J.C.R. Bennett and H. Zhang, "Hierarchical packet fair queuing algorithms, " Proceedings of ACM SIGCOMM'96, pp.143--156, August 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....higher pricing will lead to lower loads and hence lower loads in the higher classes. A second approach, careful capacity provisioning, involves higher classes having more forwarding resources relative to their expected loads through the use of schedulers such as Weighted Fair Queuing [12] 13] [14]. However, each of these two approaches has the same problem when dealing with Internet traffic. Because of the bursty nature of Internet traffic, a higher class may be overloaded, thus performing worse than a lower class [15] 10] The third approach, strict prioritization, provides consistent ....

J.C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queuing Algorithms, " IEEE/ACM Transactions on Networking, vol. 5, pp. 675--689, Oct. 1997.

....[10] to prevent link starvation of lower classes by high priority tra#c. Each core node in the network is independent of the other core nodes in order to keep scheduling simple and e#cient. The underlying scheduling mechanism is left up to the manufacturer with various options to choose from [14, 33, 34, 35, 36]. The IETF has defined several PHBs to be backward compatible with the original ToS markings. These PHBs include an Expedited Forwarding (EF) PHB [10] for low loss, lowdelay tra#c such as VoIP. Another PHB is the Assured Forwarding (AF) PHB [11] which defines only the congestion priority of the ....

J.C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queuing Algorithms," IEEE/ACM Transactions on Networking, vol. 5, pp. 675--689, Oct. 1997.

....Queuing (PQ) In PQ [12] a number of distinct queues is created and a level of priority is assigned to each one. Packets are scheduled from a particular priority queue in First ComeFirst Served order only when all queues of higher priority are empty. Weighted Fair Queuing (WFQ) In WFQ [13, 14, 15], the allocation of link bandwidth is considered fair, since the bandwidth is allocated in proportion to the weights associated with each queue and free bandwidth is fairly shared between the queues, based also on their weights. PQ WFQ: It is a combination of the PQ and WFQ, where TCL1 is ....

J. Bennett, H. Zhang, "Hierarchical packet fair queuing algorithms," Proc of SIGCOMM 1996.

....higher pricing will lead to lower loads and hence lower loads in the higher classes. A second approach, careful capacity provisioning, involves higher classes having more forwarding resources relative to their expected loads through the use of schedulers such as Weighted Fair Queuing [13] 14] [15]. However, each of these two approaches has the same problem when dealing with Internet trafc. Because of the bursty nature of Internet trafc, a higher class may be overloaded, thus performing worse than a lower class [16] 17] In [1] two principles were proposed in order for a differentiated ....

J.C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queuing Algorithms," IEEE/ACM Trans. Networking, vol.5, pp.675-689, Oct. 1997.

....that higher pricing will lead to lower loads and hence lower loads in the higher classes. A second approach, careful capacity provisioning, involves higher classes having more forwarding resources relative to their expected loads through the use of schedulers such as Weighted Fair Queuing [13, 14, 15]. However, each of these two approaches has the same problem when dealing with Internet tra#c. Because of the bursty nature of Internet tra#c, a higher class may be overloaded, thus performing worse than a lower class [16, 17] In [1] two principles were proposed in order for a di#erentiated ....

J.C.R. Bennett and H. Zhang, "Hierarchical Packet Fair Queuing Algorithms, " IEEE/ACM Trans. Networking, vol.5, pp.675-689, Oct. 1997.

....and to enforce them during runtime. For example, for CPU capacity, there exist the Dynamic Soft Real Time (DSRT) CPU scheduling framework [1] and the hierarchical CPU scheduler based on Starttime Fair Queuing [2] for network bandwidth, RSVP protocol [3] and various packet scheduling algorithms [4,5] are responsible for bandwidth reservation and enforcement, respectively; and for disk I O bandwidth, Cello [6] provides a disk scheduling framework. However, what is missing is a higher level framework that coordinates the reservations of multiple resources required in a distributed service. One ....

J. Bennett and H. Zhang, Hierarchical packet fair queuing algorithms, IEEE/ACM Transactions on Networking 5(5) (1997) 675--689.

....and a buffer manager, respectively. In this model, a proportional delay scheduler dynamically distributes the link bandwidth to the N classes, attempting to maintain the proportional delay constraints of Eq. 1. This is a fundamentally different scheduling approach from WFQ [11] CBQ [20] H PFQ [21], or H FSQ [22] schedulers, in which each class is guaranteed a certain minimum bandwidth. These latter schedulers are designed in the linksharing context, where different organizations or users are guaranteed a certain fraction of a link s capacity, sharing dynamically any available excess ....

J. C. R. Bennett and H. Zhang, "Hierarchical Packet Fair Queuing Algorithms, " IEEE/ACM Trans. Net., vol. 5, Oct. 1997, pp. 675--89.

....hand, require network resources to be allocated such that the throughput is maximized. A network meets these requirements primarily by appropriately scheduling its resources. To appropriately schedule network bandwidth, several packet scheduling algorithms have been proposed in the literature [3, 4, 5, 7, 9, 11, 13, 14, 15, 21]. Furthermore, to enable the network to provide bounds on end to end delay as well as throughput guarantees to various flows, several analyses of these algorithms have been carried out [6, 8, 10, 13] These techniques enable a network to provide deterministic bounds on QoS. Though these techniques ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. IEEE/ACM Transactions on Networking, 5(5), October 1997.

.... packets of the same size as the separate networks, then the integrated services network can provide to applications no worse, and often significantly better, performance than the separate networks by simply employing a roundrobin scheduler for packets belonging to the two application classes (see [2] for an example of an algorithm when packets are of unequal size) This is because when both the application classes have packets to transmit, they each receive C units of bandwidth similar to the separate network scenario; but when one of the application classes does not utilize its fair ....

....in the characteristics of network links and disks: network link throughput is unaffected by the relative order of servicing requests; but for disks, the relative order of servicing requests governs the overall disk throughput. Round robin and fair scheduling algorithms (e.g. WFQ [6] WF 2 Q [2]) determine the order for servicing requests based solely on the fairness criterion; they ignore the seek time and rotational latency incurred while servicing each disk access request. Consequently, using them to arbitrate access to disk bandwidth yields poor performance. In fact, the overhead ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. In Proceedings of SIGCOMM'96, pages 143--156, August 1996.

.... to Previous Work In the recent past, a number of resource management mechanisms have been developed for predictable allocation of processor bandwidth [2, 8, 13, 15, 17, 19, 21, 31] Many of these CPU scheduling mechanisms as well as their counterparts in the network packet scheduling do1 main [4, 6, 22, 26] associate an intrinsic rate with each application and allocate resource bandwidth in proportion to this rate. For instance, many recently proposed algorithms such as start time fair queuing (SFQ) 11] borrowed virtual time (BVT) 8] and SMART [19] are based on the concept of generalized ....

....an appropriate set of weights is the subject of future research. 6 Related Work Recently the design of predictable resource allocation mechanisms has received increasing research attention, both in the context of CPU scheduling [2, 5, 8, 19, 21, 27, 28, 31] as well as network packet scheduling [4, 6, 22, 26]. Whereas each research effort has differed in the exact mechanism employed (e.g. reservations [13, 17] rate based allocation [12, 15, 24, 26] GPSbased fair allocation [8, 11, 21] their broad goals are similar predictable allocation of resource bandwidth. Most of these efforts have focused ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. In Proceedings of SIGCOMM'96, pages 143--156, August 1996.

....can more easily tolerate reductions on throughput. One option is to provide minimum rate guarantees for both streams and allow a conferencing stream to obtain extra bandwidth if all the TCP streams have met their minimum rates. Such rate guarantees can be met by deploying fair queuing schemes [2, 26, 27] at the output interface. However, one difficulty is that TCP always oscillates and transmits in bursts. Additional delays in Acks may have adverse impact on its performance. It requires further research to examine the impact of fair queuing schemes on TCP. 3.6. System Architecture of AMcast ....

J. Bennett and H. Zhang. Hierarchical packet fair queuing algorithms. IEEE/ACM Trans. on Networking, 5(5):675--689, 1997.

....can more easily tolerate reductions on throughput. 14 One option is to provide minimum rate guarantees for both streams and allow a conferencing stream to obtain extra bandwidth if all the TCP streams have met their minimum rates. Such rate guarantees can be met by deploying fair queuing schemes [2, 26, 27] at the output interface. However, one diculty is that TCP always oscillates and transmits in bursts. Additional delays in Acks may have adverse impact on its performance. It requires further research to examine the impact of fair queuing schemes on TCP. 3.6 System Architecture of AMcast Servers ....

J. Bennett and H. Zhang. Hierarchical packet fair queuing algorithms. IEEE/ACM Trans. on Networking, 5(5):675-689, 1997.

.... may incur substantial seek and rotational latency overhead (for short time slices) or yield unacceptable response times (for long time slices) see Figure 1) Finally, adapting processor and network packet scheduling techniques that support a diverse mix of application classes (for instance, see [4, 13, 20]) to service disk requests is difficult. This is because a disk is a fundamentally different resource as compared to processors and network switches. Whereas the throughput of a processor or a network switch is unaffected by the relative order in which requests are serviced, in case of disk, this ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. In Proceedings of SIGCOMM'96, pages 143--156, August 1996.

....and to enforce them during runtime. For example, for CPU capacity, there exist the Dynamic Soft Real Time (DSRT) CPU scheduling framework [1] and the hierarchical CPU scheduler based on Starttime Fair Queuing [2] for network bandwidth, RSVP protocol [3] and various packet scheduling algorithms [4][5] are responsible for bandwidth reservation and enforcement, respectively; and for disk I O bandwidth, Cello [6] provides a disk scheduling framework. However, what is missing is a higher level framework that coordinates the reservations of multiple resources required in a distributed service. ....

.... [6] 12] 5] P [ l ] C P S [ h , l ] P [ h ] S i level 3 level 2 level 1 (a) For services S1 and S4 S m c c l P c k C h g f Q Q Q Q Q Q a Q Q Q Q e d c b Q Q Q n [24] Q level 3 level 2 level 1 P [ l ] C P S [ h , l ] P [ h ] S i i [6] 3] 12, 6] [4, 2] [24, 12] 6, 3] 2] 9] 1] 4] 2] 18] 8] 4] 20, 10] 40, 20] b) For services S2 amd S3 Figure 10. QoS levels and corresponding resource requirements in di erent services For each of service components c i S ; c i P , and c i C , Figure 10 shows the Q in and Q ....

[Article contains additional citation context not shown here]

J. Bennett and H. Zhang, Hierarchical packet fair queuing algorithms, in: IEEE/ACM Trans. on Networking, 5(5), 1997, pp. 675-689.

....that higher pricing will lead to lower loads and hence lower loads in the higher classes. A second approach, careful capacity provisioning, involves higher classes having more forwarding resources relative to their expected loads through the use of schedulers such as Weighted Fair Queuing [14, 15, 16]. However, each of these two approaches has the same problem when dealing with Internet trac. Because of the bursty nature of Internet trac, a higher class may be overloaded, thus performing worse than a lower class [17, 18] The third approach, strict prioritization, provides consistent class ....

J.C.R.Bennett and H.Zhang, \Hierarchical Packet Fair Queuing Algorithms," IEEE/ACM Transactions on Networking, vol. 5, pp. 675-689, Oct. 1997.

....layers and their classes, and dynamic configuration of scheduling parameters for classes. This model of a programmable hierarchical scheduling scheme is different, in principle, from other hierarchical scheduling schemes like CBQ (Class Based Queuing) 30] HPFQ (Hierarchical Packet Fair Queuing) [31], and H FSC (Hierarchical Fair Service Curve) 32] While those other schemes apply a single, hierarchy aware scheduling principle to hierarchically organized classes, our scheme organizes multiple independent flat scheduling principles into a hierarchical structure. As a result, our scheme has ....

....power. Despite, this effect may be counterbalanced by the fact that service time for process computation is not a deterministic function in general (section 2.2.1) There exist a few process scheduling schemes that are hierarchy aware. Some like HGPS (Hierarchical Generalized Packet Sharing) [31] and H SFQ (Hierarchical Start time Fair Queuing) for CPU scheduling [33] are general enough to incorporate multiple levels of hierarchy. Each of these employs a single discipline to be applied to every level of hierarchy. Some others like split level scheduling [34] 35] Bowman [48] and ....

Jon C. R. Bennett and Hui Zhang, "Hierarchical Packet Fair Queuing Algorithms," ACM SIGCOMM'96, 1996.

.... to Previous Work In the recent past, a number of resource management mechanisms have been developed for predictable allocation of processor bandwidth [2, 7, 11, 12, 14, 16, 18, 24, 28] Many of these CPU scheduling mechanisms as well as their counterparts in the network packet scheduling domain [4, 5, 19, 23] associate an intrinsic rate with each application and allocate resource bandwidth in proportion to this rate. For instance, many recently proposed algorithms such as start time fair queuing (SFQ) 9] borrowed virtual time (BVT) 7] and SMART [16] are based on the concept of generalized ....

....falls into the latter category the system will need to employ admission control if threads desire specific performance guarantees. Assuming such a test is employed, fair proportional share schedulers have been shown to provide bounds on the throughput received and the latency incurred by threads [4, 9]. We are currently analyzing SFS to determine the performance guarantees that can be provided to a thread. Note, however, that the scheduling heuristic and the processor affinity bias can weaken the guarantees provided by SFS. Finally, proportional share schedulers such as SFS need to be combined ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. In Proceedings of SIGCOMM'96, pages 143--156, August 1996.

.... that both networks carry packets of the same size then the integrated services network can provide to applications no worse, and often significantly better, performance than the separate networks by simply employing a round robin scheduler for packets belonging to the two application classes (see [2] for an example of an algorithm when packets are of unequal size) This is because, when both the application classes have packets to transmit, they each receive C units of bandwidth similar to the separate network scenario; but when one of the application classes does not utilize its fair ....

....in the characteristics of network links and disks: network link throughput is unaffected by the relative order of servicing requests; but for disks, the relative order of servicing requests governs the overall disk throughput. Round robin and fair scheduling algorithms (e.g. WFQ [6] WF 2 Q [2]) 3 0 20 40 60 80 100 120 140 0 10 20 30 40 50 60 70 80 90 100 Response time (ms) Number of video clients 16 disks, 80 text clients Integrated, SCAN Partitioned, SCAN 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0 20 40 60 80 100 120 140 160 180 200 of deadlines violated ....

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. In Proceedings of SIGCOMM'96, pages 143--156, August 1996.

....Initiation Award CCR 9409666) NASA, Mitsubishi Electric Research Laboratories (MERL) and Sun Microsystems Inc. 1 A flow is the sequence of packets transmitted by a source to a destination. 2 We refer to the output port of a switch as a server. 1 [20] SPFQ [19] WF 2 Q [3] WF 2 Q [4], Leap Forward Virtual Clock [22] etc. Most of these algorithms are either unfair over variable rate servers or require the number of bits transmitted to be counted to achieve fairness over variable rate servers [15] Since counting the number of bits transmitted may be expensive, algorithms ....

....GSQ scheduling algorithm. Observe that the rate controller limits the service received by a flow from the GSQ in any time interval. Hence, intuitively, the difference in service received by any two flows in any time interval from the GSQ is bounded, i.e. 3 The implementation of WF 2 Q in [4] that reduces the overhead of counting the number of bits by performing this operation only at packet boundaries violates the necessary assumption stated in [2, 4] that the virtual time function should have a minimum slope of 1. 4 Subsequent to our work, a fair algorithm that achieves separation ....

[Article contains additional citation context not shown here]

J.C.R. Bennett and H. Zhang. Hierarchical Packet Fair Queuing Algorithms. IEEE/ACM Transactions on Networking, 5(5), October 1997.

No context found.

J.C.R. Bennett and H. Zhang, "Hierarchical packet fair queuing algorithms, " Proceedings of ACM SIGCOMM'96, pp.143--156, August 1996.

No context found.

J. Bennett and H. Zhang, \Hierarchical packet fair queuing algorithms," in Proceedings of SIGCOMM'96, (Stanford, California), Aug. 1996.

No context found.

J. Bennett and H. Zhang. Hierarchical packet fair queuing algorithms. IEEE/ACM Transactions on Networking, 5:675--689, 1997.

No context found.

BENNETT, J., AND ZHANG, H. Hierarchical packet fair queuing algorithms. In Proceedings of SIGCOMM'96 (Stanford, California, Aug. 1996).

No context found.

J. Bennett and H. Zhang. Hierarchical packet fair queuing algorithms. IEEE/ACM Transactions on Networking, 5:675--689, 1997.

No context found.

J. Bennett and H. Zhang. Hierarchical packet fair queuing algorithms. IEEE/ACM Transactions on Networking, 5:675--689, 1997.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC