C mode). Our schedulng algorthm then calculates the executon tme for each mode. We now defne our schedulng guarantee. A. Schedulng Guarantee Our scheduler performs admsson control, and f admtted 1, a task τ s guaranteed to run up to C o f no hgher crtcalty tasks exceeds ts C. Ths guarantee follows the separaton of the overloaded from the non-overloaded stuaton. Ths separaton allows us to make two strategc decsons. Frst, when no overload condton s present, we should schedule the task wth the objectve of maxmzng utlzaton. And secondly, we also avod ntroducng any nterference preventon untl the last nstant necessary to satsfy our guarantee. The last nstant to block nterference for a task τ s the latest tme at whch the nterference preventon mechansm can free enough cycles for τ to complete ts C o before ts deadlne. Ths nstant s dentfed as zero-slack nstant because t leaves no slack n C mode after the completon of C o. Ths s the nstant at whch the executon mode of the task swtches from N mode nto C mode. Avodng nterference preventon tll ths last nstant enables us to maxmze the total schedulable utlzaton, whle stll meetng our schedulng guarantee. 1 In the rest of ths paper, when we use the term schedulable, we refer to the property of satsfyng ths schedulng guarantee.

4 Our scheme s dvded nto two parts. The frst part s an admsson test that determnes the zero-slack nstant for each of the tasks n a task set. The second s a runtme enforcement mechansm that prevents nterference based on the crtcalty and the zero-slack nstant of each task. B. Admsson Control Our admsson control scheme takes the form of a slackdscovery algorthm. We start wth the worst-case assumpton that there s no slack n the N mode of the tasks and they always need to run n C mode. Ths s equvalent to CAPA. Then, we start movng computaton tme from the C mode (n C c) to the N mode (n Cn ) of the tasks as we dscover slack n the N mode. Ths scheme s presented n Algorthm 1. The algorthm converges when t cannot move the zeroslack nstant of any task towards ther deadlne. Ths s codfed wth two varables per task Z 0 (zero-slack before cycle) and Z 1 (zero-slack after cycle). Insde the cycle, the algorthm calculates the slack vector n for task τ n N mode (wth a set of all the orgnal tasks - Γ n ) and C mode (wth a set of hgher-crtcalty tasks - Γ c ). The slack vector contans a sequence of slack regons ordered by tme, where each slack regon contans a startng nstant and duraton. Wth these slack vectors, we obtan a new zero-slack nstant (Z 1) wth the functon GetSlackZeroInstant that takes Co and accommodates the most computaton avalable n the N mode slack vector, wth the rest n the C mode slack vector. After the convergence of ths algorthm a taskset Γ s schedulable f τ Γ T otalslack(v c) Co. Algorthm 1 Compute Fnal Zero-Slack Instants (t = D ) 1: Z 1 0 2: repeat 3: Z 0 Z1 4: for all n taskset do 5: V n GetSlackV ector(, Γ n ) 6: V c GetSlackV ector(, Γ c ) 7: Z 1 GetSlackZeroInstant(, V c, V n, t) 8: end for 9: untl Z 0 = Z1 10: return Z 1 The functon GetSlackVector used n Algorthm 1 s specfc to each prorty-based preemptve schedulng algorthm and we wll dscuss t later n the context of rate-monotonc schedulng. However, the functon GetSlackZeroInstant s generc. Ths functon s presented n Algorthm 2. At the begnnng of ths algorthm, we ntalze the C c and C n of the task to C o and 0 respectvely. The cycle n ths algorthm frst fnds the last tme nstant at whch C c unts of slack are avalable n the slack vector V c before t = D (at lne 3). Then, we calculate how much slack s avalable from zero to ths nstant n the vector V n (at lne 5). Ths avalable slack s then transferred from C c to C n (lnes 7 and 8). The next teraton of the cycle then fnds the next nstant for a smaller TABLE I SAMPLE CRITICAL TASK SET Task C C o T D Crtcalty Prorty τ τ C c (due to the transfer n lnes 7 and 8), movng ths nstant closer to t = D. The loop ends when we are not able to transfer any more computaton from C c to Cn. The functon StartOfTralngSlack returns the nstant at whch the requested slack (C o) for task τ starts such that t s avalable before deadlne t = D. The functon SlackUpToInstant, on the other hand, returns the amount of slack avalable up to the specfed nstant. It s to be noted here that for smplcty of presentaton, we assume that the orgn of tme concdes wth the release of task τ. Algorthm 2 GetSlackZeroInstant(,V c, V n, t): Calculate Instant of Slack = 0 before tme t 1: C c Co ; C n 0 2: repeat 3: t 1 StartOfT ralngslack(, C c, V c ) 4: f t 1 0 and t 1 t then 5: k u SlackUpT oinstant(v n, t 1 ) C n 6: k u = max(mn(k u, C c ), 0) 7: C c Cc k u 8: C n C n + k u 9: else 10: k u 0 11: end f 12: untl k u = 0 13: return t 1 C. A Two-Task Example We now llustrate the admsson control algorthm usng a two-task example and deadlne-monotonc schedulng. Consder the task set n Table I. In ths table, we present the parameters of two tasks, wth an addtonal column for ther deadlne-monotonc prorty. Note that the prorty and the crtcalty of the tasks n ths table are the nverse of each other. Wth these parameters we proceed to calculate the C c and C n for each task τ n the set. For our task set n Table I, C2 n and C2 c are obtaned by frst calculatng the slack for τ 2 n both the N mode and the C mode. In N mode, the slack up to the end of ts deadlne (8) s 4 whch s not enough to schedule ts C2 o = 5 (ths s, n fact, the problem when usng pure deadlne-monotonc schedulng). Ths means that we need to swtch ths task to C mode at some pont to compensate. To do ths, we start by executng ts C2 c = 5 tme n C mode. Then, we evaluate the last nstant Z 2 that C2 c tme unts are avalable n C mode. Snce τ 2 has the hghest crtcalty, t s the only task runnng n ts C mode, and hence ths last nstant s ts deadlne (8) C2(5) c = Z 2 = 3. Then, we check how much tme s avalable n N mode up to

5 Fg. 1. τ τ Swtch to C mode Two-Task Example of the Zero-Slack Schedule tme 3, whch n ths case s 1 (.e., 3 C1 o = 1). We then make ths the ntal tme for C2 n and subtract t from C2. c Next, we repeat the operaton to fnd the last nstant to schedule C2 c = 4 unts of tme n C mode (Z 2 = 4). Ths, n turn, ncrements the avalable slack n N mode (up to nstant 4) to 2, movng another unt to C2 n (.e., C2 n = 2 and C2 c = 3). The last ncrement then leads us to the last nstant when C2 c = 3 s avalable n C mode (Z 2 = 5). Ths new nstant does not ncrement the avalable slack n N mode but defnes the fnal zero-slack nstant Z 2 for task τ 2 to swtch to C mode. The fnal schedule can be seen n Fgure 1. In ths fgure, we can see τ 1 runnng frst for ts whole executon tme C1 o = 2. Then, τ 2 runs for 2 unts and then gets preempted by the second arrval of τ 1 at tme 4. Then, τ 1 runs for one more unt. At ths pont we reach the zero-slack nstant Z 2 of τ 2 and suspend τ 1. When there s no overload n the system, τ 2 wll run only for 0.5 unts more and resume τ 1. However, n an overload stuaton, τ 2 wll have enough cycles to complete ts C2 o = 5 before the end of ts deadlne (at 8). It s worth notng that our computaton of the zero-slack nstant n ths example starts by assumng that there s no slack n the N mode and, as t dscovers more slack, keeps movng computaton to N mode. The key property of ths technque s that the computaton executed n C mode decreases monotoncally. A second mportant observaton s that the ntal executon dstrbuton (between C n and C c) may not be schedulable. In ths example, the ntal startng pont s, n fact, not schedulable,.e., f τ 2 runs always n C mode and only uses up to C 2 (2.5) unts of tme startng at zero, τ 1 msses ts deadlne because only 1.5 unts are left before ts deadlne (4) and our guarantee s not honored (ths s how CAPA would schedule the task set). However, wth the fnal schedulng parameters the task set honors our guarantee. That s, f τ 2 only uses C 2, then t means that at ts zero-slack nstant (5), t wll only run for 1 2 a tme unt more leavng tme unts for τ 1 (that only needs 1 more) to complete ts C2. o D. Proof of Convergence of the Zero-Slack Computaton Because our scheme reles on the convergence of the zeroslack computaton algorthm, we state the assumptons of the algorthm and prove ts convergence. The zero-slack computaton algorthm assumes the use of a non-anomalous 2 fxed-prorty schedulng algorthm. These algorthms ensure that reducng any C j of τ j Γ never 2 Some multprocessor schedulng schemes are known to be anomalous, where ths property does not hold. ncreases the response tme W τ Γ. Unprocessor schedulng algorthms lke RMS and CAPA are known to be nonanomalous wth respect to computaton tme [20]. Assumng the use of ths algorthm, we can prove convergence. Lemma 1: When non-anomalous schedulng algorthms are used to calculate the slack vectors, the Compute Fnal Zero- Slack Instants algorthm (Algorthm 1) converges. Proof: The proof follows from the followng three loop nvarants for Algorthm 1: 1) Z 1 0 τ Γ 2) Z 1 Z0 τ Γ 3) Z 1 D τ Γ Gven these three loop nvarants, we can observe that Z 1 starts at 0, never decreases snce Z 1 Z 0, and s bounded to be less than or equal to D. Therefore, Algorthm 1 converges. We now prove that these three loop nvarants hold true for Algorthm 1. Z 1 Z0 τ Γ: At the begnnng of the teraton, task τ s splt nto C c unts of executon n the crtcal mode (C), and C o Cc unts of executon n the normal mode (N). Durng the teraton, the executon tme n the crtcal mode C c never ncreases. For any non-anomalous schedulng algorthm, the slack avalable n the crtcal mode (C) can never decrease snce C c never ncreases. Therefore, the zero slack nstant Z1 calculated usng the slack avalable n the crtcal mode (C) s never pushed earler than Z 0.e. decreased n value. Z 1 0 τ Γ: ths follows from the ntalzaton condton Z 1 = 0 and the prevously establshed nvarant that Z 1 Z0 τ Γ. Z 1 D τ Γ: Consder a task τ wth a zero-slack schedulng nstant of D. The amount of tme spent n the crtcal mode (C) s 0. Therefore, there s no addtonal slack avalable n the crtcal mode (C). Ths mples that the zeroslack schedulng nstant can be moved no further than D. E. Runtme Behavor of The Zero-Slack Schedulng The central pece of the runtme mechansm of our scheme s the dual-mode executon mplementaton. It s mportant to keep the overhead of such a mechansm as lean as possble. We can acheve ths by usng the zero-slack nstants calculated offlne as tmers to swtch to C mode. These tmers then mark the tme when the lower-crtcalty tasks are suspended. For ths reason, we need to keep track of the crtcalty level of the tasks to be able to dstngush whch tasks need to be suspended. At the same tme, we need to keep track of whch tasks are n C mode n order to know whch tasks to resume when a task fnshes ts C mode. In a resource reservaton framework, the admsson test can be performed at the tme of the resource reservaton and the correspondng zero-slack nstants can be computed. F. Effectveness of Scheme To evaluate the effectveness of our schedulng algorthm, we defned a metrc to measure a reducton of the blockng densty factors (P B and CB ) defned n Secton III. For

6 CB, we know that a task τ j ζ < ζ j mposes a blockng term cb j = Co j n the worst case. For P B, a task τ j ζ j < ζ mposes a blockng term pb j = C j n the worst case. Because the blockng factors at each task level may nvolve parts of the blockng factors of other tasks, the fnal tasksetwse blockng, s calculated by takng the maxmum blockng among all the tasks: T B Γ = max τ Γ (CB + P B ) Ths total blockng s the maxmum blockng present n the task set Γ whose crtcalty-awareness s mproved by zero-slack schedulng. We call ths combned blockng mxedcrtcalty blockng (MC blockng). It s worth notng that ths value depends on the level of nverson between prortes and crtcaltes, as well as ther budgets and perods. For nstance, for the task set n Table I, the blockng factors are defned as: P B 1 = C2 D 1 = = P B 2 = 0 (because t s the lowest prorty) CB 1 = 0 (because t s the lowest crtcalty) CB 2 = Co 1 D 1 = 2 4 = 0.5 Hence, the MC blockng of the task set s the maxmum of the sum of the factors that s equal to ) Resdual MC blockng: The effectveness of an algorthm n reducng MC blockng s calculated by the amount of resdual MC blockng that the algorthm s not able to remove. For nstance, n the task set of Table I, when crtcalty s used as the schedulng prorty (CAPA), both CB 1 and CB 2 are zero. However, P B 1 = C2 D 1 = = (and P B 2 = 0). Takng the maxmum of the factors the resdual MC blockng left by the algorthm s On the other hand, when RMS s used, then P B 1 = P B 2 = 0. But CB 2 = Co 1 D 1 = 2 4 = 0.5 (and CB 1 = 0), s the resdual MC blockng left by the algorthm. Fnally, n the case of the zero-slack scheduler, the factors are reduced to P B 1 = Cc 2 D 1 = = 0.125, P B 2 = 0, CB 1 = 0, and CB 2 = C1 D 2 = C1 D D 2 = = ( 1 D 2 s the fracton of C 1 that executes n the second arrval of τ 1 before the zero-slack nstant). Hence, the resdual MC blockng s It s worth notng that the zero-slack scheduler reduces both P B and CB. P B s reduced because only a fracton of the C2 o (0.5) s run wth a prorty greater than the schedulng prorty (deadlne-monotonc prorty). CB s reduced because the swtchng to crtcal mode at the zeroslack nstant of τ 2 reduces ths blockng from 2 4 to 3 8 (2 n the frst perod of τ 1 and 1 n ts second perod). The reducton n crtcalty blockng s just enough to enable the completon of C2. o Furthermore, the fact that P B 1 was reduced from to 0.125, when compared to CAPA, ndcates that the zero-slack algorthm has further capacty to tolerate a larger prorty blockng wthout breakng the schedulng guarantee. We capture ths tolerance wth a metrc we call laxty densty. 2) Laxty Densty: The laxty densty of a task τ measures the avalable densty for τ after we dscount the preemptons of hgher-prorty tasks and the blockng factors left by the schedulng algorthm of nterest for ths task. The laxty densty for a task τ s then defned as: LaxU = A() Co (P B + CB ) C j (3) D D j where: τ j H hc A() s the avalable densty for task τ for a specfc prorty-based preemptve scheduler. For nstance, for rate-monotonc schedulng wth n mplct-deadlne tasks, ths s n(2 1 n 1) (see [21]) Co D s the overload densty of τ P B + CB s the blockng factor for the schedulng algorthm of nterest C j τ j H hc D j s the densty consumed by the hgherprorty and hgher-crtcalty tasks (H hc ) The laxty densty allows us to see how an algorthm spreads the laxty of the system across the dfferent tasks. Hence a successful algorthm wll keep the mnmum laxty across the tasks of the task set always greater than or equal to zero. Ths means that, an algorthm X s better than an algorthm Y, f there exst task sets where the mnmum laxty under Y goes below zero whle under X s greater than or equal to zero, but f there exsts a task set wth a mnmum laxty under X below zero, ts mnmum laxty under Y s also below zero. We wll use ths metrc to evaluate the performance of our zero-slack RM once we dscuss ts detals n the next secton. V. THE ZERO-SLACK RATE-MONOTONIC SCHEDULING (ZERO-SLACK-RM) The zero-slack schedulng polcy works on top of a prortybased preemptve scheduler and reles on the calculaton of slack vectors provded by such scheduler. In ths secton, we present the slack vector calculaton for the rate-monotonc scheduler. We chose the RM scheduler to smplfy the explanaton of the zero-slack scheme. Hence, n ths secton we wll assume mplct deadlnes (D = T ). A. Worst-Case Phasng of Dual-Mode Tasks Key to the calculaton of the slack vectors n the ratemonotonc schedulng s the phasng of the tasks. For a snglemode executon, Lu and Layland [21] proved that the phasng that creates the maxmum preempton for a task τ happens when every task τ j prorty(τ j ) < prorty(τ ) arrves at the same tme as τ. However, n a dual-mode task, ths worstcase phasng does not hold. Ths s because, when tasks reach ther zero-slack nstants, they wll suspend lower-crtcalty tasks. On the one hand, ths suspenson acts, as ntended, to avod preemptons suffered by task τ from lower-crtcalty tasks. However, t also acts as a preempton when hghercrtcalty tasks suspend τ. Hence, to calculate the worst-case delay mposed by ths type of preempton, we need to algn all the suspensons n the same way as the perod arrvals. Unfortunately, f we algn the zero-slack nstants of the hghercrtcalty tasks, we may msalgn the arrval of hgher-prorty tasks. In other words, t s not always possble to algn both

7 TABLE II ZERO-SLACK-RM SCHEDULED TASK SET Task C C o T D Crtcalty Prorty ZS Instant τ τ τ τ 0 τ 1 τ the worst-case arrval of the tasks and the zero-slack nstants. The mplcaton of ths msalgnment s that we cannot create a sngle ntegrated crtcal zone based on the algnment of both types of preemptons. As a result, we take a pessmstc approach by assumng that the effects of both algnments always happen. Although the worst-case phasng may not exst, t provdes an upper bound on the total nterference mposed on task τ. Ths can be shown as follows. Before the zero-slack nstant, the maxmum nterference from hgher-prorty tasks happens when they are released smultaneously wth τ (from [21]). After the zero-slack nstant, τ effectvely blocks all the lower - crtcalty tasks. Therefore, the nterference can only arse from hgher-crtcalty tasks. By swtchng all the hgher-crtcalty tasks to ther crtcal mode (C) along wth τ, the nterference suffered by τ n the crtcal mode (C) s also maxmzed. B. A Zero-Slack-RM Schedulng Example Let us use an example to llustrate the characterstcs of the zero-slack-rm scheduler. Table II presents a task set wth the prortes assgned by the rate-monotonc scheduler and the zero-slack nstants calculated by our algorthm. Due to space lmtatons, we wll focus our dscussons on τ 1. Fgure 2 presents the crtcal zone of ths task. In ths fgure, we can see the preempton from τ 0 n the N mode of τ 1 for 50 unts of tme. After ths, τ 1 runs for 10 unts and then reaches ts zero-slack nstant at tme 60, swtchng to C mode. In C mode, t suspends the lowercrtcalty task τ 0, but at the same tme t s suspended by the hgher-crtcalty task τ 2. Ths suspenson s the pessmstc approach we use due to the absence of an exact worst-case phasng (as dscussed n Subsecton V-A). τ 2 then runs for C 2 (40) unts and resumes the lower-crtcalty tasks. However, n order to mantan the crtcalty order, ths resumpton s mplemented as a stack, meanng that t only returns to the prevous crtcalty level (leavng τ 0 suspended). Then, τ 1 can contnue executng completng ts C1 o (100) at tme 190. Each task n the task set has ts own (pessmstc) crtcal zone smlar to the one presented n Fgure 2, but they are unfortunately not necessarly algned wth each other. C. Calculatng The Slack Vectors n Zero-Slack-RM In order to calculate the slack vectors n zero-slack-rm, we frst defne an nterferng taskset for each task τ under zero-slack-rm. For each of the two executon modes (C and N=RM), there could be dfferent tasksets that could nterfere wth (.e. preempt) τ. We use Cj E to represent the effectve executon tme that wll be consdered for each nterferng task Zero-slack nstant Fg. 2. Crtcal Zone of Task 1 τ j. In partcular, n the RM mode, the task set that would nterfere wth task τ s defned as: Γ n = Γ rm = H lc H hc L hc S where, * H lc s the set of tasks wth hgher rate-monotonc prortes but lower crtcalty. These tasks attempt to run for C o unts. That s, τ j H lc, CE j Cj o. * H hc s the set of tasks wth hgher rate-monotonc prortes and hgher crtcalty. These tasks run for ther respectve non-overloaded computaton tme. If a hgher-crtcalty task executes beyond ts non-overloaded computaton tme, then the schedulng guarantee of lower-crtcalty task τ need not be honored. Hence, τ j H hc, Cj E C j. * L hc s the set of tasks wth lower rate-monotonc prortes and hgher crtcalty. These tasks are consdered to be runnng n ther C mode. In fact, we only need to consder the part of computaton that runs n ts C mode. That s, assumng that a rm, j contans the slack avalable to task τ j τ j L hc n RM mode, whle honorng the guarantee of task τ, then Cj E (C j a rm, j ). Intally, we assume that there s no slack avalable n RM mode (a rm, j = 0), but as we move the zero-slack nstants of the tasks towards the end of ther perod we wll dscover addtonal slack n RM mode reducng Cj E further, n turn the avalable slack n RM mode ncreases. * S s the set of tasks wth the same level of crtcalty but hgher prorty. These tasks are runnng at ther C o,.e., τ j S, Cj E Cj o. The nterferng task set for C mode of task τ s defned as: Γ c = H hc L hc S wth the same set of defntons as before. Note that the low crtcalty tasks are excluded snce the lower-crtcalty tasks are blocked durng τ C mode of executon. Wth these task sets, we create a slack vector for each executon zone usng Algorthm 3. Ths algorthm receves as parameters the ndex of the task we are evaluatng and the nterferng task set, ordered by ncreasng order of prorty. Algorthm 3 calculates the slack vector gven a task τ and ts nterferng task set Γ, for the executon mode under consderaton. The algorthm uses a parameter C v to accumulate the total slack avalable to τ over the entre perod of duraton t. The outer loop of Algorthm 3, adds new entres to the slack

8 vector. The nner loop computes the tme nstant R current by whch C v unts of slack are avalable, and the tme nstant b at whch the next nterferng task I m arrves. Dfference between b and R current s the next avalable slack regon for τ. Algorthm 3 GetSlackVector(, Γ, t = T ): Slack Vector Calculaton 1: ndex 0 ; C v 0 2: repeat 3: R current C v l b 0 4: repeat 5: R prevous R current 6: R current C v + j Γ Rprevous 7: b t ; I m 8: for (j Γ) do 9: A Rprevous T j T j 10: f (A < b) then 11: b A ; I m j 12: end f 13: end for 14: f (R prevous = R current ) then T j C e j 15: R current R current + CI e m 16: end f 17: untl (R prevous = R current or R current t) 18: V [ndex].slack mn(b, t) R current 19: V [ndex].tme R current 20: C v Cv + (mn(b, t) R current) 21: ndex : untl (R current t) 23: return V D. Propertes of The Zero-Slack-RM Scheduler We now present the propertes of our zero-slack RM scheduler. 1) Subsumes CAPA: Any task set schedulable under Crtcalty-As-Prorty Assgnment(CAPA) s also schedulable under the zero-slack schedulng scheme. Proof: Algorthm 1 starts wth assgnng Z 0 = Z 1 = 0 for all tasks τ. Under ths assgnment of zero-slack nstants, the zero-slack scheduler behaves essentally lke a CAPA scheme, snce whenever τ s released all the lower crtcalty tasks are mmedately blocked due to τ swtchng to ts crtcal mode (Z 1 = 0). Therefore, f the task set s schedulable under CAPA, t should be schedulable at the frst teraton of Algorthm 1. In ths scenaro, we now nductvely prove that each task τ remans schedulable over subsequent teratons. Durng subsequent teratons of the Compute Fnal zeroslack Instant algorthm, the Algorthm 2 s used to transfer addtonal computaton of up to k u from the crtcal mode (C) to the normal mode (N). Ths transfer s performed only to use up the slack of k u avalable n the normal mode (N) as calculated usng SlackUpToInstant tll the zero-slack nstant t 1. Consderng any task τ, ths transfer of computaton does not ncrease the blockng terms suffered by τ from hgher crtcalty tasks executng n ther C mode. The normal mode N of τ remans unaffected, snce the addtonal computaton s transferred to only fll up the avalable slack. Therefore, the response tme of task τ only reduces wth subsequent teratons. Hence, f task τ was schedulable n the prevous teraton, t contnues to reman schedulable. Ths completes the proof by nducton. 2) Corollary: Graceful Degradaton: When a task τ j executes for a C such that C j < C Cj o, then only tasks τ wth ζ > ζ j can mss ther deadlne. Ths follows from the schedulng guarantee provded by zero-slack schedulng, whch was descrbed n subsecton IV-A, and the property that rate-monotonc schedulng s non-anomalous. 3) Subsumes RM Scheduler: Any task set schedulable under rate-monotonc schedulng s also schedulable under the zero-slack schedulng scheme. Proof: For any task set Γ consstng of tasks τ = (C, T ) schedulable under the RM schedulng scheme, consder an equvalent Γ z wth tasks τ z = (C, C o, T, ζ ) wth C = C o = C and ζ = π, where π s the prorty assgned to task τ under RM schedulng. Schedulng the task set Γ z usng CAPA produces the same schedule as the RM scheduler, snce the prortes are completely algned wth the crtcalty under the chosen ζ values. Hence, Γ z s also schedulable under CAPA, snce t s schedulable under RM schedulng. Usng the property that zero-slack schedulng subsumes CAPA, t follows that Γ z s also schedulable under zero-slack schedulng. τ z 4) Utlzaton Propertes: A task set Γ z wth wth tasks = (C, C o, T, ζ ) s schedulable under zero-slack-rm f: U rm U z = Co k C j + + T k T j ζ k =ζ andt k <T ζ j<ζ andt j<t C o + ζ p<ζ andt p T C p T (4) From the perspectve of τ, under overloaded condtons, the utlzaton s Co T. The worst-case nterferng utlzaton arsng from the equal-crtcalty hgher-prorty tasks τ k s gven by Co k T k, snce they could potentally msbehave. The worst-case nterferng utlzaton arsng from hgher-crtcalty hgher-prorty tasks τ s gven by Cj T j, and the blockng utlzaton from hgher-crtcalty lower-prorty tasks s gven by Cp T. We do not have to consder the lower-crtcalty tasks snce they cannot affect the schedulablty of task τ. If the lower-crtcalty tasks could mpose any nterference on τ that causes t to potentally become unschedulable, then the zeroslack nstant would have been moved earler to ensure the schedulablty of τ. It s mportant to note here that Equaton (4) s only used to present a property of the Zero-Slack RM Scheduler. For the purposes the admsson control, we use the response-tme test based slack dscovery algorthm descrbed n Algorthm 3. E. Laxty Utlzaton Comparson The propertes of the zero-slack-rm algorthm can be depcted graphcally n a comparson of ts laxty utlzaton

9 Laxty Utlzaton (0,1,2) (0,2,1) (1,0,2) (1,2,0) (2,0,1) (2,1,0) Fg. 3. Crtcalty Vector Laxty Utlzaton vs Crtcalty-to-Prorty Inverson RMSMn CAPAMn ZSMn RMSTotal CAPATotal ZSTotal aganst that of both RMS and CAPA. Fgure 3 depcts ths comparson. We decded not to nclude perod transformaton n the comparson gven that n the end t wll behave as RMS wth overloaded computaton tme, as descrbed n secton II. For the comparson, we use the task set presented n Table II. We then vary the crtcalty of the task set (as a crtcalty vector [ζ 0, ζ 1, ζ 2 ]) to explore all the dfferent degrees of prorty-to-crtcalty nverson. Because ths task set s harmonc, we use 1 as the avalable utlzaton A() for each task τ, the maxmum for a harmonc task set under ratemonotonc schedulng. The assgned crtcalty vector s shown n the X-axs, and the mnmum and total laxty utlzaton values are plotted. It s to be noted here that the total laxty utlzaton can be greater than 1 because the laxty avalable to a task could also be avalable to other tasks n the system. Fgure 3 shows that there are crtcalty assgnments for whch ether RMS or CAPA or both algorthms are unable to acheve a mnmum laxty greater than or equal to 0 (to make t schedulable), whereas zero-slack schedulng ensures a mnmum laxty greater than or equal to zero n all cases. In partcular, ths fgure shows that, for a crtcalty assgnment [1, 0, 2], both RMS and zero-slack schedulng algorthms are feasble, whereas, CAPA renders the task set unschedulable. However, for the crtcalty assgnment [1, 2, 0], RMS s unschedulable, whereas, CAPA and zero-slack schedulng contnue to reman feasble. Ths llustrates that zero-slack schedulng subsumes both RMS and CAPA. We observe that the total laxty avalable under zero-slack schedulng s bounded by the maxmum total laxty avalable under RMS and CAPA. Specfcally, Fgure 3 shows that zeroslack schedulng always has a total laxty that s ether upper bounded by the laxty of RMS or by the laxty of CAPA. However, there are crtcalty assgnments lke [2, 1, 0], where both RMS and CAPA have negatve mnmum laxty but zeroslack schedulng remans feasble. The reason for ths behavor s that the key factor determnng feasblty s the dstrbuton of avalable laxty, as opposed to the total avalable laxty tself. For the crtcalty assgnment [2, 1, 0], whle CAPA has more laxty than zero-slack schedulng, the latter dstrbutes the laxty more evenly among the tasks by controllng the swtchng between dfferent executon modes. F. Relaxng Constrants For smplcty of presentaton, the dscusson so far has assumed mplct-deadlne constrants and accommodated ratemonotonc schedulng. It s mportant to note here that our slack vector calculaton n Algorthm 3 uses generc statcprorty preemptve schedulng results [22], therefore enablng t to be readly extended to other prorty assgnments. For nstance, t can be extended to deadlne-monotonc schedulng, whch s an optmal schedulng algorthm for systems wth constraned deadlnes (D T ). In the context of deadlnemonotonc schedulng, t s straght forward to use t = D nstead of T n Algorthm 3, after assgnng deadlne-monotonc prortes to tasks. Due to space constrants, we do not dscuss the densty-based propertes of such a Zero-Slack Deadlne- Monotonc (ZSDM) scheduler. Subsequent support for release jtter also follows from exstng results for fxed-prorty schedulng based on those presented n [22]. Ths s due to the fact that we employ ther basc analyss n step 6 of Algorthm 3. Replacng ths response-tme computaton wth the results n [22] wll enable such extensons. In order to smplfy the presentaton of our results, we avod a detaled dscusson of such extensons. VI. IMPLEMENTATION The zero-slack-rm scheduler and ts runtme enforcement mechansm have been mplemented n Lnux/RK [8]. Lnux/RK s a resource kernel that creates tme (or space) parttons of resources, e.g. CPU, provdng temporal solaton between these parttons. These parttons are called resource reserves. Tasks are then assocated wth these reserves. Tme reserves are specfed as a consumpton tme (C) of the resource over a perod of tme (T ). Wth ths specfcaton, Lnux/RK ensures that the tasks do not consume more tme than the allocated one over the specfed perod. These reserves are mplemented as perodc servers [23] where ther budget s replenshed perodcally and f ths budget s exhausted before the end of the perod, the assocated tasks are stopped. These tasks are resumed once the reserve s replenshed. Our runtme enforcement s mplemented as a new type of reserve n Lnux/RK wth two executon modes. Ths reserved type s called a crtcalty reserve. A crtcalty reserve adds two addtonal parameters to the regular CPU reserve: the crtcalty (ζ) and the zero-slack nstant (ZS), and uses C o as the C consumpton tme. A. Zero-Slack Enforcement The dual executon mode s mplemented as a zero-slack enforcement. Specfcally, a tmer s setup at the start of the perod of every task τ to expre accordng at the zero-slack nstant. If, when the tmer expres, τ has not fnshed ts actvaton, t suspends all the tasks wth a crtcalty lower that ts own, thereby enterng ts C mode. It also sets a global

10 crtcalty-level varable to equal ts own crtcalty ndcatng that no task wth lower crtcalty should be runnng (storng the prevous value as set by any prevous task n C mode n a stack). In addton, t puts the suspended tasks n a lst that s ordered by crtcalty. Once τ fnshes ts executon, then t resumes the tasks n C mode for the next crtcalty level, or all the tasks f no other task s n C mode. Fnally, when a reserve s replenshed, the crtcalty level of the assocated task s compared aganst the current crtcalty level and t s added to the suspended lst f t s lower. VII. CONCLUSIONS In ths paper, we studed the problems ntroduced by mxedcrtcalty task sets. We also dentfed the shortcomngs of tradtonal temporal solaton schemes to prevent nter-task nterference n mxed-crtcalty systems. In partcular, we characterzed and quantfed the crtcalty nverson problem that arses when a hgher-crtcalty tasks msses ts deadlne because t s forced to wat for a lower-crtcalty task. We then presented a new schedulng scheme wth a new protecton polcy, whch we call as asymmetrc protecton that only prevents nterference to hgher-crtcalty tasks from lowercrtcalty tasks. Our scheme has a generc offlne zero-slack computaton algorthm and runtme enforcement mechansm that can be used wth any prorty-based preemptve scheduler that has non-anomalous schedulng behavor. The other part of our scheme s a slack analyss algorthm that s specfc to each prorty-based preemptve scheduler. We developed the slack analyss algorthm for RMS and proved multple propertes of the ntegrated algorthm (zero-slack RM). Ths algorthm provdes the same level of protecton aganst crtcalty nverson as the best known prorty assgnment for ths purpose, crtcalty as prorty (CAPA). Zero-slack RM provdes the same level of schedulable utlzaton as RMS f crtcalty levels are equal to prortes, and better f they are dfferent and we use our schedulng guaranteed. We also developed two metrcs to evaluate such algorthms. Frst, a mxedcrtcalty blockng (MC blockng) metrc measures how much blockng reducton s possble n a specfc task set. Secondly, laxty utlzaton measures how well a schedulng algorthm reduces the MC blockng of the task set to make the task set schedulable. These metrcs are then used to show that zeroslack RM always outperforms or equals CAPA and RMS to spread the MC blockng reducton of a task set to the tasks that need t the most. Fnally, we demonstrate the practcalty of our runtme enforcement wth an mplementaton n Lnux/RK. REFERENCES [1] D. Homan, Desgnng future systems for arworthness certfcaton, cdgll/cpsweek09 MCAR/MCAR%20overvew%20BoF%20CPS%20PA% 20approved.pdf, [2] P. Meja-Alvarez, R. Melhem, and D. Mosse, An ncremental approach to schedulng durng overloads n real-tme systems, n In Proceedngs of the 21st, IEEE RTSS, [3] G. Buttazzo, M. Spur, and F. Sensn, Value vs deadlne schedulng n overload condtons, n In Proceedngs of the 16st, IEEE Real-Tme Systems Symposum, [4] G. Koren and D. Shasha, Dover; an optmal on-lne schedulng algorthm for overloaded real-tme systems, n In Proceedngs of the IEEE Real-Tme Systems Symposum, [5] S. Baruah and J. Hartsa, Schedulng for overload n real-tme systems, IEEE Transactons on Computers, vol. 46, no. 9, pp , [6] M. Gardner and L. J.W.S., Performance algorthms for schedulng real-tme systems wth overrun and overload, n In Proceedngs of the 11th Euromcro Conference on Real-Tme Systems, [7] C.-S. Shh, P. Gant, and L. Sha, Schedulablty and farness for computaton tasks n survellance radar systems, n In Proceedngs of the 10th RealTme and Embedded Computng Systems and Applcatons Conference, [8] S. Okawa and R. Rajkumar, Lnux/rk: A portable resource kernel n lnux, In Proceedngs of the 19th, IEEE Real-Tme Systems Symposum, [9] A. Sundaram, P. Chandra, P. Goyal, Shenoy, J. Sahn, and H. Vn, Applcaton performance n the qlnux multmeda operatng system, In Proceedngs of the Eghth ACM Conference on Multmeda, November [10] X. Feng, Towards real-tme enabled mcrosoft wndows, Proceedngs of the 5th Internatonal Conference On Embedded Software, pp , [11] M. J. Paul, M. B. Jones, P. J. Leach, R. P. Draves, and J. S. Barrera, Modular real-tme resource management n the ralto operatng system, Mcrosoft Research, Advanced Technology Dvson, pp , [12] G. Buttazzo, G. Lpar, and L. Aben, Elastc task model for adaptve rate control, IEEE Real-Tme Systems Symposum, pp , [13] R. Davs and A. Wellngs, Dual prorty schedulng, In Proceedngs of the 16th IEEE RTSS, pp , Dec [14] S. Cho, S.-K. Lee, A. Han, and K.-J. Ln, Effcent realtme schedulng algorthms for multprocessor systems, IEICE Trans. Communcatons, E85-B(12), pp , Dec [15] M. Crne and T. Baker, Edzl schedulng analyss, In Proceedngs of ECRTS, pp. 9 18, July [16] S. Vestal, Preemptve schedulng of mult-crtcalty systems wth varyng degrees of executon tme assurance, In Proceedngs of the Real-Tme Systems Symposum (Tucson, AZ, December 2007), pp , [17] L. Sha, J. P. Lehoczky, and R. Rajkumar, Task schedulng n dstrbuted real-tme systems, n Proceedngs of the IEEE Industral Electroncs Conference. IEEE, [18] S. Baruah and S. Vestal, Schedulablty analyss of sporadc tasks wth multple crtcalty specfcatons, Euromcro Conference on Real-Tme Systems, pp , July [19] L. Sha, R. Rajkumar, and J. Lehoczky, Prorty nhertance protocols: an approach to real-tme synchronzaton, Computers, IEEE Transactons on, vol. 39, no. 9, pp , Sep [20] S. Baruah and A. Burns, Sustanable schedulng analyss, n RTSS 06: Proceedngs of the 27th IEEE Internatonal Real- Tme Systems Symposum, 2006, pp [21] C. L. Lu and J. W. Layland, Schedulng algorthms for multprogrammng n a hard-real-tme envronment, J. ACM, vol. 20, no. 1, pp , [22] K. Tndell, A. Burns, and A. Wellngs, An extendble approach for analysng fxed prorty hard real-tme tasks, The Journal of Real-Tme Systems, 6(2), pp , March [23] T. M. Ghazale and T. P. Baker, Aperodc servers n a deadlne schedulng envronment, Real-Tme Syst., vol. 9, no. 1, pp , 1995.

Real-Time Systems. Multiprocessor scheduling. Multiprocessor scheduling. Multiprocessor scheduling

Real-Time Systems. Multiprocessor scheduling. Multiprocessor scheduling. Multiprocessor scheduling Real-Tme Systems Multprocessor schedulng Specfcaton Implementaton Verfcaton Multprocessor schedulng -- -- Global schedulng How are tasks assgned to processors? Statc assgnment The processor(s) used for

More information

Embedded Systems. 4. Aperiodic and Periodic Tasks

Embedded Systems. 4. Aperiodic and Periodic Tasks Embedded Systems 4. Aperodc and Perodc Tasks Lothar Thele 4-1 Contents of Course 1. Embedded Systems Introducton 2. Software Introducton 7. System Components 10. Models 3. Real-Tme Models 4. Perodc/Aperodc

More information

Last Time. Priority-based scheduling. Schedulable utilization Rate monotonic rule: Keep utilization below 69% Static priorities Dynamic priorities

Last Time. Priority-based scheduling. Schedulable utilization Rate monotonic rule: Keep utilization below 69% Static priorities Dynamic priorities Last Tme Prorty-based schedulng Statc prortes Dynamc prortes Schedulable utlzaton Rate monotonc rule: Keep utlzaton below 69% Today Response tme analyss Blockng terms Prorty nverson And solutons Release

More information

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud

Resource Allocation with a Budget Constraint for Computing Independent Tasks in the Cloud Resource Allocaton wth a Budget Constrant for Computng Independent Tasks n the Cloud Wemng Sh and Bo Hong School of Electrcal and Computer Engneerng Georga Insttute of Technology, USA 2nd IEEE Internatonal

More information

Two Methods to Release a New Real-time Task

Two Methods to Release a New Real-time Task Two Methods to Release a New Real-tme Task Abstract Guangmng Qan 1, Xanghua Chen 2 College of Mathematcs and Computer Scence Hunan Normal Unversty Changsha, 410081, Chna qqyy@hunnu.edu.cn Gang Yao 3 Sebel

More information

Improved Worst-Case Response-Time Calculations by Upper-Bound Conditions

Improved Worst-Case Response-Time Calculations by Upper-Bound Conditions Improved Worst-Case Response-Tme Calculatons by Upper-Bound Condtons Vctor Pollex, Steffen Kollmann, Karsten Albers and Frank Slomka Ulm Unversty Insttute of Embedded Systems/Real-Tme Systems {frstname.lastname}@un-ulm.de

More information

Module 9. Lecture 6. Duality in Assignment Problems

Module 9. Lecture 6. Duality in Assignment Problems Module 9 1 Lecture 6 Dualty n Assgnment Problems In ths lecture we attempt to answer few other mportant questons posed n earler lecture for (AP) and see how some of them can be explaned through the concept

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Desgn and Analyss of Algorthms May 4, 2015 Massachusetts Insttute of Technology 6.046J/18.410J Profs. Erk Demane, Srn Devadas, and Nancy Lynch Problem Set 9 Solutons Problem Set 9 Solutons Ths problem

More information

Kernel Methods and SVMs Extension

Kernel Methods and SVMs Extension Kernel Methods and SVMs Extenson The purpose of ths document s to revew materal covered n Machne Learnng 1 Supervsed Learnng regardng support vector machnes (SVMs). Ths document also provdes a general

More information

Structure and Drive Paul A. Jensen Copyright July 20, 2003

Structure and Drive Paul A. Jensen Copyright July 20, 2003 Structure and Drve Paul A. Jensen Copyrght July 20, 2003 A system s made up of several operatons wth flow passng between them. The structure of the system descrbes the flow paths from nputs to outputs.

More information

ECE559VV Project Report

ECE559VV Project Report ECE559VV Project Report (Supplementary Notes Loc Xuan Bu I. MAX SUM-RATE SCHEDULING: THE UPLINK CASE We have seen (n the presentaton that, for downlnk (broadcast channels, the strategy maxmzng the sum-rate

More information

Improving the Sensitivity of Deadlines with a Specific Asynchronous Scenario for Harmonic Periodic Tasks scheduled by FP

Improving the Sensitivity of Deadlines with a Specific Asynchronous Scenario for Harmonic Periodic Tasks scheduled by FP Improvng the Senstvty of Deadlnes wth a Specfc Asynchronous Scenaro for Harmonc Perodc Tasks scheduled by FP P. Meumeu Yoms, Y. Sorel, D. de Rauglaudre AOSTE Project-team INRIA Pars-Rocquencourt Le Chesnay,

More information

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Pre-emptive Scheduling for Sporadic Tasksets with Arbitrary Deadlines

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Pre-emptive Scheduling for Sporadic Tasksets with Arbitrary Deadlines Quantfyng the Sub-optmalty of Unprocessor Fxed Prorty Pre-emptve Schedulng for Sporadc Tasksets wth Arbtrary Deadlnes Robert Davs, Sanjoy Baruah, Thomas Rothvoss, Alan Burns To cte ths verson: Robert Davs,

More information

Partitioned Scheduling of Multi-Modal Mixed- Criticality Real-Time Systems on Multiprocessor Platforms

Partitioned Scheduling of Multi-Modal Mixed- Criticality Real-Time Systems on Multiprocessor Platforms Unversty of Pennsylvana ScholarlyCommons Departmental Papers (CIS) Department of Computer & Informaton Scence 4-2014 Parttoned Schedulng of Mult-Modal Mxed- Crtcalty Real-Tme Systems on Multprocessor Platforms

More information

Computer Control: Task Synchronisation in Dynamic Priority Scheduling

Computer Control: Task Synchronisation in Dynamic Priority Scheduling Computer Control: Task Synchronsaton n Dynamc Prorty Schedulng Sérgo Adrano Fernandes Lopes Department of Industral Electroncs Engneerng School Unversty of Mnho Campus de Azurém 4800 Gumarães - PORTUGAL

More information

AN EXTENDIBLE APPROACH FOR ANALYSING FIXED PRIORITY HARD REAL-TIME TASKS

AN EXTENDIBLE APPROACH FOR ANALYSING FIXED PRIORITY HARD REAL-TIME TASKS AN EXENDIBLE APPROACH FOR ANALYSING FIXED PRIORIY HARD REAL-IME ASKS K. W. ndell 1 Department of Computer Scence, Unversty of York, England YO1 5DD ABSRAC As the real-tme computng ndustry moves away from

More information

Lecture Notes on Linear Regression

Lecture Notes on Linear Regression Lecture Notes on Lnear Regresson Feng L fl@sdueducn Shandong Unversty, Chna Lnear Regresson Problem In regresson problem, we am at predct a contnuous target value gven an nput feature vector We assume

More information

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound Fxed-Prorty Multprocessor Schedulng wth Lu & Layland s Utlzaton Bound Nan Guan, Martn Stgge, Wang Y and Ge Yu Department of Informaton Technology, Uppsala Unversty, Sweden Department of Computer Scence

More information

Partitioned Scheduling of Multi-Modal Mixed-Criticality Real-Time Systems on Multiprocessor Platforms

Partitioned Scheduling of Multi-Modal Mixed-Criticality Real-Time Systems on Multiprocessor Platforms Parttoned Schedulng of Mult-Modal Mxed-Crtcalty Real-Tme Systems on Multprocessor Platforms Donso de Nz SEI, Carnege Mellon Unversty Lnh T.X. Phan Unversty of Pennsylvana Abstract Real-tme systems are

More information

Lecture 4: November 17, Part 1 Single Buffer Management

Lecture 4: November 17, Part 1 Single Buffer Management Lecturer: Ad Rosén Algorthms for the anagement of Networs Fall 2003-2004 Lecture 4: November 7, 2003 Scrbe: Guy Grebla Part Sngle Buffer anagement In the prevous lecture we taled about the Combned Input

More information

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS Avalable onlne at http://sck.org J. Math. Comput. Sc. 3 (3), No., 6-3 ISSN: 97-537 COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

More information

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE Analytcal soluton s usually not possble when exctaton vares arbtrarly wth tme or f the system s nonlnear. Such problems can be solved by numercal tmesteppng

More information

Partitioned Mixed-Criticality Scheduling on Multiprocessor Platforms

Partitioned Mixed-Criticality Scheduling on Multiprocessor Platforms Parttoned Mxed-Crtcalty Schedulng on Multprocessor Platforms Chuanca Gu 1, Nan Guan 1,2, Qngxu Deng 1 and Wang Y 1,2 1 Northeastern Unversty, Chna 2 Uppsala Unversty, Sweden Abstract Schedulng mxed-crtcalty

More information

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals

Simultaneous Optimization of Berth Allocation, Quay Crane Assignment and Quay Crane Scheduling Problems in Container Terminals Smultaneous Optmzaton of Berth Allocaton, Quay Crane Assgnment and Quay Crane Schedulng Problems n Contaner Termnals Necat Aras, Yavuz Türkoğulları, Z. Caner Taşkın, Kuban Altınel Abstract In ths work,

More information

Improving the Quality of Control of Periodic Tasks Scheduled by FP with an Asynchronous Approach

Improving the Quality of Control of Periodic Tasks Scheduled by FP with an Asynchronous Approach Improvng the Qualty of Control of Perodc Tasks Scheduled by FP wth an Asynchronous Approach P. Meumeu Yoms, L. George, Y. Sorel, D. de Rauglaudre AOSTE Project-team INRIA Pars-Rocquencourt Le Chesnay,

More information

Two-Phase Low-Energy N-Modular Redundancy for Hard Real-Time Multi-Core Systems

Two-Phase Low-Energy N-Modular Redundancy for Hard Real-Time Multi-Core Systems 1 Two-Phase Low-Energy N-Modular Redundancy for Hard Real-Tme Mult-Core Systems Mohammad Saleh, Alreza Ejlal, and Bashr M. Al-Hashm, Fellow, IEEE Abstract Ths paper proposes an N-modular redundancy (NMR)

More information

The Schedulability Region of Two-Level Mixed-Criticality Systems based on EDF-VD

The Schedulability Region of Two-Level Mixed-Criticality Systems based on EDF-VD The Schedulablty Regon of Two-Level Mxed-Crtcalty Systems based on EDF-VD Drk Müller and Alejandro Masrur Department of Computer Scence TU Chemntz, Germany Abstract The algorthm Earlest Deadlne Frst wth

More information

Limited Preemptive Scheduling for Real-Time Systems: a Survey

Limited Preemptive Scheduling for Real-Time Systems: a Survey Lmted Preemptve Schedulng for Real-Tme Systems: a Survey Gorgo C. Buttazzo, Fellow Member, IEEE, Marko Bertogna, Senor Member, IEEE, and Gang Yao Abstract The queston whether preemptve algorthms are better

More information

Global EDF Scheduling for Parallel Real-Time Tasks

Global EDF Scheduling for Parallel Real-Time Tasks Washngton Unversty n St. Lous Washngton Unversty Open Scholarshp Engneerng and Appled Scence Theses & Dssertatons Engneerng and Appled Scence Sprng 5-15-2014 Global EDF Schedulng for Parallel Real-Tme

More information

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming

EEL 6266 Power System Operation and Control. Chapter 3 Economic Dispatch Using Dynamic Programming EEL 6266 Power System Operaton and Control Chapter 3 Economc Dspatch Usng Dynamc Programmng Pecewse Lnear Cost Functons Common practce many utltes prefer to represent ther generator cost functons as sngle-

More information

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity LINEAR REGRESSION ANALYSIS MODULE IX Lecture - 30 Multcollnearty Dr. Shalabh Department of Mathematcs and Statstcs Indan Insttute of Technology Kanpur 2 Remedes for multcollnearty Varous technques have

More information

Scheduling Motivation

Scheduling Motivation 76 eal-me & Embedded Systems 7 Uwe. Zmmer - he Australan Natonal Unversty 78 Motvaton n eal-me Systems Concurrency may lead to non-determnsm. Non-determnsm may make t harder to predct the tmng behavour.

More information

Chapter - 2. Distribution System Power Flow Analysis

Chapter - 2. Distribution System Power Flow Analysis Chapter - 2 Dstrbuton System Power Flow Analyss CHAPTER - 2 Radal Dstrbuton System Load Flow 2.1 Introducton Load flow s an mportant tool [66] for analyzng electrcal power system network performance. Load

More information

The Minimum Universal Cost Flow in an Infeasible Flow Network

The Minimum Universal Cost Flow in an Infeasible Flow Network Journal of Scences, Islamc Republc of Iran 17(2): 175-180 (2006) Unversty of Tehran, ISSN 1016-1104 http://jscencesutacr The Mnmum Unversal Cost Flow n an Infeasble Flow Network H Saleh Fathabad * M Bagheran

More information

Minimizing Energy Consumption of MPI Programs in Realistic Environment

Minimizing Energy Consumption of MPI Programs in Realistic Environment Mnmzng Energy Consumpton of MPI Programs n Realstc Envronment Amna Guermouche, Ncolas Trquenaux, Benoît Pradelle and Wllam Jalby Unversté de Versalles Sant-Quentn-en-Yvelnes arxv:1502.06733v2 [cs.dc] 25

More information

Clock-Driven Scheduling (in-depth) Cyclic Schedules: General Structure

Clock-Driven Scheduling (in-depth) Cyclic Schedules: General Structure CPSC-663: Real-me Systems n-depth Precompute statc schedule o-lne e.g. at desgn tme: can aord expensve algorthms. Idle tmes can be used or aperodc jobs. Possble mplementaton: able-drven Schedulng table

More information

Real-Time Operating Systems M. 11. Real-Time: Periodic Task Scheduling

Real-Time Operating Systems M. 11. Real-Time: Periodic Task Scheduling Real-Tme Operatng Systems M 11. Real-Tme: Perodc Task Schedulng Notce The course materal ncludes sldes downloaded from:! http://codex.cs.yale.edu/av/os-book/! and! (sldes by Slberschatz, Galvn, and Gagne,

More information

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound

Fixed-Priority Multiprocessor Scheduling with Liu & Layland s Utilization Bound Fxed-Prorty Multprocessor Schedulng wth Lu & Layland s Utlzaton Bound Nan Guan, Martn Stgge, Wang Y and Ge Yu Department of Informaton Technology, Uppsala Unversty, Sweden Department of Computer Scence

More information

Appendix B: Resampling Algorithms

Appendix B: Resampling Algorithms 407 Appendx B: Resamplng Algorthms A common problem of all partcle flters s the degeneracy of weghts, whch conssts of the unbounded ncrease of the varance of the mportance weghts ω [ ] of the partcles

More information

Lecture 4. Instructor: Haipeng Luo

Lecture 4. Instructor: Haipeng Luo Lecture 4 Instructor: Hapeng Luo In the followng lectures, we focus on the expert problem and study more adaptve algorthms. Although Hedge s proven to be worst-case optmal, one may wonder how well t would

More information

CHAPTER 17 Amortized Analysis

CHAPTER 17 Amortized Analysis CHAPTER 7 Amortzed Analyss In an amortzed analyss, the tme requred to perform a sequence of data structure operatons s averaged over all the operatons performed. It can be used to show that the average

More information

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1] DYNAMIC SHORTEST PATH SEARCH AND SYNCHRONIZED TASK SWITCHING Jay Wagenpfel, Adran Trachte 2 Outlne Shortest Communcaton Path Searchng Bellmann Ford algorthm Algorthm for dynamc case Modfcatons to our algorthm

More information

Keynote: RTNS Getting ones priorities right

Keynote: RTNS Getting ones priorities right Keynote: RTNS 2012 Gettng ones prortes rght Robert Davs Real-Tme Systems Research Group, Unversty of York rob.davs@york.ac.uk What s ths talk about? Fxed Prorty schedulng n all ts guses Pre-emptve, non-pre-emptve,

More information

CSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography

CSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography CSc 6974 and ECSE 6966 Math. Tech. for Vson, Graphcs and Robotcs Lecture 21, Aprl 17, 2006 Estmatng A Plane Homography Overvew We contnue wth a dscusson of the major ssues, usng estmaton of plane projectve

More information

Assortment Optimization under MNL

Assortment Optimization under MNL Assortment Optmzaton under MNL Haotan Song Aprl 30, 2017 1 Introducton The assortment optmzaton problem ams to fnd the revenue-maxmzng assortment of products to offer when the prces of products are fxed.

More information

Resource Reservation for Mixed Criticality Systems

Resource Reservation for Mixed Criticality Systems Resource Reservaton for Mxed Crtcalty Systems Guseppe Lpar 1, Gorgo C. Buttazzo 2 1 LSV, ENS - Cachan, France 2 Scuola Superore Sant Anna, Italy Abstract. Ths paper presents a reservaton-based approach

More information

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification

2E Pattern Recognition Solutions to Introduction to Pattern Recognition, Chapter 2: Bayesian pattern classification E395 - Pattern Recognton Solutons to Introducton to Pattern Recognton, Chapter : Bayesan pattern classfcaton Preface Ths document s a soluton manual for selected exercses from Introducton to Pattern Recognton

More information

Parametric Utilization Bounds for Fixed-Priority Multiprocessor Scheduling

Parametric Utilization Bounds for Fixed-Priority Multiprocessor Scheduling 2012 IEEE 26th Internatonal Parallel and Dstrbuted Processng Symposum Parametrc Utlzaton Bounds for Fxed-Prorty Multprocessor Schedulng Nan Guan 1,2, Martn Stgge 1, Wang Y 1,2 and Ge Yu 2 1 Uppsala Unversty,

More information

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan

Winter 2008 CS567 Stochastic Linear/Integer Programming Guest Lecturer: Xu, Huan Wnter 2008 CS567 Stochastc Lnear/Integer Programmng Guest Lecturer: Xu, Huan Class 2: More Modelng Examples 1 Capacty Expanson Capacty expanson models optmal choces of the tmng and levels of nvestments

More information

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg prnceton unv. F 17 cos 521: Advanced Algorthm Desgn Lecture 7: LP Dualty Lecturer: Matt Wenberg Scrbe: LP Dualty s an extremely useful tool for analyzng structural propertes of lnear programs. Whle there

More information

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers

Psychology 282 Lecture #24 Outline Regression Diagnostics: Outliers Psychology 282 Lecture #24 Outlne Regresson Dagnostcs: Outlers In an earler lecture we studed the statstcal assumptons underlyng the regresson model, ncludng the followng ponts: Formal statement of assumptons.

More information

Economics 101. Lecture 4 - Equilibrium and Efficiency

Economics 101. Lecture 4 - Equilibrium and Efficiency Economcs 0 Lecture 4 - Equlbrum and Effcency Intro As dscussed n the prevous lecture, we wll now move from an envronment where we looed at consumers mang decsons n solaton to analyzng economes full of

More information

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition

Single-Facility Scheduling over Long Time Horizons by Logic-based Benders Decomposition Sngle-Faclty Schedulng over Long Tme Horzons by Logc-based Benders Decomposton Elvn Coban and J. N. Hooker Tepper School of Busness, Carnege Mellon Unversty ecoban@andrew.cmu.edu, john@hooker.tepper.cmu.edu

More information

Perfect Competition and the Nash Bargaining Solution

Perfect Competition and the Nash Bargaining Solution Perfect Competton and the Nash Barganng Soluton Renhard John Department of Economcs Unversty of Bonn Adenauerallee 24-42 53113 Bonn, Germany emal: rohn@un-bonn.de May 2005 Abstract For a lnear exchange

More information

The Order Relation and Trace Inequalities for. Hermitian Operators

The Order Relation and Trace Inequalities for. Hermitian Operators Internatonal Mathematcal Forum, Vol 3, 08, no, 507-57 HIKARI Ltd, wwwm-hkarcom https://doorg/0988/mf088055 The Order Relaton and Trace Inequaltes for Hermtan Operators Y Huang School of Informaton Scence

More information

Difference Equations

Difference Equations Dfference Equatons c Jan Vrbk 1 Bascs Suppose a sequence of numbers, say a 0,a 1,a,a 3,... s defned by a certan general relatonshp between, say, three consecutve values of the sequence, e.g. a + +3a +1

More information

= z 20 z n. (k 20) + 4 z k = 4

= z 20 z n. (k 20) + 4 z k = 4 Problem Set #7 solutons 7.2.. (a Fnd the coeffcent of z k n (z + z 5 + z 6 + z 7 + 5, k 20. We use the known seres expanson ( n+l ( z l l z n below: (z + z 5 + z 6 + z 7 + 5 (z 5 ( + z + z 2 + z + 5 5

More information

Annexes. EC.1. Cycle-base move illustration. EC.2. Problem Instances

Annexes. EC.1. Cycle-base move illustration. EC.2. Problem Instances ec Annexes Ths Annex frst llustrates a cycle-based move n the dynamc-block generaton tabu search. It then dsplays the characterstcs of the nstance sets, followed by detaled results of the parametercalbraton

More information

Queueing Networks II Network Performance

Queueing Networks II Network Performance Queueng Networks II Network Performance Davd Tpper Assocate Professor Graduate Telecommuncatons and Networkng Program Unversty of Pttsburgh Sldes 6 Networks of Queues Many communcaton systems must be modeled

More information

Chapter Newton s Method

Chapter Newton s Method Chapter 9. Newton s Method After readng ths chapter, you should be able to:. Understand how Newton s method s dfferent from the Golden Secton Search method. Understand how Newton s method works 3. Solve

More information

Worst-case response time analysis of real-time tasks under fixed-priority scheduling with deferred preemption

Worst-case response time analysis of real-time tasks under fixed-priority scheduling with deferred preemption Real-Tme Syst (2009) 42: 63 119 DOI 10.1007/s11241-009-9071-z Worst-case response tme analyss of real-tme tasks under fxed-prorty schedulng wth deferred preempton Render J. Brl Johan J. Lukken Wm F.J.

More information

Lecture 14: Bandits with Budget Constraints

Lecture 14: Bandits with Budget Constraints IEOR 8100-001: Learnng and Optmzaton for Sequental Decson Makng 03/07/16 Lecture 14: andts wth udget Constrants Instructor: Shpra Agrawal Scrbed by: Zhpeng Lu 1 Problem defnton In the regular Mult-armed

More information

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module 3 LOSSY IMAGE COMPRESSION SYSTEMS Verson ECE IIT, Kharagpur Lesson 6 Theory of Quantzaton Verson ECE IIT, Kharagpur Instructonal Objectves At the end of ths lesson, the students should be able to:

More information

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium?

Welfare Properties of General Equilibrium. What can be said about optimality properties of resource allocation implied by general equilibrium? APPLIED WELFARE ECONOMICS AND POLICY ANALYSIS Welfare Propertes of General Equlbrum What can be sad about optmalty propertes of resource allocaton mpled by general equlbrum? Any crteron used to compare

More information

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009

College of Computer & Information Science Fall 2009 Northeastern University 20 October 2009 College of Computer & Informaton Scence Fall 2009 Northeastern Unversty 20 October 2009 CS7880: Algorthmc Power Tools Scrbe: Jan Wen and Laura Poplawsk Lecture Outlne: Prmal-dual schema Network Desgn:

More information

Mixed-Criticality Scheduling with I/O

Mixed-Criticality Scheduling with I/O Mxed-Crtcalty Schedulng wth I/O Erc Mssmer, Katherne Mssmer and Rchard West Computer Scence Department Boston Unversty Boston, USA Emal: mssmer,kzhao,rchwest}@cs.bu.edu Abstract Ths paper addresses the

More information

NP-Completeness : Proofs

NP-Completeness : Proofs NP-Completeness : Proofs Proof Methods A method to show a decson problem Π NP-complete s as follows. (1) Show Π NP. (2) Choose an NP-complete problem Π. (3) Show Π Π. A method to show an optmzaton problem

More information

Energy-Efficient Scheduling Fixed-Priority tasks with Preemption Thresholds on Variable Voltage Processors

Energy-Efficient Scheduling Fixed-Priority tasks with Preemption Thresholds on Variable Voltage Processors Energy-Effcent Schedulng Fxed-Prorty tasks wth Preempton Thresholds on Varable Voltage Processors XaoChuan He, Yan Ja Insttute of Network Technology and Informaton Securty School of Computer Scence Natonal

More information

Utilization-Based Scheduling of Flexible Mixed-Criticality Real-Time Tasks

Utilization-Based Scheduling of Flexible Mixed-Criticality Real-Time Tasks 1 Utlzaton-Based Schedulng of Flexble Mxed-Crtcalty Real-Tme Tasks Gang Chen, Nan Guan, D Lu, Qngqang He, Ka Huang, Todor Stefanov, Wang Y arxv:1711.00100v1 [cs.dc] 29 Sep 2017 Abstract Mxed-crtcalty models

More information

Schedulability Analysis of Task Sets with Upper- and Lower-Bound Temporal Constraints

Schedulability Analysis of Task Sets with Upper- and Lower-Bound Temporal Constraints Schedulablty Analyss of Task Sets wth Upper- and Lower-Bound Temporal Constrants The MIT Faculty has made ths artcle openly avalable. Please share how ths access benefts you. Your story matters. Ctaton

More information

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity Week3, Chapter 4 Moton n Two Dmensons Lecture Quz A partcle confned to moton along the x axs moves wth constant acceleraton from x =.0 m to x = 8.0 m durng a 1-s tme nterval. The velocty of the partcle

More information

4DVAR, according to the name, is a four-dimensional variational method.

4DVAR, according to the name, is a four-dimensional variational method. 4D-Varatonal Data Assmlaton (4D-Var) 4DVAR, accordng to the name, s a four-dmensonal varatonal method. 4D-Var s actually a drect generalzaton of 3D-Var to handle observatons that are dstrbuted n tme. The

More information

MMA and GCMMA two methods for nonlinear optimization

MMA and GCMMA two methods for nonlinear optimization MMA and GCMMA two methods for nonlnear optmzaton Krster Svanberg Optmzaton and Systems Theory, KTH, Stockholm, Sweden. krlle@math.kth.se Ths note descrbes the algorthms used n the author s 2007 mplementatons

More information

Resource-Efficient Scheduling Of Multiprocessor Mixed-Criticality Real-Time Systems

Resource-Efficient Scheduling Of Multiprocessor Mixed-Criticality Real-Time Systems Unversty of Pennsylvana ScholarlyCommons Publcly Accessble Penn Dssertatons 2017 Resource-Effcent Schedulng Of Multprocessor Mxed-Crtcalty Real-Tme Systems Jaewoo Lee Unversty of Pennsylvana, jaewoo@cs.upenn.edu

More information

A Robust Method for Calculating the Correlation Coefficient

A Robust Method for Calculating the Correlation Coefficient A Robust Method for Calculatng the Correlaton Coeffcent E.B. Nven and C. V. Deutsch Relatonshps between prmary and secondary data are frequently quantfed usng the correlaton coeffcent; however, the tradtonal

More information

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X

3.1 Expectation of Functions of Several Random Variables. )' be a k-dimensional discrete or continuous random vector, with joint PMF p (, E X E X1 E X Statstcs 1: Probablty Theory II 37 3 EPECTATION OF SEVERAL RANDOM VARIABLES As n Probablty Theory I, the nterest n most stuatons les not on the actual dstrbuton of a random vector, but rather on a number

More information

EDF Scheduling for Identical Multiprocessor Systems

EDF Scheduling for Identical Multiprocessor Systems EDF Schedulng for dentcal Multprocessor Systems Maro Bertogna Unversty of Modena, taly As Moore s law goes on Number of transstor/chp doubles every 18 to 24 mm heatng becomes a problem Power densty (W/cm

More information

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that

Online Appendix. t=1 (p t w)q t. Then the first order condition shows that Artcle forthcomng to ; manuscrpt no (Please, provde the manuscrpt number!) 1 Onlne Appendx Appendx E: Proofs Proof of Proposton 1 Frst we derve the equlbrum when the manufacturer does not vertcally ntegrate

More information

Resource Sharing. CSCE 990: Real-Time Systems. Steve Goddard. Resources & Resource Access Control (Chapter 8 of Liu)

Resource Sharing. CSCE 990: Real-Time Systems. Steve Goddard. Resources & Resource Access Control (Chapter 8 of Liu) CSCE 990: Real-Tme Systems Resource Sharng Steve Goddard goddard@cse.unl.edu http://www.cse.unl.edu/~goddard/courses/realtmesystems Resources & Resource Access Control (Chapter 8 of Lu) Real-Tme Systems

More information

The Geometry of Logit and Probit

The Geometry of Logit and Probit The Geometry of Logt and Probt Ths short note s meant as a supplement to Chapters and 3 of Spatal Models of Parlamentary Votng and the notaton and reference to fgures n the text below s to those two chapters.

More information

Feature Selection: Part 1

Feature Selection: Part 1 CSE 546: Machne Learnng Lecture 5 Feature Selecton: Part 1 Instructor: Sham Kakade 1 Regresson n the hgh dmensonal settng How do we learn when the number of features d s greater than the sample sze n?

More information

Solution Thermodynamics

Solution Thermodynamics Soluton hermodynamcs usng Wagner Notaton by Stanley. Howard Department of aterals and etallurgcal Engneerng South Dakota School of nes and echnology Rapd Cty, SD 57701 January 7, 001 Soluton hermodynamcs

More information

A FAST HEURISTIC FOR TASKS ASSIGNMENT IN MANYCORE SYSTEMS WITH VOLTAGE-FREQUENCY ISLANDS

A FAST HEURISTIC FOR TASKS ASSIGNMENT IN MANYCORE SYSTEMS WITH VOLTAGE-FREQUENCY ISLANDS Shervn Haamn A FAST HEURISTIC FOR TASKS ASSIGNMENT IN MANYCORE SYSTEMS WITH VOLTAGE-FREQUENCY ISLANDS INTRODUCTION Increasng computatons n applcatons has led to faster processng. o Use more cores n a chp

More information

O-line Temporary Tasks Assignment. Abstract. In this paper we consider the temporary tasks assignment

O-line Temporary Tasks Assignment. Abstract. In this paper we consider the temporary tasks assignment O-lne Temporary Tasks Assgnment Yoss Azar and Oded Regev Dept. of Computer Scence, Tel-Avv Unversty, Tel-Avv, 69978, Israel. azar@math.tau.ac.l??? Dept. of Computer Scence, Tel-Avv Unversty, Tel-Avv, 69978,

More information

Some Comments on Accelerating Convergence of Iterative Sequences Using Direct Inversion of the Iterative Subspace (DIIS)

Some Comments on Accelerating Convergence of Iterative Sequences Using Direct Inversion of the Iterative Subspace (DIIS) Some Comments on Acceleratng Convergence of Iteratve Sequences Usng Drect Inverson of the Iteratve Subspace (DIIS) C. Davd Sherrll School of Chemstry and Bochemstry Georga Insttute of Technology May 1998

More information

Minimisation of the Average Response Time in a Cluster of Servers

Minimisation of the Average Response Time in a Cluster of Servers Mnmsaton of the Average Response Tme n a Cluster of Servers Valery Naumov Abstract: In ths paper, we consder task assgnment problem n a cluster of servers. We show that optmal statc task assgnment s tantamount

More information

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Non-Pre-emptive Scheduling

Quantifying the Sub-optimality of Uniprocessor Fixed Priority Non-Pre-emptive Scheduling Quantfyng the Sub-optmalty of Unprocessor Fxed Prorty Non-Pre-emptve Schedulng Robert I Davs Real-Tme Systems Research Group, Department of Computer Scence, Unversty of York, York, UK robdavs@csyorkacuk

More information

(1 ) (1 ) 0 (1 ) (1 ) 0

(1 ) (1 ) 0 (1 ) (1 ) 0 Appendx A Appendx A contans proofs for resubmsson "Contractng Informaton Securty n the Presence of Double oral Hazard" Proof of Lemma 1: Assume that, to the contrary, BS efforts are achevable under a blateral

More information

Critical sections. Using semaphores. Using semaphores. Using semaphores. How long is blocking time? 17/10/2016. Problems caused by mutual exclusion

Critical sections. Using semaphores. Using semaphores. Using semaphores. How long is blocking time? 17/10/2016. Problems caused by mutual exclusion rtcal sectons Problems caused by mutual excluson crtcal secton wat(s) x = ; y = 5; sgnal(s) wrte global memory buffer nt x; nt y; read wat(s) a = x+; b = y+; c = x+y; crtcal secton sgnal(s) Usng semaphores

More information

Managing Capacity Through Reward Programs. on-line companion page. Byung-Do Kim Seoul National University College of Business Administration

Managing Capacity Through Reward Programs. on-line companion page. Byung-Do Kim Seoul National University College of Business Administration Managng Caacty Through eward Programs on-lne comanon age Byung-Do Km Seoul Natonal Unversty College of Busness Admnstraton Mengze Sh Unversty of Toronto otman School of Management Toronto ON M5S E6 Canada

More information

THE SUMMATION NOTATION Ʃ

THE SUMMATION NOTATION Ʃ Sngle Subscrpt otaton THE SUMMATIO OTATIO Ʃ Most of the calculatons we perform n statstcs are repettve operatons on lsts of numbers. For example, we compute the sum of a set of numbers, or the sum of the

More information

find (x): given element x, return the canonical element of the set containing x;

find (x): given element x, return the canonical element of the set containing x; COS 43 Sprng, 009 Dsjont Set Unon Problem: Mantan a collecton of dsjont sets. Two operatons: fnd the set contanng a gven element; unte two sets nto one (destructvely). Approach: Canoncal element method:

More information

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems Numercal Analyss by Dr. Anta Pal Assstant Professor Department of Mathematcs Natonal Insttute of Technology Durgapur Durgapur-713209 emal: anta.bue@gmal.com 1 . Chapter 5 Soluton of System of Lnear Equatons

More information

Quantifying the Exact Sub-Optimality of Non-Preemptive Scheduling

Quantifying the Exact Sub-Optimality of Non-Preemptive Scheduling Quantfyng the Exact Sub-Optmalty of Non-Preemptve Schedulng Robert I. Davs 1, Abhlash Thekklakattl 2, Olver Gettngs 1, Radu Dobrn 2, and Saskumar Punnekkat 2 1 Real-Tme Systems Research Group, Unversty

More information

An Upper Bound on SINR Threshold for Call Admission Control in Multiple-Class CDMA Systems with Imperfect Power-Control

An Upper Bound on SINR Threshold for Call Admission Control in Multiple-Class CDMA Systems with Imperfect Power-Control An Upper Bound on SINR Threshold for Call Admsson Control n Multple-Class CDMA Systems wth Imperfect ower-control Mahmoud El-Sayes MacDonald, Dettwler and Assocates td. (MDA) Toronto, Canada melsayes@hotmal.com

More information

Overhead-Aware Compositional Analysis of Real-Time Systems

Overhead-Aware Compositional Analysis of Real-Time Systems Overhead-Aware ompostonal Analyss of Real-Tme Systems Lnh T.X. Phan, Meng Xu, Jaewoo Lee, nsup Lee, Oleg Sokolsky PRESE enter Department of omputer and nformaton Scence Unversty of Pennsylvana ompostonal

More information

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA 4 Analyss of Varance (ANOVA) 5 ANOVA 51 Introducton ANOVA ANOVA s a way to estmate and test the means of multple populatons We wll start wth one-way ANOVA If the populatons ncluded n the study are selected

More information

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for P Charts. Dr. Wayne A. Taylor

Copyright 2017 by Taylor Enterprises, Inc., All Rights Reserved. Adjusted Control Limits for P Charts. Dr. Wayne A. Taylor Taylor Enterprses, Inc. Control Lmts for P Charts Copyrght 2017 by Taylor Enterprses, Inc., All Rghts Reserved. Control Lmts for P Charts Dr. Wayne A. Taylor Abstract: P charts are used for count data

More information

An Admission Control Algorithm in Cloud Computing Systems

An Admission Control Algorithm in Cloud Computing Systems An Admsson Control Algorthm n Cloud Computng Systems Authors: Frank Yeong-Sung Ln Department of Informaton Management Natonal Tawan Unversty Tape, Tawan, R.O.C. ysln@m.ntu.edu.tw Yngje Lan Management Scence

More information

Speeding up Computation of Scalar Multiplication in Elliptic Curve Cryptosystem

Speeding up Computation of Scalar Multiplication in Elliptic Curve Cryptosystem H.K. Pathak et. al. / (IJCSE) Internatonal Journal on Computer Scence and Engneerng Speedng up Computaton of Scalar Multplcaton n Ellptc Curve Cryptosystem H. K. Pathak Manju Sangh S.o.S n Computer scence

More information