Schedulability Bound of Weighted Round Robin Schedulers for Hard Real-Time Systems

Size: px
Start display at page:

Download "Schedulability Bound of Weighted Round Robin Schedulers for Hard Real-Time Systems"

Transcription

1 Schedulablty Bound of Weghted Round Robn Schedulers for Hard Real-Tme Systems Janja Wu, Jyh-Charn Lu, and We Zhao Department of Computer Scence, Texas A&M Unversty {janjaw, lu, Abstract We derve a parameterzed, closed-formed schedulablty bound for weghted round robn schedulers n hard real-tme computng systems. The schedulablty bound uses several parameters to represent mportant system behavors: (1) Number of tasks; (2) Normalzed deadlne that measures the tghtness of task deadlnes; (3) Tasks set workload burst-ness; (4) Overhead rato, whch measures the porton of overhead tme consumed n round robn operatons; (5) Normalzed token rotaton frequency, that measures the number of token rotatons n a tme nterval of the shortest relatve deadlne. Our work follows the network calculus representaton framework and s based on a generalzed workload and servce approxmaton models. We derve a closed form expresson of the utlzaton bound, so that one can analyze the mpact of partcular system parameters on the schedulablty bound and schedulablty bounds for specfc system confguratons can be easly obtaned by smply pluggng n proper parameters. 1

2 TABLE OF CONTENT 1. INTRODUCTION SYSTEM MODEL TASK, TASK SET, WORKLOAD, AND SERVICE FUNCTION TASK MODEL Workload Constrant Functon S-shaped Workload Constrant Functons Parameters of Task Set SCHEDULER MODEL Servce Constrant Functon Weghted Round Robn Scheduler Servce Constrant Functon of WRR Scheduler Parameters of Weght Round Robn Schedulers SCHEDULABILITY BOUND SCHEDULABILITY BOUND A LOWER BOUND OF SCHEDULABILITY BOUND SCHEDULABILITY BOUND OF WEIGHTED ROUND ROBIN SCHEDULERS PARAMETERIZED SCHEDULABILITY BOUND OF WRR SCHEDULER PARAMETER SENSITIVITY ANALYSIS OF THE WRR SCHEDULABILITY BOUND OPTIMAL TTRT SELECTION TO MAXIMIZE SCHEDULABILITY BOUND COMPARISON WITH EXISTING RESULTS COMPARISON WITH STATIC PRIORITY SCHEDULERS COMPARISON WITH TIMED TOKEN PROTOCOL FINAL REMARKS...28 REFERENCES...29 APPENDIX

3 1. INTRODUCTION Weghted round robn (WRR) schedulng dscplne has been mplemented n a broad range of msson crtcal computng and communcaton systems. In a WRR scheduler, tasks are performed n a cyclc order, n whch the tme a task can execute wthn each round s proportonal to the weght assgned to t. Weghted round robn schedulers have two major advantages: Ablty to mprove the system robustness, gven ther guarantee of a mnmum servce rate for each task. For weghted round robn schedulers, the maxmum amount of servce every task can receve n each round s upper-bounded by ts allocaton. As such, no task can consume more servce than what has been assgned. Ablty for mplementaton n a dstrbuted fashon. A WRR scheduler can be easly realzed n a dstrbuted envronment through a logc rng lke the tmed token protocol [2]. By passng a token along the rng, nodes can access resource n turn based on ther weght assgnments. It has been proven n [21] that wth proper weght assgnment, weghted round robn schedulers can provde deadlne guarantees for real-tme computng systems. Yet a major challenge s how to devse a low complexty schedulablty test, whch would guarantee the deadlne requrements, acheve hgh level of resource utlzaton, and s applcable to a broad range of system confguratons. Accordng to [22], a schedulablty test can be drect or ndrect. A drect schedulablty test explctly calculates the precse worst-case delay of each task n order to determne the permssblty of a new task. Despte ts accuracy, ths type of test has hgh run-tme cost n calculatng the worst case task delays, and therefore may not be sutable for on-lne admsson control. In contrast, an ndrect schedulablty test would use one or more system ndcators to determne the schedulablty of a new task wthout computng the worst case task delays drectly. The utlzaton based schedulablty test s the most common ndrect schedulablty test, n whch a new task can be admtted only f the total system utlzaton (as the system ndcator) s lower than a pre-determned bound. Utlzaton bound based schedulablty test s hghly desrable for large, 1

4 complex systems, because of ts extreme effcency and the ablty to provde a safe operaton margn by settng the system utlzaton bound to a value lower than the proven utlzaton bound [22]. Despte ts extreme smplcty n on-lne admsson control, lmted results have been obtaned on schedulablty bound of WRR schedulers. Most exstng work about WRR schedulers has focused on maxmzng the resource utlzaton through clever weght assgnments, wthout consderng the complexty and overhead of the schedulablty test algorthms. In many large-scale systems, t s desrable to trade some level of resource utlzaton for a smple and fast schedulablty test. To acheve ths goal, ths paper proposes to employ the general framework developed n [22] to analyze the schedulablty bound of the weghted round robn schedulers. We focus on a specal weght assgnment scheme n whch task weghts are assgned n proporton to ther resource demand rates. Ths weght assgnment scheme matches many resource allocaton requrements n practce,.e. assgnng larger weghts to users wth greater resource consumpton rates. We derve a closed-form expresson for the schedulablty bound of WRR schedulers wth normalzed weght assgnments. The schedulablty bound s parameterzed for number of tasks, the normalzed deadlne, whch measures the tghtness of deadlne assgnment, the tasks set workload burst-ness, the overhead rato, whch measures the porton of overhead tme consumed n round robn operatons per round, and the normalzed token rotaton frequency, whch measures the number of token rotatons n a tme nterval of the shortest relatve deadlne of the tasks. Unlke the relatvely developed knowledge body on the schedulablty bounds of statc prorty schedulers [1], [4], [5], [6], [7], [12], [15], [16], [17], [18], [19], [22], earlest deadlne frst schedulers [17], tmed token rng schedulers [2], [3], [11], [24], [25], [26], [27], [28], [29], [3], and ther varatons, no known bounds could be found for the weghted round robn schedulers n the lterature. Ths paper presents the frst systematc results of schedulablty bounds for weghted round robn schedulers, derved usng a formal modelng and optmzaton technque developed for general real-tme systems. Its closedform expresson allows one to easly analyze the mpact of system parameters over the schedulablty 2

5 bounds, and to effcently and effectvely fnd the schedulablty bounds for specfc system confguratons through smple substtuton of parameter values. The hghly versatle modelng and optmzaton method presented n ths paper can be talored for analyss of other types of real-tme systems. The rest of the paper s organzed as follows. Secton 2 ntroduces our general system model. Secton 3 dscusses the schedulablty bound analyss methodologes and derves a general bound result for arbtrary schedulers. In Secton 4, a closed-form schedulablty bound s derved and analyzed for WRR scheduler. Detaled comparsons of the newly derved bound wth the known results of other schedulers are provded n Secton 5, followed by conclusons n Secton SYSTEM MODEL The closed-form expresson of the schedulablty bound for WRR based real-tme schedulers s derved by applyng a constraned optmzaton technque to the workload-servce models proposed n [22]. To make ths paper self contaned, we ntroduce the generalzed system model n ths secton. The detals on dervaton of the schedulablty bound for WRR schedulers usng the general model wll be gven n the next secton. 2.1 Task, Task Set, Workload, and Servce Functon Ths paper consders sngle processor systems. We use Γ = { T1, T2,..., T n } to denote a task set, where T s the th task. When the context s clear, we may omt ndex n the subsequent dscussons. Each task s composed of a sequence of jobs. The worst-case executon tme of a job s called the job sze, whch s measured n second. A job can start ts executon after ts release tme, t r, and must be fnshed by ts absolute deadlne t d = t r + D where D s called relatve deadlne. For a job, the tme elapsed from the release tme t r to the completon tme t f s called the delay of the job, and the worst-case (.e., largest) 3

6 delay of all jobs n a task s denoted by d *. Wthn a task, the jobs have the same relatve deadlne, but may not necessarly be the same sze. Jobs wthn a task are executed n a frst come, frst served order. To characterze the resource demand of task T analytcally, we defne f() t, the workload functon for T, as follows, f ( t) = the summaton of the szes of all the jobs from T n [, t]. (2.1) Smlarly, to characterze the actual processor tme receved by task T, we defne g() t, the servce functon for T, as follows, gt ( ) = the total executon tme rendered to jobs of task Tdurng [, t]. (2.2) Based on the defntons of d *, f() t and g() t, the followng worst case delay formula can be easly derved [7]: ( ( τ τ )) * d t f t g t = sup nf ( ) ( + ). (2.3) From [8] and [22], we know that n (2.3) nf ( τ f( t) g( t τ )) and thus * d s the worst case delay of all the jobs from task T. + s the delay of the jobs arrvng at tme t A basc functon of the schedulablty test algorthm s to determne whether the followng nequalty holds: * d D. (2.4) One may want to use (2.3) to calculate d * and then compare the result wth D to test the schedulablty. However, ths method may not be sutable for onlne operaton because the exact forms of f () t and g() t may not be avalable when schedulablty test s made. Furthermore, even f f () t and g() t 4

7 are avalable, they are often too cumbersome to handle. A practcal soluton s usng some alternatve forms of f () t and g() t that can be obtaned durng schedulablty test. 2.2 Task Model Workload Constrant Functon Much work has been performed to fnd alternatves to f () t n order to model task workload for delay analyss. For example, a typcal alternatve s the workload constrant functon F( I ) ntroduced n [13] and [14] (under the name of workload curve). F( I ) s sad to be a workload constrant functon for task T f for any I t, f () t f( t I) F( I). (2.5) It s obvous that, gven any F that satsfes (2.5), F( I ) s an upper bound of total sze of jobs that can be released n any tme wndow [ t I, t]. We use I n (2.5) because F s defned on the doman of tme ntervals, whle f () t s defned on the doman of tme. Note that (2.5) defnes a group of functons, and for analyss convenence, we wll focus on those whch are non-decreasng and satsfy F () = S-shaped Workload Constrant Functons In ths paper, we consder a specal class of F, called s-shaped workload constrant functon. As ts name suggests, an s-shaped workload constrant functon conssts of segmented peces, and resembles a starcase. The values of an s-shaped workload constrant functon ncrease only at border ponts of segments. We assume that the segment length S s fxed and the ncrements may not be dentcal for the frst L segments where L s a parameter n the functon. 5

8 F() I L = 4 C C C 3 C 4 C 1 C 2 S 2S 3S 4S 5S I Fgure 1. An Example S-Shaped Workload Constrant Functon Formally, an s-shaped workload constrant functon can be expressed as follows: F( I) = h j C h L j = 1 L j C + ( a L) C h > L j = 1, (2.6) where h = I / S, j C s the ncrement at the begnnng of the j th segment, and C s the constant ncrement after the L th segment. Fgure 1 shows an example of the s-shaped workload constrant functon. When L = 1, an s-shaped constrant functon reduces to the classcal perodc task model F( I) = I / P C. In ths paper, we assume 1 2 L C C C C L. (2.7) Parameters of Task Set To characterze the task set n the sense of ts real-tme demand, we ntroduce several parameters. These parameters wll be used n the derved schedulablty result Workload Rate 6

9 Recall that the goal of ths paper s to derve a bound based schedulablty test, such as the utlzaton based method. To do so, we need to generalze the classcal utlzaton defned for perodc tasks to accommodate other non-perodc tasks usng the workload constrant functon defned above. Generally speakng, utlzaton s the resource consumpton rate n a measurng tme wndow. For perodc systems, task perod s typcally used as the length of the measurng wndow. Ths approach s not applcable to non-perodc tasks snce one may not have a well defned "perod". As such, n [1] and [4], the authors proposed to use the task relatve deadlne as the length of the measurng wndow. Whle ths choce s smple and convenent for some cases, we fnd that t s too restrctve for the desgn of a versatle utlzaton bound analyss system. To relax the constrants, we propose to defne the length of the measurng wndow as a lnear scale of the relatve deadlne. That s, the length of the measurng wndow wll be expressed as θ D, where θ > s called the scalng parameter and D s the relatve deadlne of the task. To avod confuson wth notatons from other lteratures, we refer to ths generalzed utlzaton as the scaled workload rate, and formally express t as follows: F( θ D) W ( θ ) =, (2.8) θ D and the task set workload rate as follows: n W( θ, Γ ) = W( θ ). (2.9) = 1 When the context of dscusson s clear, the term scaled may be omtted. Snce F ( θ D ) s an upper bound of the sze of jobs that can be released n any tme wndow of length θ, W ( θ, Γ ) can be treated as an upper bound of the job releasng rate averaged n a wndow of length D θ D. Introducng θ nto the modelng process parameterzes the utlzaton measurement. For example, when θ = 1, (2.9) reduces to 7

10 the defnton provded n [1] and [4]. Ths parameterzed measurement of utlzaton enables flexble representaton of dfferent schedulng and workload scenaros, and more mportantly, leads to a unform analyss system of schedulablty bounds Normalzed Deadlne To capture the tghtness of the task deadlne requrements of dfferent systems, we defne the normalzed deadlne k for T as follows: k D / S =, (2.1) where D s the relatve deadlne of task T and S s the segment length n the s-shaped workload constrant functon defned n (2.6). We follow the conventon that for = 1, 2,..., n k = k, (2.11) k can be vewed as the deadlne usng S as the measurement unt, and t characterzes tghtness of the deadlne requrements. The smaller the k, the more dffcult t s to schedule the task Burst-ness To characterze the burst-ness of dfferent tasks we ntroduce burst-ness parameter μ for T as: FS ( )/ S μ =, (2.12) F( k S)/( k S) and the burst-ness for the task set as: μ = max ( μ ). (2.13) = 1, 2,..., n By (2.7), one can notce that μ 1. (2.14) 8

11 μ s rato between the workload rate n a tme wndow S, and that n k S. Larger μ means more bursty workload. 2.3 Scheduler Model Recall that g defned n (2.2) characterzes the amount of servces a task may receve va a WRR scheduler. That s, g reflects the behavor of the scheduler. However, we cannot practcally use the form of g as defned n (2.2) whch was ponted out n Secton 2.1. Thus, n ths secton, we wll start modelng the scheduler by consderng the alternatves of g. We wll then formally defne the WRR scheduler and present an analytcal model for t Servce Constrant Functon A common alternatve to g() t s the generalzed servce constrant ntroduced n [8], [9], and [1] (under the name of servce curve). GI ( ) s sad to be a generalzed servce constrant functon f for any t, there exsts I t that preserves the property gt () f( t I) + GI ( ). (2.15) Typcally, we assume that GI ( ) s non-decreasng and G(). (2.15) means that for any t, we can fnd I, where I t, such that 1) all the jobs released n [, t I ] have been served, and 2) for jobs released n [ t I, t], at least G(I) amount of jobs have been served, as llustrated n Fgure 2. jobs released and served n ths nterval at least G(I ) of jobs released and served n ths nterval t - I t Fgure 2: Components n The Generalzed Servce Constrant Functon 9

12 Weghted Round Robn Scheduler Under the WRR schedulng dscplne, task servces are tme-multplexed n a cyclc, round robn fashon. A token s rotated n the cycle and a task can execute only f has the token. After recevng the token, task T can run for up to H tme unt where H s the tme allocated to a task. Typcally, H s calculated as ( ) H = O TTRT τ, (2-16) where O, O 1, s the weght of task T, and the TTRT s the target token rotaton tme, whch s the desred tme to complete one round of token rotaton, and τ s the tme overhead for token rotaton and other round robn operatons n each round, (e.g. the context swtchng cost n a sngle processor system or the propagaton delay n a dstrbuted system). Generally speakng, τ can be expressed as τ = n τ, (2-17) where n s the number of tasks n the system and τ s a overhead constant whch models the cost of context swtch or/and token propagaton delay. To use the WRR schedulng dscplne to schedule hard real-tme computng tasks, weghts O must be properly allocated. In ths paper, we wll show that an effectve weght assgnment scheme s based on the normalzed weght assgnments: O W (1) = W (1, Γ ), (2-18) where W(1) = F( D) / D s the workload of task T and W(1, Γ ) = W(1) s the workload rate of the task set defned n (2.9). Intutvely, the weghts of the tasks are assgned n proporton to ther n = 1 1

13 workload rates. Ths weght assgnment scheme matches many resource allocaton requrements n practce,.e. assgnng larger weghts to users havng greater resource consumpton rates Servce Constrant Functon of WRR Scheduler For the WRR scheduler, we have the followng result on ts servce constrant functon. Theorem 1. For WRR schedulers, a servce constrant functon for task T, s gven by I G( I) = n H. nτ + H j= 1 j (2-19) Proof: See Appendx. GI () I Fgure 3. An Example Servce Constrant Functon of Weghted Round Robn Scheduler We can make the followng observaton over (2-19): The servce constrant functons are of perodc shapes,.e. the values of the servce constrant n functon ncrease only at multples of the perod n + H 1 j at the amount of H for task τ j= T. Ths feature can be attrbuted to the fact that a WRR scheduler grants servces to each task T up to H amount of tme n each round, whose length s equal to + H. An example n τ j= 1 j servce constrant functon s gven n Fgure 3. 11

14 The value of the servce constrant functon s a decreasng functon of overhead constantτ. But an ncreasng functon of assgned weght H. Both are ntutve snce the lower the overhead and/or the bgger the weght, the more the servce tme for the task. When the overhead constant τ s zero and TTRT, the scheduler becomes the well-known Generalzed Process Sharng (GPS) system and the servce constrant functon reduces to G( I) = O I where O s the weght of task T Parameters of Weght Round Robn Schedulers To derve schedulablty bounds that can acheve hgh system resource utlzaton and s easly applcable to varous system settngs, t s mportant to select proper system parameters to capture the dynamcs of the system confguratons. We wll ntroduce two parameters,.e. overhead rato and normalzed token rotaton frequency for ths purpose. To measure the porton of tme consumed n round robn operatons relatve to the length of rotaton, we defne the overhead rato α as follows: τ TTRT α =, (2.2) where τ s the overhead constant defned n (2-17). To capture the effect of token rotaton speed on the schedulablty bound, we defne the second system parameter normalzed token rotaton frequency as follows: γ = D / TTRT mn, (2.21) where Dmn = mn( D ), (2.22) 12

15 and we assume that Dmn TTRT. (2.23) γ s the number of rounds the token rotates wthn a tme nterval of length D mn. The larger the γ, the faster the token rotates. 3 SCHEDULABILITY BOUND Wth workload and servce constrant functons defned n (2.5) and (2.15) as chosen alternatves of f () t and g() t for modelng task workload and scheduler servce, a general result on the schedulablty test usng F( I ) and GI ( ) s derved n [22], whch can be stated as n the followng theorem. Theorem 2: A task s schedulable f for any I FI ( ) GI ( + D). (3.1) where D s the relatve deadlne of the task. Though one can use (3.1) for each task to decde ts schedulablty test, t may be tme consumng, snce (3.1) needs to be checked for all I. An alternatve s the utlzaton based schedulablty test. 3.1 Schedulablty Bound For a gven system, we say that W ( θ ) s a schedulablty bound f an arbtrary task set Γ s schedulable when the followng condton holds: W( θ, ) W ( θ ) Γ <. (3.2) 13

16 The challenge s how to derve W * ( θ ) for a broad range of workload patterns and schedulng dscplnes. Let the space of all task sets be denoted as Ω,.e., Ω = {Γ}. Ω can be parttoned nto two subsets, Ω s and Ω ns, where Ω s = {Γ Γ s schedulable} (3.3) and Ω ns = {Γ Γ s not schedulable}. (3.4) W * ( θ ) s a lower bound of the workload rate of these task sets that belong to Ω ns. That s ( W θ ) W * ( θ) nf (, Γ ). (3.5) Γ Ω ns Fgure 4 llustrates ths concept. W ( θ, Γ) Ω ns W * ( θ ) Ω s Fgure 4. Illustraton of Schedulablty Bound 3.2 A Lower Bound of Schedulablty Bound In practce, t s often very dffcult to obtan an exact expresson of Ω ns. Instead, t may be desrable to use a substtuton of Ω ns n (3.5) for schedulablty analyss. Specfcally, we want to derve the schedulablty bound usng the followng nequty: 14

17 Γ Ω* ( W θ ) * W ( θ) nf (, Γ ). (3.6) In order to mantan the correctness of the schedulablty bound, Ω * must satsfy the followng constrant: * Ωns Ω. (3.7) That s, Ω * must contan all elements n Ω ns so that the derved bound wll be no hgher than that obtaned wth orgnal Ω ns. Fgure 5 llustrates ths concept. W * ( θ ) Ω ns W ( θ, Γ) * Ω Ω s Fgure 5. Relatonshp Between Ω ns and Ω *. If the Ω * s selected properly, the bound obtaned wth the substtuton could be close to the tght schedulablty bound obtaned wth Ω ns, f not the same. An nterestng queston s how to select Ω * that can work for dfferent type of tasks and schedulers. Clearly, we do not want the expresson of Ω * to depend on f and g, snce they are dffcult to obtan and handle. Instead, t s desrable to defne Ω * usng F and G. It presents an nterestng and challengng task to perform a full nvestgaton of the optons of Ω * and ther effects on the resultng schedulablty bound. Ths nvestgaton s yet to be seen. Nevertheless, we notce that the followng defnton works well for WRR schedulers: { T such that F( I) G ( I D) } * Ω = Γ Γ > +. (3.8) 15

18 That s, Ω * s the collecton of task set whch has at least one task not satsfyng (3.1). The followng Theorem proves that substtutng Ω ns wth Ω * s feasble. Theorem 3. Gven a collecton of task set Ω, a schedulablty bound wth scalng parameter θ s gven by: Γ Ω* ( W θ ) * W ( θ) = nf (, Γ ). (3.9) where * Ω and W ( θ, ) Γ are defned n (3.8) and (2.9) respectvely. Proof: By a close comparson of (3.9) and (3.5), we know that we just need to prove the followng: * Ωns Ω. (3.1) Assume (3.1) does not hold, then there must exst a non-schedulable task set Γ ' such that * Γ ' Ω. However, by (3.8) we know that for all, =1, 2,..., n, F( I) G ( I + D). By Theorem 2, we know that Γ ' s schedulable. Ths contradcts the fact that Γ ' s non-schedulable. Then the Theorem follows. Q.E.D. In the followng secton, we wll derve W for weghted round robn schedulers usng Theorem 3. 4 SCHEDULABILITY BOUND OF WEIGHTED ROUND ROBIN SCHEDULERS In ths secton, we wll analyze the schedulablty bound of WRR scheduler usng the general system model and methodology ntroduced n the prevous sectons. 16

19 4.1 Parameterzed Schedulablty Bound of WRR Scheduler Based on the task parameters defned n (2.1) and (2.13), the scheduler parameters defned n (2.2) and (2.21), we derve our schedulablty bound by substtutng the servce constrant functons of WRR scheduler derved n (2-19) nto Theorem 3 and solvng the resultng optmzaton problem. The result s stated formally n the next Theorem Theorem 4. A lower bound of schedulablty bound wth scalng parameter round robn scheduler wth normalzed weght assgnment, and s-shaped tasks s gven by θ = k / k for weghted W * ( k / k) = 1 ( 1 nα ) 1 mn( 1, k ) 1/ γ + 1 μ. (4-1) Proof: See Appendx. Q.E.D. (4-1) s a parameterzed bound result. By substtutng specfc values of these parameters nto (4-1), one can obtan the dfferent schedulablty bound results. To llustrate how the bound result can be used n practcal systems, we ntroduce the followng two examples. Example 1: Consder a smple real-tme robot controller who s responsble for the routng control of three data samplng robots, A, B, and C. The three robots are drvng at dfferent speeds and the controller communcates wth the robot at a dfferent frequency,.e. once every 2., 5., and 2. seconds, for A, B, and C, respectvely. For the three robots, the controller takes.25,.3, and.65 seconds to fnsh the route selecton and communcaton. The route selecton and communcaton must be fnshed wthn 4., 3., and 2. seconds for A, B, and C to avod robot damages. The controller uses a WRR schedulng dscplne wth target token rotaton tme of 2. seconds per round. There s a cost of.2 seconds per context swtchng (changng the robot to be served). The weghts for robots are assgned n normalzed fashon usng (2-18). Now we need to decde whether the controller can fnsh all the routng tasks wthn ther deadlnes. 17

20 From the above descrpton, we know there are three perodc tasks { T, T, T } Γ= where F( I) = I / P C (4-2) and C =.25 P = 4. D = C =.3 P = 5. D = C =.65 P = 2. D = (4-3) For ths set of tasks, by (2.2), (2.21), (2.1), and (2.12), we have α =.1, γ = 2, k = 1, and μ = 1. To decde whether the task set s schedulable or not, we must calculate the total system workload rate as follows: W F( D) C C C (4-4) D P P P (1, Γ ) = = + + = + + =.155. = By substtutng n = 3, α =.1, γ = 2, k = 1, and μ = 1 nto (4-1), we have a schedulablty bound of.66. Snce.155<.66, we conclude that task set s schedulable. Example 2: Now we consder a more complex scenaro of the above robot controller. The controller needs to send routng commands to robot A, B, C every 4., 5., and 2. seconds. The frst tme the controller communcates wth the robots takes longer and requres 1., 1.2, 1.3 seconds for A, B, and C, respectvely. After the frst tme, the tme reduces to.5,.6, and.65 seconds. Ths tme, the routng and communcaton for robot A, B, C must be fnshed wthn 8., 1., and 4. seconds (ncludng the frst tme). Agan, the controller uses a WRR dscplne wth normalzed weght assgnments and TTRT=2, τ =.2. We need to decde whether the controller can serve the three robots whle guaranteeng ther deadlnes. 18

21 From the above descrpton, we know that we can model the communcaton and routng selecton wth three s-shaped tasks: { T, T, T } Γ= where C I P F ( I) = 1 C + ( I / P 1 ) C I > P, (4.5) and C = 1. C =.5 P = 4. D = C = 1.2 C =.6 P = 5. D = 1. C = 1.3 C =.65 P = 2. D = 4.. (4-6) To decde whether the task set s schedulable or not, we should calculate the total system workload rate as follows: W F( D) C C C (4-7) D D D D (1, Γ ) = = + + = + + =.17. = For ths set of tasks, by (2.2), (2.21), (2.1), and (2.12), we have α =.1, γ = 4, k = 2, and μ = 1.5 By substtutng n = 3, α =.1, γ = 4, k = 2, and μ = 1.5 nto (4-1), we obtan a schedulablty bound of.53. Snce.53 >.17, we conclude that the task set s schedulable. The above two examples llustrate how to use the newly derved schedulablty bound for schedulablty analyss. In the next secton, we wll dscuss n detal how the dfferent parameters affect the schedulablty bound as well as a comparson wth the well known schedulablty bound of the rate monotonc scheduler. 19

22 4.2 Parameter Senstvty Analyss of the WRR Schedulablty Bound By (4-1), we can make the followng observatons on the senstvty of the schedulablty bound on the dfferent system parameters: For gven values of n, k, μ, and α, the schedulablty bound ncreases wth the token rotaton frequency. Thus, the faster the token rotates, the better chance a task set can be scheduled. Fgure 6 llustrates ths trend for the case of α =. For gven values of n, k, μ, and α, the schedulablty bound decreases wth the task set burst-ness μ. Increasng of μ mples a more bursty workload, whch tends to be more dffcult to schedule than less bursty ones. The schedulablty bound s maxmzed when μ =1, whch corresponds to perodc tasks. For gven values of n, k, γ, and μ, the schedulablty bound ncreases when overhead rato α decreases. The schedulablty bound s maxmzed when α =, whch s an deal case and means that there s no overhead n round robn operatons. For gven values of k, γ, μ, and α, the schedulablty bound decreases when the number of tasks n the system ncrease and approaches zero when n 1/ α whch means all processor tme wll be devoted to round robn operatons. For perodc tasks wth relatve deadlnes equal to perods,.e., D =P, and token rotaton frequency γ = 2, (token rotates twce per D mn nterval), the schedulablty bound s 67%. When the normalzed deadlne ncreases from 1. to 1., the schedulablty bound remans at 67%. Ths s llustrated as a seral of ponts n Fgure 7. Note that ths does not mply that k has no effect on the schedulablty test, snce the workload rate s measured dfferently. For perodc tasks wth relatve deadlnes equal to ther perod (D =P ), and token rotaton frequency γ = 1, (token rotates at least once per D mn nterval), the schedulablty bound s 5%. Ths s hghlghted on Fgure 8 as a pont. 2

23 Fgure 6. Schedulablty Bound Wth k = 1 Fgure 7. Schedulablty Bound Wth γ = 2 21

24 Schedulablty Bound k 1 μ = 1, α = % γ 1 2 Fgure 8. Schedulablty Bound Wth μ = Optmal TTRT Selecton to Maxmze Schedulablty Bound In a WRR system, for a gven task set, parameters n, k and μ are fxed. In order acheve a hgher schedulablty bound, one can adjust TTRT. By (2.2) and (2.21), we know that settng larger TTRT may reduce operaton overhead rato but at the same tme leadng to lower rotaton frequency. Thus, TTRT should be adjusted n a way that balances the operaton overhead rato and the token rotaton frequency. The followng theorem gves the value of TTRT that maxmzes the schedulablty bound. Theorem 4. The schedulablty bound of a WRR scheduler, for a gven set of tasks, s maxmzed when the value of TTRT s selected as follows: TTRT = D / γ *, (4-8) mn where 22

25 1 f nτ / Dmn 1/3 f nτ / Dmn < 1/ 3 and γ* = Dmn / nτ 1 + 1, (4-9) Z ( Dmn / nτ 1 1 ) <Z ( Dmn / nτ ) Dmn / nτ otherwse and 1 nτ x. (4-1) 1/ x+ 1 D ( ) = 1 Z x mn When the TTRT takes the optmal value, the schedulablty bound s gven by 1 nτ 1 W * ( k / k) = 1 γ * mn( 1, k ). (4-11) 1/ γ * + 1 D μ mn Proof: See Appendx. Q.E.D. By a careful observaton of Theorem 4, we notce that the optmal TTRT value that maxmzes the schedulablty bound only depends on rato nτ / Dmn. Ths rato s the percentage of tme consumed n round robn operatons wth a wndow of length D mn. When rato n Dmn τ / 1/3, the optmal TTRT equals to D mn and when n Dmn τ / < 1/3, the optmal TTRT value would be ether D Dmn / nτ mn or D Dmn / nτ mn. Fgure 9 plots the trend of the maxmum schedulablty bound of WRR wth the optmal TTRT selecton for the case of k = 1 and μ = 1. It s clear from the Fgure 9 that, when the optmal value of TTRT s used, the schedulablty bound of WRR s a monotonc decreasng functon of nτ / Dmn. The 23

26 rate of decrease s low for small values of nτ / Dmn, say, less than.1, and the rate gradually ncreases nτ / D becomes larger. When nτ / Dmn approaches 1., the schedulablty bound s very close when mn to %. Schedulablty Bound nτ /D mn Fgure 9. Schedulablty Bound wth the Optmal TTRT Selecton 5. COMPARISON WITH EXISTING RESULTS We now compare the newly derved schedulablty bound wth the results of statc prorty scheduler and the tmed token rng scheduler. 5.1 Comparson wth Statc Prorty Schedulers In ths secton, we wll compare the newly derved schedulablty bound wth those of the statc prorty scheduler. For the sake of smplcty, we focus on perodc tasks and assume that overhead rato α of the weghted round robn schedule s. 24

27 a). Consder a set of perodc tasks wth deadlnes equal to ther perods. A schedulablty bound for the rate monotonc scheduler s 69% [17]. For the same task system and WRR scheduler, we know that μ = 1 and k = 1. By substtutng these parameters nto (4-1), we have a schedulablty bound of γ /( γ + 1). The curve of ths functon n plotted n Fgure 1. As can be seen from Fgure 1, the bound of the weghted round robn scheduler s lower than the bound of statc prorty schedulers when γ < When γ = 2.26, the weghted sound robn scheduler acheves the same 69% bound. When γ > 2.26, WRR out-performs the rate monotonc scheduler n term of hgher schedulablty bound. Ths phenomenon can be explaned by the nature of the two types of schedulers. Statc prorty scheduler renders servce to a job only f no hgher prorty job s watng to be executed, even though completng the lower prorty job frst may avod mssng a deadlne. In other words, statc prorty scheduler may allocate more servce tme to hgh prorty tasks than what s needed to guarantee the deadlnes, whle WRR scheduler assgns task servce tme n proportonal to ts workload rate whch avods overallocaton of servce tme to certan tasks. As such, WRR scheduler can out-perform statc prorty scheduler under certan stuaton. b). For a set of perodc tasks wth deadlnes beng half as long as ther perods, a schedulablty bound for the rate monotonc scheduler s 5% [16]. For the same task system wth the WRR scheduler, we know that μ = 1, and k=1/2. By substtutng these parameters nto (4-1), we have a bound of γ /( 2( γ + 1) ). Snce ( ) γ / 2( γ + 1) 1/2, we know that the rate monotonc scheduler out-performs the WRR scheduler n ths case. However, when we ncrease token rotaton frequency γ, the schedulablty bound of WRR scheduler s approachng the 5% bound and attans t when γ. c). For a set of perodc tasks wth deadlnes beng twce as long as ther perods, a schedulablty bound for the rate monotonc scheduler s 81% [16]. For the same task system wth the WRR 25

28 scheduler, we know that μ = 1, and k=2. By substtutng these parameters nto (4-1), we have a bound of γ /( γ + 1). It s easy to see that when γ = 4.32, the WRR scheduler acheves the same bound as the rate monotonc scheduler and out-performs t when γ further ncreases. 1.9 Schedulablty Bound % γ Fgure 1. Schedulablty Bound Wth μ = 1 Based on the above analyss, we know that for the perodc tasks, weghted round robn scheduler can acheve the same or even hgher schedulablty bound than the rate monotonc schedulers. We should note that the above analyss focuses on a very smple case n whch tasks are of a perodc shape and token rotaton overhead s neglgble. However, smlar trends stll hold for more complex cases. The fact that weghted round robn schedulers can solate ll-behaved tasks and can acheve the same or hgher schedulablty bound than the well-known rate monotonc scheduler makes ths type of scheduler applcable to practcal real-tme systems. 26

29 5.2 Comparson wth Tmed Token Protocol A close related varant of the weghted round robn scheduler s the tmed token rng scheduler used n FDDI networks [2], [3], [24], [25], [26], [27], [28], [29], [3]. In a typcal FDDI tmed token rng system, there are n communcaton nodes connected nto a rng. Each node has two types of packets: real-tme and non real-tme. Real-tme packets have hard deadlnes, e.g. packets must be sent before ther deadlnes, whle non real-tme packets do not have deadlne requrements. Smlar to the weghted round robn scheduler, a token s rotated among the nodes n the system and the desred tme to fnsh one round of token rotaton s denoted as TTRT. Upon recevng the token, a node wll frst send ts real-tme packets up to the allocated unts of tme. Each node also keeps track the last token rotaton tme, denoted by TTR. If the token arrves earler n the last round,.e. TTR < TTRT, then t wll send ts non real-tme packets up to the TTRT-TRT amount of tme. The ratonal behnd ths s to "steal" the unused tme slots n the last token rotaton. Due to the nterference of the non real-tme packets, the servce avalable to each node may be less, compared wth what s provded by a weghted round robn scheduler. In turn, the schedulablty bound wll be lower. It has been proven n [3] that for perodc tasks wth relatve deadlne equals to ther perods and normalzed weght assgnment scheme wth token rotaton at lease twce for any nterval of length of mnmum task perods, the tmed token rng scheduler has a schedulablty bound of (1 nα ) / 3 1. For ths system, by (2.21), (2.1), and (2.12), we know that k=1, μ =1, and γ =2. By substtutng them nto (4-1), we know that a schedulablty bound of the weghted round robn scheduler s 2(1 nα ) / 3, twce of the tmed token rng. Fgure 11 llustrates ths dfference for the case of n=1. 1 In [3], the authors assumes the scheduler spend a const amount of tme n round robn operatons and derved a bound of (1 α ) / 3 where α = τ / TTRT. Ths bound converts to (1 nα ) / 3 n our new defnton of α = τ / TTRT snce τ = nτ. 27

30 Schedulablty Bound Overhead Rato α Fgure 11. Schedulablty Bound Comparson Between WRR and Tmed Token Rng Scheduler 6. FINAL REMARKS In summary, n ths paper, we derve a closed-form expresson for the schedulablty bound of WRR schedulers wth normalzed weght assgnment scheme for s-shaped tasks. The schedulablty bound s parameterzed for round robn overhead rato α, normalzed token rotaton rato γ, the normalzed deadlne k, and the task set workload burst-ness μ. From the general result, one can easly obtan schedulablty bounds for specfc system confguraton by smple plug-n of proper parameters. Our work reported here s the frst that systematcally derves the schedulablty bound for WRR systems. Our results are general and can be appled to a wde range of systems. Nevertheless, these results are prelmnary and can be easly extended to other type of task workloads and weght assgnment scheme. 28

31 REFERENCES [1] T. Abdelzaher and C. Lu, Schedulablty analyss and utlzaton bounds for hghly scalable real-tme servces, Proc. 7th Real-Tme Technology and Applcatons Symposum, Tape, Tawan, 21, pp [2] G. Agrawal, B. Chen, and W. Zhao, Local synchronous capacty allocaton schemes for guaranteeng message deadlnes wth the tmed token protocol, n Proc. IEEE INFOCOM'93, San Francsco, CA, Mar. 1993, pp [3] G. Agrawal, B. Chen, W. Zhao, and S. Davar, Guaranteeng synchronous message deadlnes wth the tmed token protocol, IEEE Trans. Computers, vol. 43, no. 3, pp , Mar [4] B. Andersson, Statc-prorty schedulng on multprocessors, Ph.D. dssertaton, Dept. Computer Engneerng., Chalmers Unv. of Technology, Göteborg, Sweden, 23. [5] T. P. Baker, Multprocessor EDF and deadlne monotonc schedulablty analyss, Proc. 24th IEEE Int. Real-Tme Systems Symposum, Cancun, Mexco, 23, pp [6] E. Bn, Schedulablty analyss of perodc fxed prorty systems, IEEE Trans. Computers, vol. 53, no.11, pp , Nov. 24. [7] E. Bn, G. C. Buttazzo, and G. Buttazzo, Rate monotonc analyss: the hyperbolc bound, IEEE Trans. Computers, vol 52, no. 7, pp , Jul. 23. [8] J. Y. Le Boudec and P. Thran, Network Calculus, a Theory of Determnstc Queung Systems for the Internet, New York: Sprnger-Verlag, 21. [9] S. Chang, On determnstc traffc regulaton and servce guarantee: a systematc approach by flterng, IEEE Trans. Inform. Theory, vol. 44, pp , Aug [1] S. Chang, Performance Guarantees n Communcaton Networks, New York: Sprnger-Verlag, 2. [11] B. Chen, G. Agrawal, and W. Zhao, Optmal synchronous capacty allocaton for hard real-tme communcatons wth the tmed token protocol, Proc. IEEE Real-Tme Systems Symposum (RTSS'92), Phoenx, AZ, Dec. 1992, pp [12] D. Chen, A. K. Mok, and T.-W. Kuo, Utlzaton bound revsted, IEEE Trans. Computers, vol. 52, no. 3, pp , Mar. 23. [13] R. L. Cruz, A calculus for network delay, part I: network elements n solaton, IEEE Transactons on Informaton Theory, 37(1), pp , Jan [14] R. L. Cruz, A calculus for network delay, part II: network elements n solaton, IEEE Transactons on Informaton Theory, 37(1), pp , Jan [15] J. P. Lehoczky, Fxed prorty schedulng of perodc task sets wth arbtrary deadlnes, n Proc. 11th IEEE Real Tme Systems Symposum, Pscataway, NJ, 199, pp [16] J. P. Lehoczky and L. Sha, Performance of real-tme bus schedulng algorthms, ACM SIGMETRICS Performance Evaluaton Revew, vol. 14, no. 1, May 1986, pp [17] C. L. Lu and J. W. Layland, Schedulng algorthms for multprogrammng n a hard-real-tme envronment, J. ACM, vol. 2, no. 1, pp , Jan [18] X. Lu and T. Abdelzaher, "On Non-Utlzaton Bounds for Arbtrary Fxed Prorty Polces," Proc. 12th IEEE Real-Tme and Embedded Technology and Applcatons Symposum (RTAS'6), CA, Apr. 26, pp [19] D. Oh and T. P. Bakker, Utlzaton bounds for n-processor rate monotone schedulng wth statc processor assgnment, Real Tme Systems J., vol. 15, no. 1, pp , Nov [2] A. Parekh and R. G. Gallager, A generalzed processor sharng approach to flow control n ntegrated servces networks: the sngle node case, IEEE/ACM Trans. Networkng, vol. 1, no. 3, pp , Jun [21] A. Raha, N. Malcolm, We Zhao, "Hard real-tme communcatons wth weghted round robn servce n ATM local area networks," ceccs, Frst IEEE Internatonal Conference on Engneerng of Complex Computer Systems (ICECCS'95), pp , Aug [22] J. Wu, J.-C. Lu and W. Zhao, On schedulablty bounds of statc prorty schedulers, Proc. 11th IEEE Real-Tme and Embedded Technology and Applcatons Symposum, San Francsco, CA, Mar, 25, pp [23] D. Xuan, C. L, R. Bettat, J. Chen, and W. Zhao, Utlzaton-based admsson control for real-tme applcatons, Proc. 2 Int. Conf. Parallel Processng, Toronto, Canada, 2, pp [24] S. Zhang and A. Burns, An optmal synchronous bandwdth allocaton scheme for guaranteeng synchronous message deadlnes wth the tmed-token MAC protocol, IEEE/ACM Trans. Networkng, vol. 3, no. 6, pp , [25] S. Zhang and A. Burns, Tmng propertes of the tmed token MAC protocol, Proc. 6th Internatonal Conference on Computer Communcatons and Networks, Sep, 1997, pp [26] S. Zhang and A. Burns, and T.-H. Cheng, Cycle-tme propertes of the tmed token medum access control protocol, IEEE Transactons on Computers, vol.51, no.11, pp , Nov. 22. [27] S. Zhang, A. Burns, J. Chen, and E. S. Lee, Hard real-tme communcaton wth the tmed token protocol: current state and challengng problems, Real-Tme System, vol. 27, no. 3, Sep. 24, pp [28] S. Zhang, A. Burns, A. Mehaoua, E. S. Lee, and H. Yang, Testng the schedulablty of synchronous traffc for the tmed token medum access control protocol,, Real-Tme Systems, vol.22, no.3, pp , May

32 [29] S. Zhang and E. S. Lee, Effcent global allocaton of synchronous bandwdths for hard real-tme communcaton wth the tmed token MAC protocol, Proc. 6th Int. Conf. Real-Tme Computng Systems and Applcatons, Hong Kong, Chna, 1999, pp [3] Q. Zheng and K. G. Shn, Synchronous bandwdth allocaton n FDDI networks, IEEE Transactons on Parallel and Dstrbuted Systems, vol.6 no.12, pp , Dec

33 APPENDIX Theorem 3. For WRR schedulers, a servce constrant functon for task T, s I G( I) = n H. τ + H j= 1 j (A-1) Proof. We prove ths theorem based on the defnton of servce constrant functon. Let t be an arbtrary tme nstant. If at tme t, all the jobs from task T have been served, then we can let s =t and (2.15) s true. Now we focus the case that at tme t, task T has backlog. Let s be the last tme before t such that T dd not backlog. That s to say, at tme s, we have g () s = f () s. (A-2) n In tme nterval [ s, t ], the scheduler served at least I / ( τ + H j= 1 j ) length of H each round. In other words, task servce. Formally, we have rounds wth a servng tme n T receved at least I / ( τ + H j= 1 j ) H j amount of I g() t g() s n H. τ + H j= 1 j (A-3) By substtutng (A-2) nto(a-3), we have I g() t f() s n H. τ + H j= 1 j (A-4) 31

34 By comparng (A-4) wth (2.15), we know that the theorem follows. Q.E.D. Theorem 4. A lower bound of schedulablty bound wth scalng parameter round robn scheduler wth normalzed weght assgnment, and s-shaped tasks s gven by θ = k / k for weghted k = ( α ) ( ) * 1 1 W ( / k) 1 n mn 1, k 1/ γ + 1 μ. (A-5) Proof. By Theorem 3, we known that a servce constrant functon provded by task T by a WRR scheduler s: I G( I) = n H. τ + H j= 1 j (A-6) For normalzed weght assgnment, by (2-18) and (2-16), we have H W (1) = ( TTRT τ ) W (1, Γ) (A-7) By substtutng (A-7) nto (A-6), we have the servce constrant functon as I W (1) G ( I) = ( TTRT τ ). TTRT W (1, Γ) (A-8) By Theorem 2, we know that a schedulablty bound for WRR wth scalng parameter θ = k / k s ( ) * W ( k / k) = nf W( k / k, Γ ). (A-9) Γ Ω* where 32

35 { T such that F( I) G ( I D) } * Ω = Γ Γ > +. (A-1) By substtutng (A-8) nto (A-1) and rearrange the resultng equaton, we know that a schedulablty bound s ( ) * W ( k / k) = nf W( k / k, Γ ). (A-11) Γ Ω* where * I + D (1) W Ω = Γ T Γ such that W(1, Γ ) > ( TTRT τ ) TTRT. (A-12) F ( I) * Γ Ω, let us defne I + D TTRT τ Z () = mn I W(1) TTRT. (A-13) F ( I ) Then by (A-12), we have = 1, 2,..., n ( Z ) W(1, Γ ) > mn ( ). (A-14) By substtutng (2.2) and (2.2) nto (A-13), we have I + D I D F( D) Z () ( 1 n ) mn TTRT + = α I. (A-15) I + D F( I ) D TTRT I + D I + D Snce 1 TTRT + TTRT, we have 33

36 I + D ( ) () ( 1 ) mn TTRT I + D F D Z nα I. (A-16) I + D ( ) + 1 F I D TTRT It s easy to verfy that I + D TTRT = =. (A-17) I + D / γ TTRT I + D Dmn TTRT TTRT where have D mn and γ are defned n (2.22) and (2.21), respectvely. By substtutng (A-17) nto (A-16), we 1 I + D F( D) Z () ( 1 nα ) mni 1 1/ γ F( I ) D +. (A-18) Now let I = ms + ω where ω < S, S s the segment length of the s-shaped workload constrant functon defned n (2.6), and m s a non-negatve nteger. By (2.6), we have F( I) F(( m+ 1) S ). (A-19) By substtutng (A-19) nto (A-18) and rearrangng t, we have 1 ms + ω + D F( D) Z () ( 1 nα ) mni 1/ γ 1 F( ( m 1) S) D. (A-2) + + Snce ω, we have 34

37 1 ms + D F( D) Z () ( 1 nα ) mni 1/ γ 1 F( ( m 1) S) D. (A-21) + + By (2.6) and (2.7), t can be verfed that (( + 1) ) ( ) F m S F S ( m+ 1) S S. (A-22) By substtutng (A-22) nto (A-21) and rearrangng t, we get 1 ms + D F( D) Z () ( 1 nα ) mni 1/ γ 1 ( m 1) F( S) D. (A-23) + + By defnton of s-shaped workload constrant functon n (2.6) and k n (2.11), we know that F( D) = F( ks ) = F( k S ). (A-24) By substtutng (A-24) nto (A-23), we have 1 ms + D F( k S) Z () ( 1 nα ) mni 1/ γ 1 ( m 1) F( S) ks. (A-25) + + By substtutng (2.1) nto (A-25) and rearrangng the resultng nequalty, we have 1 m+ k F( k S) Z () ( 1 nα ) mni 1/ γ 1 ( m+ 1) F( S) k. (A-26) + Rewrte (A-26) nto 1 1 F( k S) m+ k Z () ( 1 nα ) mn m=, 1, 2,... 1/ γ + 1 k F S m + 1. (A-27) ( ) 35

38 It can be verfed that m+ k mn m=, 1, 2,... mn 1, m + 1 ( k ). (A-28) By substtutng (A-28) nto (A-27), we have 1 1 F( k S) Z () ( 1 nα ) mn 1, k k 1/ γ + 1 F S ( ) ( ). (A-29) By (2.6) and (2.7), we have F( k S) k k =. (A-3) F( S ) μ μ By substtutng (A-3) nto (A-29), we get 1 k Z () ( 1 nα ) mn 1, k 1/ γ + 1 kμ ( ). (A-31) By substtutng (A-31) nto (A-14), we have 1 k W(1, Γ) ( 1 nα ) mn 1, k 1/ γ + 1 kμ ( ). (A-32) By (2.6), we know that n F( D) k n F( k S) k n k k k W ( 1, Γ) = = = W, W, = 1 = 1 = 1 Γ = Γ D k k S k k k k.(a-33) By substtutng (A-33) nto (A-32) and rearrange the resultng nequalty, we have 36

39 W k / k, Γ 1 nα 1 1 mn 1, 1/ γ + 1 μ ( ) ( ) ( k ). (A-34) Snce (A-34) s true * Γ Ω, we know that 1 1 nf * ( W( k / k, Γ )) ( 1 nα ) mn ( 1, k ) Γ Ω 1/ γ + 1 μ. (A-35) Then by substtutng (A-35) nto (A-9), we get * 1 1 W ( / k) 1 n mn 1, k 1/ γ + 1 μ k ( α ) ( ). (A-36) Comparng (A-36) wth (A-5), we have the theorem proven. Q.E.D. When the TTRT takes the optmal value, the schedulablty bound s gven by 1 nτ 1 W * ( k / k) = 1 γ * mn( 1, k ). (1-37) 1/ γ * + 1 D μ mn Theorem 4. The schedulablty bound of a WRR scheduler, for a gven set of tasks, s maxmzed when the value of TTRT s selected as follows: TTRT = D / γ *, (A-38) mn where 37

40 1 f nτ / Dmn 1/3 f nτ / Dmn < 1/ 3 and γ* = Dmn / nτ 1 + 1, (A-39) Z ( Dmn / nτ 1 1 ) <Z ( Dmn / nτ ) Dmn / nτ otherwse and 1 nτ x. (A-4) 1/ x+ 1 D ( ) = 1 Z x mn When the TTRT takes the optmal value, the schedulablty bound s gven by 1 nτ 1 W * ( k / k) = 1 γ * mn( 1, k ). (A-41) 1/ γ * + 1 D μ mn Proof: By (4-1), we know that a schedulablty bound s W * ( k / k) = 1 ( 1 nα ) 1 mn( 1, k ) 1/ γ + 1 μ. (A-42) Let us defne ω = D γ TTRT. (A-43) mn where γ s defned n (2.21). By (2.21), we have ω. (A-44) By substtutng (A-43) and (2.21) nto (A-42), we have 38

41 * 1 nτ 1 W ( k / k) = 1 mn 1, k 1/ γ + 1 ( Dmn ω) / γ μ ( ). (A-45) It s easy to see that (A-45) wll be maxmzed when ω = and the maxmum wll be 1 nτ 1 W * ( k / k) = 1 γ mn( 1, k ). (A-46) 1/ γ + 1 D μ mn Now, we need to fnd a value of γ, γ = 1, 2,..., that maxmzes (A-46). Clearly, the value of γ that maxmzes (A-46) wll also maxmze Z 1 nτ γ. (A-47) 1/ x+ 1 D ( γ ) = 1 mn We can rewrte (A-47) as follows: Z 1 nτ nτ = 1 (1 + ) ( γ + 1). (A-48) γ + 1 D D ( γ) mn mn An equvalent form of (A-48) s gven by: Z nτ nτ D 1 = 1+ 2 ( + 1) + ( γ + 1). (A-49) Dmn Dmn nτ γ + 1 mn ( γ) Snce Dmn TTRT nτ, by calculatng the dervaton of (A-49) over γ, we know that (A-49) wll be maxmzed when γ = Dmn / nτ But snce γ can only take postve nteger value, we know that (A-49) wll attan ts maxmum ether at ( D nτ ) γ = max 1, mn / + 1 1, (A-5) 39

42 or γ1 = Dmn / nτ (A-51) That s, the optmal value of TTRT that maxmzes the schedulablty bound s: TTRT mn 1 ( ) ( ) Dmn / γ f Z γ Z γ1 =. (A-52) D / γ otherwse It s easy to verfy that when Dmn / nτ 3, γ = γ1 = 1 and thus (A-52) s equvalent to TTRT = D / γ *, (A-53) mn where 1 f Dmn / nτ 3 γ * = Dmn / γ f Dmn / nτ > 3 and Z γ <Z γ1 Dmn / γ1 otherwse ( ) ( ). (A-54) By substtutng (A-54) nto (A-46), we know that the maxmum schedulablty bound 1 nτ 1 W * ( k / k) = 1 γ * mn( 1, k ). (A-55) 1/ γ * + 1 D μ mn s attaned when TTRT = D / γ * mn. Hence, the theorem follows. Q.E.D. 4

Rate Monotonic (RM) Disadvantages of cyclic. TDDB47 Real Time Systems. Lecture 2: RM & EDF. Priority-based scheduling. States of a process

Rate Monotonic (RM) Disadvantages of cyclic. TDDB47 Real Time Systems. Lecture 2: RM & EDF. Priority-based scheduling. States of a process Dsadvantages of cyclc TDDB47 Real Tme Systems Manual scheduler constructon Cannot deal wth any runtme changes What happens f we add a task to the set? Real-Tme Systems Laboratory Department of Computer

More information

Real-Time Process Scheduling

Real-Time Process Scheduling Real-Tme Process Schedulng ktw@cse.ntu.edu.tw (Real-Tme and Embedded Systems Laboratory) Independent Process Schedulng Processes share nothng but CPU Papers for dscussons: C.L. Lu and James. W. Layland,

More information

An Alternative Way to Measure Private Equity Performance

An Alternative Way to Measure Private Equity Performance An Alternatve Way to Measure Prvate Equty Performance Peter Todd Parlux Investment Technology LLC Summary Internal Rate of Return (IRR) s probably the most common way to measure the performance of prvate

More information

Project Networks With Mixed-Time Constraints

Project Networks With Mixed-Time Constraints Project Networs Wth Mxed-Tme Constrants L Caccetta and B Wattananon Western Australan Centre of Excellence n Industral Optmsaton (WACEIO) Curtn Unversty of Technology GPO Box U1987 Perth Western Australa

More information

FORMAL ANALYSIS FOR REAL-TIME SCHEDULING

FORMAL ANALYSIS FOR REAL-TIME SCHEDULING FORMAL ANALYSIS FOR REAL-TIME SCHEDULING Bruno Dutertre and Vctora Stavrdou, SRI Internatonal, Menlo Park, CA Introducton In modern avoncs archtectures, applcaton software ncreasngly reles on servces provded

More information

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ).

benefit is 2, paid if the policyholder dies within the year, and probability of death within the year is ). REVIEW OF RISK MANAGEMENT CONCEPTS LOSS DISTRIBUTIONS AND INSURANCE Loss and nsurance: When someone s subject to the rsk of ncurrng a fnancal loss, the loss s generally modeled usng a random varable or

More information

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis

The Development of Web Log Mining Based on Improve-K-Means Clustering Analysis The Development of Web Log Mnng Based on Improve-K-Means Clusterng Analyss TngZhong Wang * College of Informaton Technology, Luoyang Normal Unversty, Luoyang, 471022, Chna wangtngzhong2@sna.cn Abstract.

More information

Enabling P2P One-view Multi-party Video Conferencing

Enabling P2P One-view Multi-party Video Conferencing Enablng P2P One-vew Mult-party Vdeo Conferencng Yongxang Zhao, Yong Lu, Changja Chen, and JanYn Zhang Abstract Mult-Party Vdeo Conferencng (MPVC) facltates realtme group nteracton between users. Whle P2P

More information

Checkng and Testng in Nokia RMS Process

Checkng and Testng in Nokia RMS Process An Integrated Schedulng Mechansm for Fault-Tolerant Modular Avoncs Systems Yann-Hang Lee Mohamed Youns Jeff Zhou CISE Department Unversty of Florda Ganesvlle, FL 326 yhlee@cse.ufl.edu Advanced System Technology

More information

BERNSTEIN POLYNOMIALS

BERNSTEIN POLYNOMIALS On-Lne Geometrc Modelng Notes BERNSTEIN POLYNOMIALS Kenneth I. Joy Vsualzaton and Graphcs Research Group Department of Computer Scence Unversty of Calforna, Davs Overvew Polynomals are ncredbly useful

More information

DEFINING %COMPLETE IN MICROSOFT PROJECT

DEFINING %COMPLETE IN MICROSOFT PROJECT CelersSystems DEFINING %COMPLETE IN MICROSOFT PROJECT PREPARED BY James E Aksel, PMP, PMI-SP, MVP For Addtonal Informaton about Earned Value Management Systems and reportng, please contact: CelersSystems,

More information

Period and Deadline Selection for Schedulability in Real-Time Systems

Period and Deadline Selection for Schedulability in Real-Time Systems Perod and Deadlne Selecton for Schedulablty n Real-Tme Systems Thdapat Chantem, Xaofeng Wang, M.D. Lemmon, and X. Sharon Hu Department of Computer Scence and Engneerng, Department of Electrcal Engneerng

More information

Luby s Alg. for Maximal Independent Sets using Pairwise Independence

Luby s Alg. for Maximal Independent Sets using Pairwise Independence Lecture Notes for Randomzed Algorthms Luby s Alg. for Maxmal Independent Sets usng Parwse Independence Last Updated by Erc Vgoda on February, 006 8. Maxmal Independent Sets For a graph G = (V, E), an ndependent

More information

Recurrence. 1 Definitions and main statements

Recurrence. 1 Definitions and main statements Recurrence 1 Defntons and man statements Let X n, n = 0, 1, 2,... be a MC wth the state space S = (1, 2,...), transton probabltes p j = P {X n+1 = j X n = }, and the transton matrx P = (p j ),j S def.

More information

A New Quality of Service Metric for Hard/Soft Real-Time Applications

A New Quality of Service Metric for Hard/Soft Real-Time Applications A New Qualty of Servce Metrc for Hard/Soft Real-Tme Applcatons Shaoxong Hua and Gang Qu Electrcal and Computer Engneerng Department and Insttute of Advanced Computer Study Unversty of Maryland, College

More information

J. Parallel Distrib. Comput.

J. Parallel Distrib. Comput. J. Parallel Dstrb. Comput. 71 (2011) 62 76 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. journal homepage: www.elsever.com/locate/jpdc Optmzng server placement n dstrbuted systems n

More information

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

Module 2 LOSSLESS IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur Module LOSSLESS IMAGE COMPRESSION SYSTEMS Lesson 3 Lossless Compresson: Huffman Codng Instructonal Objectves At the end of ths lesson, the students should be able to:. Defne and measure source entropy..

More information

Performance Analysis and Comparison of QoS Provisioning Mechanisms for CBR Traffic in Noisy IEEE 802.11e WLANs Environments

Performance Analysis and Comparison of QoS Provisioning Mechanisms for CBR Traffic in Noisy IEEE 802.11e WLANs Environments Tamkang Journal of Scence and Engneerng, Vol. 12, No. 2, pp. 143149 (2008) 143 Performance Analyss and Comparson of QoS Provsonng Mechansms for CBR Traffc n Nosy IEEE 802.11e WLANs Envronments Der-Junn

More information

Support Vector Machines

Support Vector Machines Support Vector Machnes Max Wellng Department of Computer Scence Unversty of Toronto 10 Kng s College Road Toronto, M5S 3G5 Canada wellng@cs.toronto.edu Abstract Ths s a note to explan support vector machnes.

More information

A generalized hierarchical fair service curve algorithm for high network utilization and link-sharing

A generalized hierarchical fair service curve algorithm for high network utilization and link-sharing Computer Networks 43 (2003) 669 694 www.elsever.com/locate/comnet A generalzed herarchcal far servce curve algorthm for hgh network utlzaton and lnk-sharng Khyun Pyun *, Junehwa Song, Heung-Kyu Lee Department

More information

8 Algorithm for Binary Searching in Trees

8 Algorithm for Binary Searching in Trees 8 Algorthm for Bnary Searchng n Trees In ths secton we present our algorthm for bnary searchng n trees. A crucal observaton employed by the algorthm s that ths problem can be effcently solved when the

More information

Distributed Optimal Contention Window Control for Elastic Traffic in Wireless LANs

Distributed Optimal Contention Window Control for Elastic Traffic in Wireless LANs Dstrbuted Optmal Contenton Wndow Control for Elastc Traffc n Wreless LANs Yalng Yang, Jun Wang and Robn Kravets Unversty of Illnos at Urbana-Champagn { yyang8, junwang3, rhk@cs.uuc.edu} Abstract Ths paper

More information

An MILP model for planning of batch plants operating in a campaign-mode

An MILP model for planning of batch plants operating in a campaign-mode An MILP model for plannng of batch plants operatng n a campagn-mode Yanna Fumero Insttuto de Desarrollo y Dseño CONICET UTN yfumero@santafe-concet.gov.ar Gabrela Corsano Insttuto de Desarrollo y Dseño

More information

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING

ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING ANALYZING THE RELATIONSHIPS BETWEEN QUALITY, TIME, AND COST IN PROJECT MANAGEMENT DECISION MAKING Matthew J. Lberatore, Department of Management and Operatons, Vllanova Unversty, Vllanova, PA 19085, 610-519-4390,

More information

Calculation of Sampling Weights

Calculation of Sampling Weights Perre Foy Statstcs Canada 4 Calculaton of Samplng Weghts 4.1 OVERVIEW The basc sample desgn used n TIMSS Populatons 1 and 2 was a two-stage stratfed cluster desgn. 1 The frst stage conssted of a sample

More information

AN APPROACH TO WIRELESS SCHEDULING CONSIDERING REVENUE AND USERS SATISFACTION

AN APPROACH TO WIRELESS SCHEDULING CONSIDERING REVENUE AND USERS SATISFACTION The Medterranean Journal of Computers and Networks, Vol. 2, No. 1, 2006 57 AN APPROACH TO WIRELESS SCHEDULING CONSIDERING REVENUE AND USERS SATISFACTION L. Bada 1,*, M. Zorz 2 1 Department of Engneerng,

More information

PAS: A Packet Accounting System to Limit the Effects of DoS & DDoS. Debish Fesehaye & Klara Naherstedt University of Illinois-Urbana Champaign

PAS: A Packet Accounting System to Limit the Effects of DoS & DDoS. Debish Fesehaye & Klara Naherstedt University of Illinois-Urbana Champaign PAS: A Packet Accountng System to Lmt the Effects of DoS & DDoS Debsh Fesehaye & Klara Naherstedt Unversty of Illnos-Urbana Champagn DoS and DDoS DDoS attacks are ncreasng threats to our dgtal world. Exstng

More information

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1

Open Access A Load Balancing Strategy with Bandwidth Constraint in Cloud Computing. Jing Deng 1,*, Ping Guo 2, Qi Li 3, Haizhu Chen 1 Send Orders for Reprnts to reprnts@benthamscence.ae The Open Cybernetcs & Systemcs Journal, 2014, 8, 115-121 115 Open Access A Load Balancng Strategy wth Bandwdth Constrant n Cloud Computng Jng Deng 1,*,

More information

On the Interaction between Load Balancing and Speed Scaling

On the Interaction between Load Balancing and Speed Scaling On the Interacton between Load Balancng and Speed Scalng Ljun Chen, Na L and Steven H. Low Engneerng & Appled Scence Dvson, Calforna Insttute of Technology, USA Abstract Speed scalng has been wdely adopted

More information

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems

Multi-Resource Fair Allocation in Heterogeneous Cloud Computing Systems 1 Mult-Resource Far Allocaton n Heterogeneous Cloud Computng Systems We Wang, Student Member, IEEE, Ben Lang, Senor Member, IEEE, Baochun L, Senor Member, IEEE Abstract We study the mult-resource allocaton

More information

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence

How Sets of Coherent Probabilities May Serve as Models for Degrees of Incoherence 1 st Internatonal Symposum on Imprecse Probabltes and Ther Applcatons, Ghent, Belgum, 29 June 2 July 1999 How Sets of Coherent Probabltes May Serve as Models for Degrees of Incoherence Mar J. Schervsh

More information

1 Example 1: Axis-aligned rectangles

1 Example 1: Axis-aligned rectangles COS 511: Theoretcal Machne Learnng Lecturer: Rob Schapre Lecture # 6 Scrbe: Aaron Schld February 21, 2013 Last class, we dscussed an analogue for Occam s Razor for nfnte hypothess spaces that, n conjuncton

More information

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts

Power-of-Two Policies for Single- Warehouse Multi-Retailer Inventory Systems with Order Frequency Discounts Power-of-wo Polces for Sngle- Warehouse Mult-Retaler Inventory Systems wth Order Frequency Dscounts José A. Ventura Pennsylvana State Unversty (USA) Yale. Herer echnon Israel Insttute of echnology (Israel)

More information

Dominant Resource Fairness in Cloud Computing Systems with Heterogeneous Servers

Dominant Resource Fairness in Cloud Computing Systems with Heterogeneous Servers 1 Domnant Resource Farness n Cloud Computng Systems wth Heterogeneous Servers We Wang, Baochun L, Ben Lang Department of Electrcal and Computer Engneerng Unversty of Toronto arxv:138.83v1 [cs.dc] 1 Aug

More information

Fault tolerance in cloud technologies presented as a service

Fault tolerance in cloud technologies presented as a service Internatonal Scentfc Conference Computer Scence 2015 Pavel Dzhunev, PhD student Fault tolerance n cloud technologes presented as a servce INTRODUCTION Improvements n technques for vrtualzaton and performance

More information

On the Interaction between Load Balancing and Speed Scaling

On the Interaction between Load Balancing and Speed Scaling On the Interacton between Load Balancng and Speed Scalng Ljun Chen and Na L Abstract Speed scalng has been wdely adopted n computer and communcaton systems, n partcular, to reduce energy consumpton. An

More information

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP)

1. Fundamentals of probability theory 2. Emergence of communication traffic 3. Stochastic & Markovian Processes (SP & MP) 6.3 / -- Communcaton Networks II (Görg) SS20 -- www.comnets.un-bremen.de Communcaton Networks II Contents. Fundamentals of probablty theory 2. Emergence of communcaton traffc 3. Stochastc & Markovan Processes

More information

Survey on Virtual Machine Placement Techniques in Cloud Computing Environment

Survey on Virtual Machine Placement Techniques in Cloud Computing Environment Survey on Vrtual Machne Placement Technques n Cloud Computng Envronment Rajeev Kumar Gupta and R. K. Paterya Department of Computer Scence & Engneerng, MANIT, Bhopal, Inda ABSTRACT In tradtonal data center

More information

On the Optimal Control of a Cascade of Hydro-Electric Power Stations

On the Optimal Control of a Cascade of Hydro-Electric Power Stations On the Optmal Control of a Cascade of Hydro-Electrc Power Statons M.C.M. Guedes a, A.F. Rbero a, G.V. Smrnov b and S. Vlela c a Department of Mathematcs, School of Scences, Unversty of Porto, Portugal;

More information

Power Low Modified Dual Priority in Hard Real Time Systems with Resource Requirements

Power Low Modified Dual Priority in Hard Real Time Systems with Resource Requirements Power Low Modfed Dual Prorty n Hard Real Tme Systems wth Resource Requrements M.Angels Moncusí, Alex Arenas {amoncus,aarenas}@etse.urv.es Dpt d'engnyera Informàtca Matemàtques Unverstat Rovra Vrgl Campus

More information

The Greedy Method. Introduction. 0/1 Knapsack Problem

The Greedy Method. Introduction. 0/1 Knapsack Problem The Greedy Method Introducton We have completed data structures. We now are gong to look at algorthm desgn methods. Often we are lookng at optmzaton problems whose performance s exponental. For an optmzaton

More information

A Secure Password-Authenticated Key Agreement Using Smart Cards

A Secure Password-Authenticated Key Agreement Using Smart Cards A Secure Password-Authentcated Key Agreement Usng Smart Cards Ka Chan 1, Wen-Chung Kuo 2 and Jn-Chou Cheng 3 1 Department of Computer and Informaton Scence, R.O.C. Mltary Academy, Kaohsung 83059, Tawan,

More information

MAC Layer Service Time Distribution of a Fixed Priority Real Time Scheduler over 802.11

MAC Layer Service Time Distribution of a Fixed Priority Real Time Scheduler over 802.11 Internatonal Journal of Software Engneerng and Its Applcatons Vol., No., Aprl, 008 MAC Layer Servce Tme Dstrbuton of a Fxed Prorty Real Tme Scheduler over 80. Inès El Korb Ecole Natonale des Scences de

More information

A Design Method of High-availability and Low-optical-loss Optical Aggregation Network Architecture

A Design Method of High-availability and Low-optical-loss Optical Aggregation Network Architecture A Desgn Method of Hgh-avalablty and Low-optcal-loss Optcal Aggregaton Network Archtecture Takehro Sato, Kuntaka Ashzawa, Kazumasa Tokuhash, Dasuke Ish, Satoru Okamoto and Naoak Yamanaka Dept. of Informaton

More information

A Replication-Based and Fault Tolerant Allocation Algorithm for Cloud Computing

A Replication-Based and Fault Tolerant Allocation Algorithm for Cloud Computing A Replcaton-Based and Fault Tolerant Allocaton Algorthm for Cloud Computng Tork Altameem Dept of Computer Scence, RCC, Kng Saud Unversty, PO Box: 28095 11437 Ryadh-Saud Araba Abstract The very large nfrastructure

More information

"Research Note" APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES *

Research Note APPLICATION OF CHARGE SIMULATION METHOD TO ELECTRIC FIELD CALCULATION IN THE POWER CABLES * Iranan Journal of Scence & Technology, Transacton B, Engneerng, ol. 30, No. B6, 789-794 rnted n The Islamc Republc of Iran, 006 Shraz Unversty "Research Note" ALICATION OF CHARGE SIMULATION METHOD TO ELECTRIC

More information

Efficient Bandwidth Management in Broadband Wireless Access Systems Using CAC-based Dynamic Pricing

Efficient Bandwidth Management in Broadband Wireless Access Systems Using CAC-based Dynamic Pricing Effcent Bandwdth Management n Broadband Wreless Access Systems Usng CAC-based Dynamc Prcng Bader Al-Manthar, Ndal Nasser 2, Najah Abu Al 3, Hossam Hassanen Telecommuncatons Research Laboratory School of

More information

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression

A Novel Methodology of Working Capital Management for Large. Public Constructions by Using Fuzzy S-curve Regression Novel Methodology of Workng Captal Management for Large Publc Constructons by Usng Fuzzy S-curve Regresson Cheng-Wu Chen, Morrs H. L. Wang and Tng-Ya Hseh Department of Cvl Engneerng, Natonal Central Unversty,

More information

Calculating the high frequency transmission line parameters of power cables

Calculating the high frequency transmission line parameters of power cables < ' Calculatng the hgh frequency transmsson lne parameters of power cables Authors: Dr. John Dcknson, Laboratory Servces Manager, N 0 RW E B Communcatons Mr. Peter J. Ncholson, Project Assgnment Manager,

More information

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic

Institute of Informatics, Faculty of Business and Management, Brno University of Technology,Czech Republic Lagrange Multplers as Quanttatve Indcators n Economcs Ivan Mezník Insttute of Informatcs, Faculty of Busness and Management, Brno Unversty of TechnologCzech Republc Abstract The quanttatve role of Lagrange

More information

Efficient On-Demand Data Service Delivery to High-Speed Trains in Cellular/Infostation Integrated Networks

Efficient On-Demand Data Service Delivery to High-Speed Trains in Cellular/Infostation Integrated Networks IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. XX, NO. XX, MONTH 2XX 1 Effcent On-Demand Data Servce Delvery to Hgh-Speed Trans n Cellular/Infostaton Integrated Networks Hao Lang, Student Member,

More information

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm

A New Task Scheduling Algorithm Based on Improved Genetic Algorithm A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng Envronment Congcong Xong, Long Feng, Lxan Chen A New Task Schedulng Algorthm Based on Improved Genetc Algorthm n Cloud Computng

More information

Forecasting the Direction and Strength of Stock Market Movement

Forecasting the Direction and Strength of Stock Market Movement Forecastng the Drecton and Strength of Stock Market Movement Jngwe Chen Mng Chen Nan Ye cjngwe@stanford.edu mchen5@stanford.edu nanye@stanford.edu Abstract - Stock market s one of the most complcated systems

More information

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy

Answer: A). There is a flatter IS curve in the high MPC economy. Original LM LM after increase in M. IS curve for low MPC economy 4.02 Quz Solutons Fall 2004 Multple-Choce Questons (30/00 ponts) Please, crcle the correct answer for each of the followng 0 multple-choce questons. For each queston, only one of the answers s correct.

More information

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network *

Data Broadcast on a Multi-System Heterogeneous Overlayed Wireless Network * JOURNAL OF INFORMATION SCIENCE AND ENGINEERING 24, 819-840 (2008) Data Broadcast on a Mult-System Heterogeneous Overlayed Wreless Network * Department of Computer Scence Natonal Chao Tung Unversty Hsnchu,

More information

Availability-Based Path Selection and Network Vulnerability Assessment

Availability-Based Path Selection and Network Vulnerability Assessment Avalablty-Based Path Selecton and Network Vulnerablty Assessment Song Yang, Stojan Trajanovsk and Fernando A. Kupers Delft Unversty of Technology, The Netherlands {S.Yang, S.Trajanovsk, F.A.Kupers}@tudelft.nl

More information

A Lyapunov Optimization Approach to Repeated Stochastic Games

A Lyapunov Optimization Approach to Repeated Stochastic Games PROC. ALLERTON CONFERENCE ON COMMUNICATION, CONTROL, AND COMPUTING, OCT. 2013 1 A Lyapunov Optmzaton Approach to Repeated Stochastc Games Mchael J. Neely Unversty of Southern Calforna http://www-bcf.usc.edu/

More information

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center

How To Solve An Onlne Control Polcy On A Vrtualzed Data Center Dynamc Resource Allocaton and Power Management n Vrtualzed Data Centers Rahul Urgaonkar, Ulas C. Kozat, Ken Igarash, Mchael J. Neely urgaonka@usc.edu, {kozat, garash}@docomolabs-usa.com, mjneely@usc.edu

More information

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by

8.5 UNITARY AND HERMITIAN MATRICES. The conjugate transpose of a complex matrix A, denoted by A*, is given by 6 CHAPTER 8 COMPLEX VECTOR SPACES 5. Fnd the kernel of the lnear transformaton gven n Exercse 5. In Exercses 55 and 56, fnd the mage of v, for the ndcated composton, where and are gven by the followng

More information

taposh_kuet20@yahoo.comcsedchan@cityu.edu.hk rajib_csedept@yahoo.co.uk, alam_shihabul@yahoo.com

taposh_kuet20@yahoo.comcsedchan@cityu.edu.hk rajib_csedept@yahoo.co.uk, alam_shihabul@yahoo.com G. G. Md. Nawaz Al 1,2, Rajb Chakraborty 2, Md. Shhabul Alam 2 and Edward Chan 1 1 Cty Unversty of Hong Kong, Hong Kong, Chna taposh_kuet20@yahoo.comcsedchan@ctyu.edu.hk 2 Khulna Unversty of Engneerng

More information

What is Candidate Sampling

What is Candidate Sampling What s Canddate Samplng Say we have a multclass or mult label problem where each tranng example ( x, T ) conssts of a context x a small (mult)set of target classes T out of a large unverse L of possble

More information

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks

Analysis of Energy-Conserving Access Protocols for Wireless Identification Networks From the Proceedngs of Internatonal Conference on Telecommuncaton Systems (ITC-97), March 2-23, 1997. 1 Analyss of Energy-Conservng Access Protocols for Wreless Identfcaton etworks Imrch Chlamtac a, Chara

More information

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT

APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT APPLICATION OF PROBE DATA COLLECTED VIA INFRARED BEACONS TO TRAFFIC MANEGEMENT Toshhko Oda (1), Kochro Iwaoka (2) (1), (2) Infrastructure Systems Busness Unt, Panasonc System Networks Co., Ltd. Saedo-cho

More information

Sngle Snk Buy at Bulk Problem and the Access Network

Sngle Snk Buy at Bulk Problem and the Access Network A Constant Factor Approxmaton for the Sngle Snk Edge Installaton Problem Sudpto Guha Adam Meyerson Kamesh Munagala Abstract We present the frst constant approxmaton to the sngle snk buy-at-bulk network

More information

QBox: Guaranteeing I/O Performance on Black Box Storage Systems

QBox: Guaranteeing I/O Performance on Black Box Storage Systems QBox: Guaranteeng I/O Performance on Black Box Storage Systems Dmtrs Skourts skourts@cs.ucsc.edu Shnpe Kato shnpe@cs.ucsc.edu Department of Computer Scence Unversty of Calforna, Santa Cruz Scott Brandt

More information

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm

A hybrid global optimization algorithm based on parallel chaos optimization and outlook algorithm Avalable onlne www.ocpr.com Journal of Chemcal and Pharmaceutcal Research, 2014, 6(7):1884-1889 Research Artcle ISSN : 0975-7384 CODEN(USA) : JCPRC5 A hybrd global optmzaton algorthm based on parallel

More information

Robust Design of Public Storage Warehouses. Yeming (Yale) Gong EMLYON Business School

Robust Design of Public Storage Warehouses. Yeming (Yale) Gong EMLYON Business School Robust Desgn of Publc Storage Warehouses Yemng (Yale) Gong EMLYON Busness School Rene de Koster Rotterdam school of management, Erasmus Unversty Abstract We apply robust optmzaton and revenue management

More information

In some supply chains, materials are ordered periodically according to local information. This paper investigates

In some supply chains, materials are ordered periodically according to local information. This paper investigates MANUFACTURING & SRVIC OPRATIONS MANAGMNT Vol. 12, No. 3, Summer 2010, pp. 430 448 ssn 1523-4614 essn 1526-5498 10 1203 0430 nforms do 10.1287/msom.1090.0277 2010 INFORMS Improvng Supply Chan Performance:

More information

Multiple-Period Attribution: Residuals and Compounding

Multiple-Period Attribution: Residuals and Compounding Multple-Perod Attrbuton: Resduals and Compoundng Our revewer gave these authors full marks for dealng wth an ssue that performance measurers and vendors often regard as propretary nformaton. In 1994, Dens

More information

The Load Balancing of Database Allocation in the Cloud

The Load Balancing of Database Allocation in the Cloud , March 3-5, 23, Hong Kong The Load Balancng of Database Allocaton n the Cloud Yu-lung Lo and Mn-Shan La Abstract Each database host n the cloud platform often has to servce more than one database applcaton

More information

Price Competition in an Oligopoly Market with Multiple IaaS Cloud Providers

Price Competition in an Oligopoly Market with Multiple IaaS Cloud Providers Prce Competton n an Olgopoly Market wth Multple IaaS Cloud Provders Yuan Feng, Baochun L, Bo L Department of Computng, Hong Kong Polytechnc Unversty Department of Electrcal and Computer Engneerng, Unversty

More information

Fair Virtual Bandwidth Allocation Model in Virtual Data Centers

Fair Virtual Bandwidth Allocation Model in Virtual Data Centers Far Vrtual Bandwdth Allocaton Model n Vrtual Data Centers Yng Yuan, Cu-rong Wang, Cong Wang School of Informaton Scence and Engneerng ortheastern Unversty Shenyang, Chna School of Computer and Communcaton

More information

Heuristic Static Load-Balancing Algorithm Applied to CESM

Heuristic Static Load-Balancing Algorithm Applied to CESM Heurstc Statc Load-Balancng Algorthm Appled to CESM 1 Yur Alexeev, 1 Sher Mckelson, 1 Sven Leyffer, 1 Robert Jacob, 2 Anthony Crag 1 Argonne Natonal Laboratory, 9700 S. Cass Avenue, Argonne, IL 60439,

More information

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems

Joint Scheduling of Processing and Shuffle Phases in MapReduce Systems Jont Schedulng of Processng and Shuffle Phases n MapReduce Systems Fangfe Chen, Mural Kodalam, T. V. Lakshman Department of Computer Scence and Engneerng, The Penn State Unversty Bell Laboratores, Alcatel-Lucent

More information

Traffic State Estimation in the Traffic Management Center of Berlin

Traffic State Estimation in the Traffic Management Center of Berlin Traffc State Estmaton n the Traffc Management Center of Berln Authors: Peter Vortsch, PTV AG, Stumpfstrasse, D-763 Karlsruhe, Germany phone ++49/72/965/35, emal peter.vortsch@ptv.de Peter Möhl, PTV AG,

More information

The OC Curve of Attribute Acceptance Plans

The OC Curve of Attribute Acceptance Plans The OC Curve of Attrbute Acceptance Plans The Operatng Characterstc (OC) curve descrbes the probablty of acceptng a lot as a functon of the lot s qualty. Fgure 1 shows a typcal OC Curve. 10 8 6 4 1 3 4

More information

Research Article Enhanced Two-Step Method via Relaxed Order of α-satisfactory Degrees for Fuzzy Multiobjective Optimization

Research Article Enhanced Two-Step Method via Relaxed Order of α-satisfactory Degrees for Fuzzy Multiobjective Optimization Hndaw Publshng Corporaton Mathematcal Problems n Engneerng Artcle ID 867836 pages http://dxdoorg/055/204/867836 Research Artcle Enhanced Two-Step Method va Relaxed Order of α-satsfactory Degrees for Fuzzy

More information

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek

THE DISTRIBUTION OF LOAN PORTFOLIO VALUE * Oldrich Alfons Vasicek HE DISRIBUION OF LOAN PORFOLIO VALUE * Oldrch Alfons Vascek he amount of captal necessary to support a portfolo of debt securtes depends on the probablty dstrbuton of the portfolo loss. Consder a portfolo

More information

denote the location of a node, and suppose node X . This transmission causes a successful reception by node X for any other node

denote the location of a node, and suppose node X . This transmission causes a successful reception by node X for any other node Fnal Report of EE359 Class Proect Throughput and Delay n Wreless Ad Hoc Networs Changhua He changhua@stanford.edu Abstract: Networ throughput and pacet delay are the two most mportant parameters to evaluate

More information

Rapid Estimation Method for Data Capacity and Spectrum Efficiency in Cellular Networks

Rapid Estimation Method for Data Capacity and Spectrum Efficiency in Cellular Networks Rapd Estmaton ethod for Data Capacty and Spectrum Effcency n Cellular Networs C.F. Ball, E. Humburg, K. Ivanov, R. üllner Semens AG, Communcatons oble Networs unch, Germany carsten.ball@semens.com Abstract

More information

Economic Models for Cloud Service Markets

Economic Models for Cloud Service Markets Economc Models for Cloud Servce Markets Ranjan Pal and Pan Hu 2 Unversty of Southern Calforna, USA, rpal@usc.edu 2 Deutsch Telekom Laboratores, Berln, Germany, pan.hu@telekom.de Abstract. Cloud computng

More information

J. Parallel Distrib. Comput. Environment-conscious scheduling of HPC applications on distributed Cloud-oriented data centers

J. Parallel Distrib. Comput. Environment-conscious scheduling of HPC applications on distributed Cloud-oriented data centers J. Parallel Dstrb. Comput. 71 (2011) 732 749 Contents lsts avalable at ScenceDrect J. Parallel Dstrb. Comput. ournal homepage: www.elsever.com/locate/pdc Envronment-conscous schedulng of HPC applcatons

More information

An Analysis of Central Processor Scheduling in Multiprogrammed Computer Systems

An Analysis of Central Processor Scheduling in Multiprogrammed Computer Systems STAN-CS-73-355 I SU-SE-73-013 An Analyss of Central Processor Schedulng n Multprogrammed Computer Systems (Dgest Edton) by Thomas G. Prce October 1972 Techncal Report No. 57 Reproducton n whole or n part

More information

An Adaptive Cross-layer Bandwidth Scheduling Strategy for the Speed-Sensitive Strategy in Hierarchical Cellular Networks

An Adaptive Cross-layer Bandwidth Scheduling Strategy for the Speed-Sensitive Strategy in Hierarchical Cellular Networks An Adaptve Cross-layer Bandwdth Schedulng Strategy for the Speed-Senstve Strategy n erarchcal Cellular Networks Jong-Shn Chen #1, Me-Wen #2 Department of Informaton and Communcaton Engneerng ChaoYang Unversty

More information

A Performance Analysis of View Maintenance Techniques for Data Warehouses

A Performance Analysis of View Maintenance Techniques for Data Warehouses A Performance Analyss of Vew Mantenance Technques for Data Warehouses Xng Wang Dell Computer Corporaton Round Roc, Texas Le Gruenwald The nversty of Olahoma School of Computer Scence orman, OK 739 Guangtao

More information

A Dynamic Energy-Efficiency Mechanism for Data Center Networks

A Dynamic Energy-Efficiency Mechanism for Data Center Networks A Dynamc Energy-Effcency Mechansm for Data Center Networks Sun Lang, Zhang Jnfang, Huang Daochao, Yang Dong, Qn Yajuan A Dynamc Energy-Effcency Mechansm for Data Center Networks 1 Sun Lang, 1 Zhang Jnfang,

More information

Multi-Source Video Multicast in Peer-to-Peer Networks

Multi-Source Video Multicast in Peer-to-Peer Networks ult-source Vdeo ultcast n Peer-to-Peer Networks Francsco de Asís López-Fuentes*, Eckehard Stenbach Technsche Unverstät ünchen Insttute of Communcaton Networks, eda Technology Group 80333 ünchen, Germany

More information

Dynamic Pricing for Smart Grid with Reinforcement Learning

Dynamic Pricing for Smart Grid with Reinforcement Learning Dynamc Prcng for Smart Grd wth Renforcement Learnng Byung-Gook Km, Yu Zhang, Mhaela van der Schaar, and Jang-Won Lee Samsung Electroncs, Suwon, Korea Department of Electrcal Engneerng, UCLA, Los Angeles,

More information

General Auction Mechanism for Search Advertising

General Auction Mechanism for Search Advertising General Aucton Mechansm for Search Advertsng Gagan Aggarwal S. Muthukrshnan Dávd Pál Martn Pál Keywords game theory, onlne auctons, stable matchngs ABSTRACT Internet search advertsng s often sold by an

More information

IWFMS: An Internal Workflow Management System/Optimizer for Hadoop

IWFMS: An Internal Workflow Management System/Optimizer for Hadoop IWFMS: An Internal Workflow Management System/Optmzer for Hadoop Lan Lu, Yao Shen Department of Computer Scence and Engneerng Shangha JaoTong Unversty Shangha, Chna lustrve@gmal.com, yshen@cs.sjtu.edu.cn

More information

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS

INVESTIGATION OF VEHICULAR USERS FAIRNESS IN CDMA-HDR NETWORKS 21 22 September 2007, BULGARIA 119 Proceedngs of the Internatonal Conference on Informaton Technologes (InfoTech-2007) 21 st 22 nd September 2007, Bulgara vol. 2 INVESTIGATION OF VEHICULAR USERS FAIRNESS

More information

2008/8. An integrated model for warehouse and inventory planning. Géraldine Strack and Yves Pochet

2008/8. An integrated model for warehouse and inventory planning. Géraldine Strack and Yves Pochet 2008/8 An ntegrated model for warehouse and nventory plannng Géraldne Strack and Yves Pochet CORE Voe du Roman Pays 34 B-1348 Louvan-la-Neuve, Belgum. Tel (32 10) 47 43 04 Fax (32 10) 47 43 01 E-mal: corestat-lbrary@uclouvan.be

More information

How To Calculate The Accountng Perod Of Nequalty

How To Calculate The Accountng Perod Of Nequalty Inequalty and The Accountng Perod Quentn Wodon and Shlomo Ytzha World Ban and Hebrew Unversty September Abstract Income nequalty typcally declnes wth the length of tme taen nto account for measurement.

More information

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits

Linear Circuits Analysis. Superposition, Thevenin /Norton Equivalent circuits Lnear Crcuts Analyss. Superposton, Theenn /Norton Equalent crcuts So far we hae explored tmendependent (resste) elements that are also lnear. A tmendependent elements s one for whch we can plot an / cure.

More information

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services

An Evaluation of the Extended Logistic, Simple Logistic, and Gompertz Models for Forecasting Short Lifecycle Products and Services An Evaluaton of the Extended Logstc, Smple Logstc, and Gompertz Models for Forecastng Short Lfecycle Products and Servces Charles V. Trappey a,1, Hsn-yng Wu b a Professor (Management Scence), Natonal Chao

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and Ths artcle appeared n a journal publshed by Elsever. The attached copy s furnshed to the author for nternal non-commercal research and educaton use, ncludng for nstructon at the authors nsttuton and sharng

More information

Economic Models for Cloud Service Markets Pricing and Capacity Planning

Economic Models for Cloud Service Markets Pricing and Capacity Planning Economc Models for Cloud Servce Markets Prcng and Capacty Plannng Ranjan Pal 1 and Pan Hu 2 1 Unversty of Southern Calforna, USA, rpal@usc.edu 2 Deutsch Telekom Laboratores, Berln, Germany, pan.hu@telekom.de

More information

A Probabilistic Theory of Coherence

A Probabilistic Theory of Coherence A Probablstc Theory of Coherence BRANDEN FITELSON. The Coherence Measure C Let E be a set of n propostons E,..., E n. We seek a probablstc measure C(E) of the degree of coherence of E. Intutvely, we want

More information

Demographic and Health Surveys Methodology

Demographic and Health Surveys Methodology samplng and household lstng manual Demographc and Health Surveys Methodology Ths document s part of the Demographc and Health Survey s DHS Toolkt of methodology for the MEASURE DHS Phase III project, mplemented

More information

Case Study: Load Balancing

Case Study: Load Balancing Case Study: Load Balancng Thursday, 01 June 2006 Bertol Marco A.A. 2005/2006 Dmensonamento degl mpant Informatc LoadBal - 1 Introducton Optmze the utlzaton of resources to reduce the user response tme

More information