A Power-Aware And QoS-Aware Service Model On Wireless

2y ago
42 Views
2 Downloads
365.68 KB
11 Pages
Last View : 12d ago
Last Download : 6m ago
Upload by : Kian Swinton
Transcription

A Power-Aware and QoS-Aware Service Model onWireless NetworksHao Zhu and Guohong CaoDepartment of Computer Science & EngineeringThe Pennsylvania State UniversityUniversity Park, PA 16802E-mail: {hazhu, gcao}@cse.psu.eduAbstract— Many studies show that the wireless network interface (WNI) accounts for a significant part of the power consumedby mobile terminals. Thus, putting the WNI into sleep when it isidle is an effective technique to save power. To support streamingapplications, existing techniques cannot put the WNI into sleepdue to strict delay requirements. In this paper, we present anovel power-aware and QoS-aware service model over wirelessnetworks. In the proposed model, mobile terminals use proxies tobuffer data so that the WNIs can sleep for a long time period. Toachieve power-aware communication while satisfying the delayrequirement of each flow, a scheduling scheme, called prioritybased bulk scheduling (PBS), is designed to decide which flowshould be served at which time. Through analysis, we prove thatthe PBS service model can provide delay assurance and achievepower efficiency. We use Audio-on-Demand and Web access ascase studies to evaluate the performance of the PBS servicemodel. Experimental results show that PBS achieves excellentQoS provision for each flow and significantly reduces the powerconsumption.Index Terms: Power-aware, simulations, QoS, scheduling,buffer management, wireless networks.I. I NTRODUCTIONWith the advent of the third generation wireless infrastructure, and the rapid growth of wireless communication technology, pervasive computing becomes possible: people with battery powered mobile terminals (cellular phones, PDAs, handheld computers, etc.) can access various kinds of services atany time any place. However, the goal of achieving ubiquitousconnectivity with small-size and low-cost mobile terminals(MTs) is challenged by the power constraints. Most MTs arepowered by battery, but the rate at which battery performanceimproves is fairly slow [18]. Aside from major breakthroughs,it is doubtful that significant improvement can be expectedin the foreseeable future. Instead of trying to improve theamount of energy that can be packed into a power source,we can carefully design communication protocols so that theMTs can perform the same functions and provide the sameservices while minimizing their overall power consumption.Understanding the power characteristics of the wirelessnetwork interface (WNI) used in MTs is important for designing power efficient communication protocols. A typicalThis work was supported in part by the National Science Foundation(CAREER CCR-0092770 and ITR-0219711).0-7803-8356-7/04/ 20.00 (C) 2004 IEEEWNI may exist in active or sleep state. In the active state, theWNI may be in the transmit, receive and idle modes. Manystudies [3], [12], [22], [24] show that the power consumedin the active state is similar, which is significantly higherthan the power consumed in the sleep state. As a result,most of the work on power management concentrates onputting the WNI into sleep when it is idle. Stemm and Katz[22] studied transport layer approaches and application-drivenapproaches to help power down the WNI. Protocols [9], [10],[19], [21], [26] are proposed to put the WNI into sleep underthe following conditions: the WNI is idle, the use of the WNImay collide with other MTs, or the use of the WNI suffersfrom interference. Most of these works focus on reducingthe power consumption. This can be applied to most dataapplications which do not have QoS requirements, but maynot be enough for streaming applications (e.g. audio/video-ondemand) due to lack of QoS provisions. Steaming applicationshave been very popular over the Internet, and it is becomingpossible for mobile environments. For example, Nokia modelsfrom 5510 [15] start to support MP3. IDC [5] predicts that by2006, in Europe alone, there will be 37 million users listento music delivered through wireless channel. Compared to thepower consumed on music playing, the WNI consumes muchmore energy (as much as 70% of the total power in Nokia5510). Thus, to achieve Audio-on-Demand (AoD) in wirelessnetworks, reducing the power consumption of the WNI is oneof the major issue. Since streaming applications have QoSrequirement, how to achieve power saving without violatingQoS is a big challenge for supporting streaming applicationson wireless networks.It takes some time for a WNI to wake up from sleep. Asreported in [20], the transition time from active to sleep andback to active is on the order of tens of milliseconds. Due tothis state transition delay, existing power management schemesmay not be directly applied to streaming applications since theinter-packet arrival time maybe too small and the MT cannotpower off the WNI without violating the delay requirement.To deal with the transition delay, we propose to support powersave mode using buffers. In the proposed solution, a proxy isadded at the MT side. This proxy buffers enough data fromthe server, and lets the WNI enter sleep until the buffereddata run out. Then, it wakes up the WNI and downloads moreIEEE INFOCOM 2004

data to fill the buffer. Since a base station (BS) may serve alarge number of MTs, simple solutions may not be able tomanage the proxy without violating the QoS. For example, asa simple solution, the MT sends a request to the BS whenits buffer is near empty, and the BS serves the MT’s requestuntil the MT’s buffer is full and then serving another one.Although this approach can minimize the power consumption,it may violate the QoS when multiple MTs send requests atalmost the same time. Under this condition, since the BS canonly serve one MT at a time, some MTs may have to wait fora long time and the delay requirement may be violated.In this paper, we propose a new scheduling scheme, calledpriority-based bulk scheduling (PBS) to utilize the client bufferto save power and provide QoS. Under PBS, the schedulerkeeps track of the amount of prefetched (buffered) data foreach flow. The flow with sufficient buffered data will be suspended until the buffered data runs out, whereas the flow withinsufficient buffered data will be served based on its priority,which is determined by its delay requirement. With buffereddata, the WNI can sleep long enough to offset the impact of thestate transition delay. By suspending some flows, other activeflows sharing the channel can obtain more bandwidth and takeless time to fill the buffer. We also extend the service modelto consider channel errors and applications that do not havestrict delay requirements. Through analysis, we prove that PBScan provide delay assurance for real-time applications and ismore power efficient than conventional rate-based fair queuingmodels. We use Audio-on-Demand and Web access as casestudies to evaluate the performance of the PBS service model.Experimental results show that PBS achieves excellent QoSprovision for each flow and significantly reduces the powerconsumption.The rest of the paper is organized as follows. In the nextsection, we describe the system model. In Section III, wedescribe the details of the PBS service model. In Section IV,we evaluate its performance. Section V presents related work,and Section VI concludes the paper.II. S YSTEM M ODELThe geographical area is divided into cells in a wirelessnetwork. Inside each cell, the base station (BS) communicateswith the mobile terminals (MTs) through uplink and downlinkchannels. Both uplink and downlink channels are divided intotime slots. For uplink, we assume a channel can be eitherrandomly accessed by MTs or granted through downlink (e.g.,by reservation). Location-dependent errors may happen dueto channel fading, interference, etc [13]. For simplicity, weassume the modulation techniques and coding techniques arefixed. Certainly, link adaptation schemes which can select themost suitable modulation and coding technique based on thecurrent channel quality, can be complement to our solution. Inorder to achieve reliable transmission, each downlink packetmust be acknowledged. The channel condition can be measured by the number of packet retransmissions.Since a WNI spends significantly higher amount of power(from 10 to 100 times [3], [12], [22], [24]) during active0-7803-8356-7/04/ 20.00 (C) 2004 IEEEcompared to sleep, we use the accumulated WNI sleep timeto measure the power efficiency of different schemes. In orderto accurately measure the power consumption, we use thefollowing notations. Tof f on : time spent to transit from sleep to active. Ton of f : time spent to transit from active to sleep.We assume that the power consumed during the state transitionfrom active to sleep, or from sleep to active is similar tothat in the active mode, which is much larger than the powerconsumed in the sleep mode.III. T HE PBS S ERVICE M ODELIn this section, we present the PBS service model. Beforelooking into details, let’s first look at the drawbacks of existingscheduling algorithms in terms of power efficiency and QoSprovision.A. Background and MotivationIn order to provide guaranteed service over a shared link,several rate-based service disciplines have been proposed [25].The principle of these service models is to provide each flowwith a guaranteed data rate without being affected by othermis-behaving flows sharing the link. A scheduling algorithmcan be classified as work-conserving or non-work-conserving.In the work-conserving scheduling, a server is never idlewhen there is a packet to send. In the non-work-conservingscheduling, a packet is not served until it is eligible [25], eventhough the server is idle at that time.When applying the existing scheduling algorithms towireless networks, power issues should be considered. Asexplained in Section I, putting the WNI into sleep is the mostwidely used method to save power. When work-conservingservice discipline is used, it is difficult to put the WNI intosleep. The reason is as follows. When an MT shares the linkwith other MTs, if a work-conserving guaranteed servicemodel is applied, the service sequence of the link depends onthe scheduling pattern of all flows. The scheduling pattern isrelated to the number of backlogged flows and the deadlinesof the head-of-line packets of these flows. Unfortunately, theMT does not know the following service sequence due tolack of global information regarding the scheduling patternof all flows in the system. Thus, the WNI has to stay inactive since it does not know when the next packet willarrive. This may cause the WNI to waste a lot of power.For example, as shown in Figure 1, suppose three flowsFlow1(0.2)MT 1Flow2(0.3)LINKFlow3(0.5)Fig. 1.111100000000111100001111MT 200001111000011110000111100001111MT 3000111000111000111An example of work-conserving link sharingIEEE INFOCOM 2004

share a link of capacity C, and each flow has an MT as itsreceiver. Suppose all flows want to send B bits and theirdata rates are: 0.2C, 0.3C and 0.5C respectively. Under thework-conserving service model, if B is quite large, M T1 ’sactive time is approximately 5B/C, but M T1 ’s effectivereceive time is B/C. It is easy to see that almost 80% of thepower has been wasted.Non-workNon-work-conserving Virtual Clock (NVC):conserving scheduling can be used to save power. The basicidea is to let the WNI enter sleep when it is not used. Onesimple approach is to let the BS and the MT mutually agreeon a scheduling pattern. When the BS sends a packet tothe MT, the scheduler piggybacks the information about theeligible time for the next packet to be transmitted. Notethat the eligible time can be calculated based on the flow’sdata rate. The scheduler works in a non-work-conserving[25] manner since it will not serve the flow before theeligible time even though the channel is idle. Thus, theMT can enter sleep for a while until the eligible time ofthe next packet. This simple approach is referred to asthe Non-work-conserving Virtual Clock (NVC) scheduling.Although NVC can save a large amount of power in theory,it may not be possible in practice. This is due to the reasonthat it takes some time and power for the WNI to transitbetween sleep and active. Suppose Tof f on 10ms, if thepacket interval time, denoted by Tintvl , of the flow is lessthan 10ms, we have Ton of f Tof f on Tintvl . Thus,the NVC scheme cannot put the WNI into sleep withoutviolating the QoS requirement of the flow1 . Even whenTon of f Tof f on Tintvl , not too much power can besaved unless Ton of f Tof f on Tintvl . From FigurePacket Non offoff onPacket N 1TintvlFig. 2. The relationship between state transition and the packet interval timeT T2, we can see that on ofTfintvlof f on of the total packetinterval time will be used for state transition. In addition,since this transition also costs a lot of power, the NVCscheduling algorithm cannot save too much power unlessTon of f Tof f on Tintvl .Bulk Scheduling (BKS): Based on the above reasoning,another approach, called Bulk Scheduling (BKS), can be usedto further reduce power. With this service policy, the channelis divided into bulk slots. A flow which needs to be servedwakes up at the beginning of a bulk slot. The schedulerrandomly selects a flow to serve at that time, and the selectedflow will be served until the end of the bulk slot. Meanwhile,other losing flows enter sleep and wake up again to wait for1 In this case, in order to provide QoS, the WNI has to stay in active andthe scheduler serves the flow in a work-conserving manner.0-7803-8356-7/04/ 20.00 (C) 2004 IEEEthe service at the beginning of the next bulk slot. Figure 3t1t2t3B/CB/CB/Ct4B/CBulk Slot 1Bulk Slot 2Bulk Slot 3Bulk Slot 4MT1 is servedMT2 is servedMT3 is servedMT2 is servedFig. 3.t5An example of bulk schedulingshows one example of bulk scheduling. In this example, threeflows share the link and the receivers are M T1 , M T2 , andM T3 respectively. Each bulk slot is equal to B/C, where Bis the number of bits and C is the capacity of the link. ForM T3 , it wakes up at t1 and enters sleep since it found thatthe scheduler has selected M T1 to serve. The same procedurehappens at t2 . At t3 , it wakes up again and starts to receivedata. Suppose M T3 wants to transmit k B bits. If B is largeenough so that the power used for state transitions betweenidle and active is negligible, the active time for M T3 is onlyk B/C, which allows M T3 to stay in sleep for the maximumamount of time. Thus, if the bulk slot is significantly largerthan Ton of f Tof f on , the power consumption of statetransition can be neglected, and bulk scheduling is an optimalservice model in terms of power efficiency. However, bulkscheduling cannot provide QoS when multiple flows requestdata at the same time. For example, at t1 , suppose M T3 andM T1 will miss their deadline if waiting for another B/Ctime slot, scheduling M T1 to serve will force M T3 to missits deadline.In order to address the drawbacks of the NVC approachand the BKS approach, we propose a priority-based bulkscheduling (PBS) service model considering QoS provisionand power efficiency. The basic idea of PBS is to let the MTbuffer as much data as possible without affecting the QoSrequirement of other flows. Relying on the buffered data, theMT can put its WNI into sleep and wake up only when theprefetched data is not enough to satisfy its QoS requirement.In this way, the WNI can power off for many time slots.Furthermore, as the MT enters sleep, other MTs can sharethe channel with fewer MTs (ideally no other MTs). Thus,MTs only spend a very small amount of time in the activemode to prefetch enough data, and then saves power.B. PBS in DetailThe PBS service model has two parts: a scheduler at theBS side and a proxy at the MT side. The scheduler is used tocontrol the channel access among multiple MTs. The proxyis used to coordinate with the scheduler at the MT side.We describe the algorithm used by the scheduler, and showhow the proxy works. Then, we present solutions to dealwith channel errors, calculate the computation complexity andprove some properties of the service model.1) The PBS Scheduler: In PBS, each packet of a flow isassigned a deadline. The scheduler orders the transmission ofpackets according to their deadlines. If a flow’s aggregatedservice goes beyond the minimum service required toIEEE INFOCOM 2004

maintain the QoS2 , it will be removed from the schedulingregion until it needs more data to maintain the QoS. As aresult, the flow is suspended periodically and the data aretransmitted in the form of several runs of packets.Notations:A: the set of flows in active stateDi : the deadline of the head-of-line packet of fit: the current timeWithout loss of generality, forService accounting:each flow, we assume that the first served packet afterentering the scheduling region is indexed by 1. Each packetis assigned a deadline according to the packet length andthe flow’s data rate. Formally, the deadlines (in seconds) arecomputed as follows:schedule()1 begin:2 if (A N U LL)3{ idle in the time slot; goto begin; }4 if (no primary flow)5select the primary flow fi according to Eq. (3)6 if ( t arg min{Dj } j i )d1idji ei dj 1ilj 1 iri(1)where pji is the j th packet of flow fi , ei is the eligible time offi , dji is the deadline of packet pji , lij is the length of pji and riis the data rate of fi (in bps). Suppose, pni is the head-of-linepacket of fi at time t, the ahead-service (in seconds) of fi ,denoted by aheadi , is computed as follows:lin(2)riwhere I(i) returns 1 if fi is being served, otherwise, it returns0. Since aheadi is used to trace the amount of prefetcheddata at the MT side, it cannot be negative at any time.aheadi (t) max(dni t, 0) I(i) 78910111213141516j Ai j; /* the deadline of the secondary flow fjwill be violated, so serve fj */p fi .deque(); /* get the packet to be transmitted */if (aheadi M axservi (aheadi φ j(fj A aheadj φ))mark(p);send(p);if (the transmission is successful){ Di Di p.length/ri ;if (p is marked){ei t aheadi ; statei idle; } }goto beginFig. 4.At the BS side, eachScheduling state management:flow has two scheduling states: idle and active. Thetransitions between scheduling states of fi (denoted bystatei ) are controlled by the scheduler. For the purpose offlow control, there is an upper limit of ahead-service for eachflow fi , denoted by M axservi . When aheadi M axservi ,the scheduler stops serving it and changes statei to idle.Suppose the scheduler provides enough ahead-service tofi ; i.e., aheadi φ, φ is a system parameter to representthe lower bound for ahead-service. If there exists anotherflow, say fj , which have not got enough ahead-service(i.e.,aheadj φ), the scheduler lets fi yield the channel toother flows by changing statei to be idle. Whenever stateiis changed to idle, the eligible time of fi is set to be thecurrent time plus aheadi . At the same time, the schedulerupdates the value of the eligible time to indicate when fiwill be moved back in scheduling region again. Whenever fiis changed to idle, the scheduler will not serve it until theeligible time expires. After the eligible time expires, its statewill be changed to active again (not shown in Figure 4).When statei is set to idle, the scheduler should notifyM Ti which can shutdown its WNI. We assume that the BScan mark the packet transmitted to the M T that will poweroff its WNI after receiving the marked packet. Since the MT’saddress is equal to the destination address of the packet, theBS can simply use one bit in the packet header to represent2 At this time, the MT of the flow has enough prefetched data. For simplicity,we say the flow has got enough ahead-service.0-7803-8356-7/04/ 20.00 (C) 2004 IEEEThe PBS Algorithmwhether the packet is marked or not. When the MT receivesthe marked packet, it replies an ACK and puts its WNI intosleep. When the BS receives the ACK, it knows that the WNIhas been powered off, and suspends the related flow.The scheduler: The PBS scheduler works as follows. Whenn (n 1) flows are active, the scheduler selects one flow asthe primary flow, and the other n 1 flows are secondary flows.At any time, the scheduler exclusively serves the primaryflow provided that the deadlines of the secondary flows willnot be violated. If the deadline of the secondary will beviolated, the scheduler has to serve the secondary flow in orderto meet the QoS requirement of the flow. In other words,the scheduler serves the primary flow in a work-conservingmanner, whereas each secondary flow is served in a noni )ri work-conserving way. As (φ aheadC j A rj (See Property 2 inSection III-C) is an approximation of how fast a flow fi canhave enough ahead-service as the primary flow, the scheduleralways selects the flow that can take the shortest time to getenough ahead-service as the primary flow. Formally, at timet, the primary flow (fprim ) is selected as follows:fprim arg min{i A(φ aheadi (t))ri }C j A rj(3)where A is the set of flows in active state and C is the channelIEEE INFOCOM 2004

capacity. The principle behind Eq (3) is similar to the shortestjob first policy, which can minimize the average waiting time.Thus, the average time for each flow to get enough aheadservice is also minimized under PBS.applications, other applications (e.g. FTP, WWW) may nothave stringent delay requirements. Suppose a non-streamingflow, say fk , requires data and the channel utilization is high.If fk is admitted into the scheduling region, from Eq (3), wecan see that the time spent by the primary flow to get enoughahead-service will be increased, and the secondary flows haveto wait longer before being the primary flow. Thus, all active471234 5f3flows spend more time in the active state, and the powerconsumption is also increased. Since fk does not have strict123 4 567f2(head2(1.5) 5.5, e2 7.0)delay requirement, it is better to postpone the serving timeof fk for a specified time period, denoted by yieldk . We123 456f1(head1(0.8) 5.2, e1 6.0)assign an integer relax factor (σi ) to each flow fi , which isbounded by σimax . For the flow with strict delay requirement,1.50.81.06.00.0time the upper bound is simply set to 0. For fi with σi 0, whenits deadline expires, the scheduler decides whether to serve itFig. 5. An illustration of the PBS schemeor not according to the current system utilization, which canbe measured by the number of active flows. If the currentFigure 5 shows how the PBS scheme works. There are three system utilization is greater than a threshold, denoted bybacklogged flows (f1 , f2 , f3 ) in the system. Each flow has µthresh , the scheduler lets fi yield the channel for a period1Kbps data rate and unlimited M axserv. Suppose φ 5.0 of yield . Otherwise, f is served.iiseconds, C 10 Kbps, and all packets have the same packetTo avoid starvation, an adaptive scheme is used to managelength of 1 Kb. At time 0.0, the eligible time of all flowsσj as follows.(f1 .f3 ) expires, and the ahead-service of each flow is 0.1) When the scheduler decides to let fj yield the channel, σjSuppose f1 is selected as the primary flow at time 0.0. To meetis decreased by one;the deadlines of f2 and f3 , the scheduler serves p12 and p13 first.2) When fj leaves the channel with the ahead-service greaterFrom time 0.2 to 0.8, without violating the deadlines of f2 andthanφ yieldj , σj is increased by one. The service loss duef3 , which are equal to 1.0, p11 .p61 are served back-to-back. Attoyieldingis compensated by decreasing the ahead-service bytime 0.8, according to Eq (2), ahead1 is d61 0.8 5.2, whichyieldi .is greater than φ. Thus, f1 is suspended and its eligible time isBy using the adaptive scheme, the maximum delay for fi toset to 6.0. At time 0.8, f2 becomes the primary flow. Followbe admitted into the scheduling region is σi yieldi .the same procedure, f2 is suspended from time 1.5 to 7.0.2) The PBS Proxy: A proxy is associated with the receiverAfter time 1.5, f3 is the only active flow, and the schedulerofeach flow. The proxy downloads data from the BS, monitorsserves f3 until time 6.0 when the eligible time of f1 expires.the amount of prefetched data, and manages the operationmodes of the WNI. The proxy coordinates with the server anddecides whether the WNI should enter sleep. Similar to the571234567 8f3scheduler, the proxy can calculate the ahead-service of eachflow according to the flow’s data rate, packet length and the6123457(head2(1.9) 5.1, e2 7.0)f2arrival time of each packet. If the proxy finds that the packet ismarked, the WNI will be shutdown for a time period equal to1234567f1(head1(1.8) 5.2, e1 7.0)the calculated ahead-service. In this way, the control overheadcan be reduced since the scheduler does not need to tell the1.81.97.0 time0.0length of the sleep period.If multiple applications are supported, the WNI is shut downFig. 6. An illustration of the WFQ schemeonly when all proxies running on the MT have requested toTo demonstrate the power efficiency of PBS, we compare do so. However, the WNI wakes up if any proxy needs it atthe time period of each flow to get enough ahead-service any time. Since the MT knows all proxies running on it, thisunder WFQ [17] and PBS. As shown in Figure 6, under can be easily implemented.WFQ, f1 needs 1.8 seconds to get enough ahead-service, and3) Dealing with Channel Errors: The wireless commuf2 needs 1.9 seconds. Comparing to the correspondent time nication channel is error prone, and the error is locationperiods in Figure 5 (e.g., f1 needs 0.8 second), we can see dependent and bursty. If channel errors exist, the probabilitythat, on average, the time period of each flow to get enough of a successful transmission becomes very low, and hence,ahead-service under PBS is much smaller than that under bandwidth and power may be wasted during re-transmissions.WFQ.Similar to [2], [13], [14], we deal with channel errors byswapping time slots from flows suffering channel errors toApplication-aware extensions: Compared with streaming flows which have good channel conditions. However, we focus0-7803-8356-7/04/ 20.00 (C) 2004 IEEEIEEE INFOCOM 2004

sji dji ( Q 1)LmaxC(4)where sji is the time when the server starts serving pji andLmax is the maximum packet length.Property 2: (Power Efficiency) Suppose Q is the set ofbacklogged flows in the system and Q r C. Consider theprocess that each flow prefetches the ahead-service greaterthan or equal to φ and leaves the channel. If φrLmax ,the average active time under PBS and under weightedfair queuing (WFQ), denoted by T̄act,P BS and T̄act,W F Qrespectively, follows:φr LmaxC/ Q r( Q i 1)(φr ( Q i 1)Lmax )i 1 Q (C ( Q i 1)r)T̄act,W F Q Q T̄act,P BS(5)Property 2 gives the lower bound ratio of the average activetime under WFQ to that under PBS. We give an exampleto show some numerical results of the ratio as a function of Q . Suppose C 400Kbps, r 50Kbps, φ 500ms, andLmax 1000bits. As shown in Figure 7, when Q increasesfrom 2 to 6, the ratio lower bound increases from 1.26 to 2.14.This shows that PBS is more power efficient than WFQ.2.22.1The lower bound of Trecv,WFQ/ Trecv,DPBSon minimizing the influence of channel errors on QoS andpower consumption of each flow under the PBS service model.In our approach, every packet is delivered in a DATA-ACKorder. If the BS did not receive the ACK of a packet, itknows that the transmission fails. After a transmission failure,the scheduler re-transmits the packet up to three times. Ifthe packet still cannot be delivered, the BS assumes that theflow, say fi , is experiencing a channel error. In this case, ifaheadi φ, the BS stops fi . At the same time, the proxy of fialso realizes the error problem from the transmission failuresand shuts down the WNI. If aheadi φ, it may not worth topower off the WNI since the actual sleep time could be tooshort. Thus, the scheduler stops serving fi for a pre-specifiedshort period, which is called backoff period, and the proxyof fi lets the WNI stay in active. After the backoff period,the scheduler will resume serving fi . If fi still suffers fromchannel errors, the same backoff procedure will be applied tofi again.Since most channel errors are bursty, after leaving the channel for some time, the flow may get good channel state whenbeing served. After a flow leaves the channel, the total numberof flows sharing the channel decreases and the allocated datarate of each remaining flow increases. As a result, these activeMTs spend less time in active, and then reduce the powerconsumption. Under rare situations, the channel error may lastlong enough to violate the QoS requirement. At this time,the only solution left is to change the modulation and codingschemes. Since this is not the major concern of this paper, wewill not further discuss it.4) Computation complexity of PBS: It only takes onedivision and one addition to get the ahead-service (in seconds)at the MT side. At the BS side, the operation consists of theselection of the primary flow at the cost of O(log(n)), and theupdates for ahead-service per-flow at the cost of O(n). Thus,PBS is computationally feasible in many wireless networksthat have a moderate number of flows per BS. MTs can easilyget the data rate of the flow, φ, Maxservi , the backoff periodand yieldi (for non-streaming applications) during sessioninitialization. Since the performance (in terms of QoS andpower efficiency) of PBS is not sensitive to the selection of φeven when the system is heavily loaded (see Section IV-D), itdoes not require precise information about the length of thestate transition delay. The only requirement is that φ is muchlarger than the delay.21.91.81.71.61.51.41.31.222.533.544.555.56 Q Fig. 7.The numerical results of the lower bound ofT̄act,W F QT̄act,P BSIV. P ERFORMANCE E VALUATIONSA. The Experimental SetupC. Property of PBSIn this section, we present some important properties of thePBS scheme. We prove that

novel power-aware and QoS-aware service model over wireless networks. In the proposed model, mobile terminals use proxies to buffer data so that the WNIs can sleep for a long time period. To achieve power-aware communication while satisfying the delay requirement of

Related Documents:

Catalyst 3750 Switch Software Configuration Guide 78-15164-04 29 Configuring QoS This chapter describes how to configure quality of service (QoS) by using automatic QoS (auto-QoS) commands or by using standard QoS commands on the Catalyst 3750 switch. With QoS, you can provide preferenti

HowtoConfigureAuto-QoS 106 ConfiguringAuto-QoS 106 EnablingAuto-QoS 106 TroubleshootingAuto-QoS 108 MonitoringAuto-QoS 109 ConfigurationExamplesforAuto-Qos 110 Examples:GlobalAuto-QoSConfiguration 110 QoS Configuration Guide, Cisco IOS Release 15.2(2)E (Catalyst 2960-X Switches) vi OL-32552-01 Contents

QoS technologies, or tools, each have speciic roles that are used in conjuncion with one another to build end-to-end network QoS policies. Skype QoS A Powerful Duo Together, Skype and QoS make a powerful duo for two reasons: 1. QoS has a signiicant impact on the network when it comes to high-bandwidth video and mulimedia transmissions.

QoS awareness for the special live-soccer event applications in smart cities, in which the QoS-aware routing mechanism is proposed, called LSEA. The main contributions of this paper are summarized as follows. (1) SDN-based QoS-aware routing model is introduced, where delay, delay jitter, and packet loss rate are considered as three objects for .

LTE - QoS LTE architecture supports hard QoS, with end-to-end quality of service and guaranteed bit rate GBR for radio bearers. Just as Ethernet and the internet have different types of QoS, for example, various levels of QoS can be applied to LTE traffic for different applications. Because the LTE MAC is fully scheduled, QoS is a natural fit.File Size: 381KB

Chapter-7 VoIP Case Study Voice Bandwidth Calculators . www.itttc.net 3 Quality of Service (QoS) Chapter 1 Cisco VoIP Implementations . 1.3 Encapsulating Voice Packets 1.4 Bandwidth Calculation 1.5 Implementing VoIP Support in Enterprise Network Chapter 2 IP QoS 2.1 Introduction to QoS 2.2 Identifying and Comparing QoS Models 2.3 QoS .

times of congestion. Quality of Service (QoS) management is the collection of mechanisms that control how traffic is prioritized and handled during these times. QoS technologies, or tools, each have specific rolls that are used in conjunction with one another to build end-to-end network QoS policies. The two most common QoS tools used to

Abrasive water jet machining (AWJM) process is one of the most recent developed non-traditional machining processes used for machining of composite materials. In AWJM process, machining of work piece material takes place when a high speed water jet mixed with abrasives impinges on it. This process is suitable for heat sensitive materials especially composites because it produces almost no heat .