Srinivasan Keshav. On the efficient implementation of fair queueing. In Internetworking.' Research and Experience Vol. 2, 157-173, September 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and zero window constraints, order highest window denominator first Equal Deadlines and equal non zero window constraints, order lowest window numerator first All other cases: first come first serve Table 1. Scheduler Decision Rules Related Work A number of efforts [15] 10] 8] and [7] have detailed and constructed priority queuing architectures and corresponding hardware in switches and network interfaces. The DWCS scheduling framework allows flexible reconfiguration of the native scheduling discipline into EDF, static priority and fair share using more complex multiple ....

Srinivasan Keshav. On the efficient implementation of fair queueing. In Internetworking.' Research and Experience Vol. 2, 157-173, September 1991.

....given that resources have been allocated in advance. Closed loop flow control requires the sender to adjust its rate based on feedback 4 Parekh [56] has shown that, if a source is shaped by a token bucket with leaky bucket rate control and scheduled by the weighted fair queuing service discipline [58], it is possible to achieve strong guarantees on delay. from the receiver [41] or network [64] Applications using closed loop flow control based protocols must be able to adapt to fluctuations in the available resources. On the other hand, applications which cannot adjust to changes in the ....

....uncompressing a video flow is computationally more intensive than displaying video to a screen) In contrast, switch and router schedulers are generally moving packets cells from queues to ports or vice versa and are optimized for that task. Therefore, techniques resident in switches (such as HRR [58]) may be inappropriate in host operating systems. 6 Conclusion In this paper, we have argued that multimedia systems designers should adopt an end to end approach to meet application level QoS requirements. To meet this challenge we have proposed a generalized QoS framework that is motivated by ....

Keshav S (1991) On the Efficient Implementation of Fair Queueing, Internetworking: Research and Experiences, Vol. 2, pp 157--173

....traffic, an input may be connected to multiple outputs requiring a list of outputs to configure an input. Because an output can still receive a cell from at most one input, it is still sufficient to indicate to each output which input is connected to. 1 N log CHAPTER 1 Introduction 8 [29] [30], 31] 44] 45] This is not generally possible with input queued switches due to variations in delay caused by contention for the switching fabric and queueing at the input.The main disadvantage of output queueing is that for a N port switch, the internal interconnect and output queues must ....

Keshav, S. "On the efficient implementation of fair queueing," Internetworking: Research and Experience, Sept. 1991, vol.2, (no.3):157-73.

.... has been done to develop the right scheduling models for the operating system to support real time video and audio applications [12] In the area of computer networks, there is work in many areas, admission control [13, 14, 15] routing [16, 17] resource reservation [18, 19] packet scheduling [20, 21], protocols [22] and network architectures [23, 13, 24, 25, 26] to name a few. I.B Group Communication An area receiving much attention nowadays is the field of computer supported cooperative work (CSCW) With the improvements in communication, it takes only a small leap of imagination to ....

S. Keshav, "On the efficient implementation of fair queueing," Journal of Internetworking, vol. 2, pp. 157--173, Sep. 1991.

.... Flow isolation is based on per flow queueing in concert with classifiers assigning the incoming packet to its flow and schedulers distributing the bandwidth of the outgoing link in a well defined manner [6] Scheduling mechanisms have been widely investigated in the past [9] 10] 11] 12] [13] [14] However, although flow isolation may solve the problem of unfairness between responsive and non responsive flows, advanced queue management is still needed to avoid bias against bursty traffic and global synchronization. Additionally, the computational cost of per flow scheduling is not ....

S. Keshav, "On the efficient implementation of Fair Queueing"

.... of Weighted Fair Queueing, referred to as Worst case Weighted Fair Queueing (W 2 FQ) was recently proposed to improve even further the fairness properties of the algorithm [11] In both approaches, a maximum of V events may be triggered in the GPS simulator during the transmission of one packet [57]. Thus, the process overhead for completing a scheduling decision is O(V ) making the implementation of the algorithm prohibitively expensive in most cases. Self Clocked Fair Queueing In order to reduce the complexity of Weighted Fair Queueing, an approximate implementation was proposed in [25, ....

S. Keshav, "On the efficient implementation of fair queueing," Internetworking: Research and Experience, vol. 2, pp. 157--173, September 1991.

....item that it needs. It is important to minimize the access time so as to decrease 1 the idle time at the client [5, 15, 25, 24, 26, 27, 19, 18, 3, 2, 38, 39] This paper makes three contributions: ffl We observe that the problem of broadcast scheduling is related to packet fair queueing [10, 28, 30]. While obvious in the hindsight, this observation has not been exploited before to design efficient broadcasting algorithms. ffl Based on the above observation, we present a O(log M) broadcast scheduling algorithm, where M is the number of information items. Simulations show that this algorithm ....

S. Keshav, "On the efficient implementation of fair queueing," Journal of Internetworking: Research and Experience, vol. 2, pp. 57--73, September 1991.

....flow found, the packet is marked for best effort handling. Beside the classifier, we need a more sophisticated scheduler than a simple FIFO mechanism. This scheduler must handle the different flows according to the reservation made for that flow. Examples of such schedulers could be found in [15] [9] [10] For the INRIA Classifier IP Forward IP Output RSVP Daemon Application KernelUser Space Space Setuptool From last Router To next Router Admission Control Interface Figure 1: Architecture of the INRIA Traffic Control Kernel Traffic Control kernel, we implemented a variant ....

Keshav, S., On the Efficient Implementation of Fair Queueing", Journal of Internetworking: Research and Experience, V2, N3, September 1991.

....packets to transmit next (scheduling management) and which packets to drop (buffer management) The simplest queue discipline is FIFO (first in first out) which results in best effort transmission. To accomplish QoS guarantees, a combination of packet scheduling such as fair weighted queuing [52] and traffic shaping at the edge of the network (e.g. token bucket with leaky bucket rate control) is necessary [77] This paper will not deal with resource management in the data transmission context. In network storage, resource management means deciding which data objects to keep in memory, ....

S. Keshav, "On the efficient implementation of fair queueing," Internetworking: Research and Experiences, vol. 2, pp. 157-173, 1991. 25

....29 References . 29 4 1 Fair Queuing In recent years, much research has been performed on the subject of packet fair queuing (PFQ) 2] 3] 9] [12], 22] 23] 26] Consider the system shown in Figure 1 where a node maintains several queues (flows 1 ) which store packets to be transmitted on an output link. A fair queuing algorithm is used to determine which queue to serve next, so as to satisfy a certain fairness criterion 2 . flow ....

S. Keshav, "On the Efficient Implementation of Fair Queueing," Journal of Internetworking: Research and Experience, vol. 2, no. 3, pp. 57-73 (September 1991)

....implementation that can be used to implement the above algorithm. Their implementation is apparently presented in [5] however, we are unable to obtain a copy of [5] at this time. It is possible that their implementation of fair queueing is analogous to the implementation summarized below. Keshav [17] also presents a heap based implementation of fair queueing. However, his fair queueing algorithm is somewhat different from that in [6] We maintain two binary heaps, HB and HC . Heap HB has item with smallest B i value, among all its items, at its root. HC has item with smallest C i value, ....

S. Keshav, "On the efficient implementation of fair queueing," Journal of Internetworking: Research and Experience, vol. 2, September 1991.

....Section 5 concludes this paper with a summary of major research results and an outline of on going work. 2 Related Work Packetized weighted fair queueing is based upon the fluid processor sharing model [3] WFQ has been discussed in [1] and an implementation and simulation study is reported in [2]. 2] discusses the performance cost of various data structures for the priority sorting required in WFQ and mainly considers insertion costs with varying number of connections and buffers. However, besides insertion costs into the appropriate data structure, another source of overhead in WFQ is ....

....5 concludes this paper with a summary of major research results and an outline of on going work. 2 Related Work Packetized weighted fair queueing is based upon the fluid processor sharing model [3] WFQ has been discussed in [1] and an implementation and simulation study is reported in [2] [2] discusses the performance cost of various data structures for the priority sorting required in WFQ and mainly considers insertion costs with varying number of connections and buffers. However, besides insertion costs into the appropriate data structure, another source of overhead in WFQ is the ....

Keshav, S., On the Efficient Implementation of Fair Queueing, Journal of Internetworking: Research and Experience, V2, N3, September 1991.

.... correction [2] and co ordination over multiple related connections [3] In communications systems, protocol support such as end to end QoS negotiation, renegotiation and indication of QoS degradation are required [4] In networks, suitable resource reservation protocols [5] service disciplines [6], flow filtering mechanisms [7] 8] are needed. The QoS A also provides a framework for the maintenance and management of QoS over all system layers. This includes management functions such as admission control for new connections and monitoring to ensure that QoS levels are being maintained by the ....

....delivery parameter remains part of our delta service contract only since it is dependent upon the type of scheduling adopted. In our implementation we have selected the earliest deadline first (EDF) scheduling discipline for the end systems and are experimenting with EDF and weighted fair queueing [6] in our ORL ATM swicthes. Promoting the delivery parameter to the generalised service contract is for further study in the QoS A project. 6. CONCLUSION AND FUTURE WORK In this paper we have described in detail our QoS architecture with particular emphasis on the enhanced transport service ....

[Article contains additional citation context not shown here]

Keshav, S., "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experiences, Vol. 2, pp 157-173, 1991

.... Packet Pair Scheme [7] BUC gateways require per conversation queues at outputports, scheduled by a fair round robin mechanism. Packets arriving at a single per conversation queue are served in FIFO order. Fair Queuing, the particular mechanism used in this paper is investigated in [5] 10] [11] and [12] Fair queueing emulates bit by bit round robin (i.e. transmission of one bit of per conversation queue one, then one bit of perconversation queue two, and so on) by computing a so called bid number for each packet. The packet to be transmitted is the one with the minimum bid number. ....

S. Keshav, "On the efficient implementation of Fair Queueing"

....from queues require a time complexity of O(log(n) While this scheme guarantees absolute fairness [GM90] the packet processing cost makes it hard to implement cheaply at high speeds. A naive FQ server would require O(log(m) where m is the number of packets in the router. However Keshav [Kes91] has shown that only one entry per flow need be inserted into a sorted queue. This still results in O(log(n) overhead. Keshav s other implementation ideas [Kes91] take at least O(log(n) time in the worst case. Stochastic Fair Queuing (SFQ) SFQ was proposed by McKenney [McK91] to address the ....

....cheaply at high speeds. A naive FQ server would require O(log(m) where m is the number of packets in the router. However Keshav [Kes91] has shown that only one entry per flow need be inserted into a sorted queue. This still results in O(log(n) overhead. Keshav s other implementation ideas [Kes91] take at least O(log(n) time in the worst case. Stochastic Fair Queuing (SFQ) SFQ was proposed by McKenney [McK91] to address the inefficiencies of Nagle s algorithm. McKenney uses a hashing scheme to map incoming packets to corresponding queues. Normally, one would use hashing with chaining to ....

Srinivasan Keshav. On the efficient implementation of fair queueing. In Internetworking: Research and Experience Vol.2, 157-173, September 1991.

....item that it needs. It is important to minimize the access time so as to decrease the idle time at the client [4, 9, 15, 14, 17, 18, 11, 10, 3, 2, 31, 32] This paper makes three contributions: ffl We observe that the problem of broadcast scheduling is closely related to packet fair queueing [6, 19, 21]. While obvious in the hindsight, this observation has not been exploited before to design efficient broadcasting algorithms. ffl Basedon the above observation, we present a O(log M) broadcast scheduling algorithm, where M is the number of information items. Simulations show that this algorithm ....

S. Keshav, "On the efficient implementation of fair queueing," Journal of Internetworking: Research and Experience, vol. 2, September 1991.

....value of its head of line packet. For networks of the future, we envision that a high speed channel will be shared by hundreds, perhaps, thousands of flows, which would require a highly efficient search algorithm. Towards this goal, several heap search algorithms have been designed and studied [7, 12]. We propose the idea of group priority to substantially reduce the work of a packet scheduler in updating heaps (or any other sorted priority data structure) The basic idea is simple. Consecutive packets in a flow are partitioned into groups. Each group is given a priority value equal to the ....

Srinivasan Keshav. On the efficient implementation of fair queueing. Journal of Internetworking Research and Experience, 1991.

....of traffic shaping at the edge of the network and scheduling in the network can provide hard performance guarantees. Parekh [ Parekh,93] has shown that if a source flow is shaped by a token bucket with leaky bucket rate control and scheduled by the weighted fair queueing service discipline [Keshav,91] it is possible to achieve strong guarantees on delay. This is the theory behind int serv s guaranteed delay service soon to be offered in an integrated services Internet [Braden,94] 2.4.2.2 Flow Scheduling Flow scheduling manages the forwarding of flows (chunks of media based on application ....

.... service disciplines such as FIFO taking advantage of statistical multiplexing gain; iii) a bounding approach, which takes into account any distortion of the flow as it traverses work conserving multiplexors (e.g. packetised generalised processor sharing [Parekh,93] and weighted fair queueing [Keshav,91] resulting in mathematically provable performance bounds for statistical and deterministic service guarantees [Cruz,91] and finally (iv) an observation based approach, which uses measured behaviour (e.g. COMET s approach [Hyman,90] and Clark s predictive service [Clark,92] of the aggregate ....

Keshav, S., "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experiences, Vol. 2, pp 157-173, 1991.

....of traffic shaping at the edge of the network and scheduling in the network can provide hard performance guarantees. Parekh [ Parekh,93] has shown that if a source flow is shaped by a token bucket with leaky bucket rate control and scheduled by the weighted fair queueing service discipline [Keshav,91] it is possible to achieve strong guarantees on delay. This is the theory behind int serv s guaranteed delay service soon to be offered in an integrated services Internet [Braden,94] 2.4.2.2 Flow Scheduling Flow scheduling manages the forwarding of flows (chunks of media based on application ....

.... service disciplines such as FIFO taking advantage of statistical multiplexing gain; iii) a bounding approach, which takes into account any distortion of the flow as it traverses work conserving multiplexors (e.g. packetised generalised processor sharing [Parekh,93] and weighted fair queueing [Keshav,91] resulting in mathematically provable performance bounds for statistical and deterministic service guarantees [Cruz,91] and finally (iv) an observation based approach, which uses measured behaviour (e.g. COMET s approach [Hyman,90] and Clark s predictive service [Clark,92] of the aggregate ....

Keshav, S., "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experiences, Vol. 2, pp 157-173, 1991

....that the combination of traffic shaping at the edge of the network and scheduling in the network can provide hard performance guarantees. Parekh [17] has shown that if a source is shaped by a token bucket with leaky bucket rate control and scheduled by the weighted fair queueing service discipline [18], it is possible to achieve strong guarantees on delay; flow scheduling manages the forwarding of flows (chunks of media based on application layer framing) in the end system [19] 20] 21] and network (packets and or cells) in an integrated manner [22] Flows are generally scheduled independently ....

.... simple service disciplines such as FIFO taking advantage of statistical multiplexing gain; iii) a bounding approach, which takes into any account distortion of the flow as traverses work conserving multiplexers (e.g. packetised generalised processor sharing [17] and weighted fair queue ing [18]) resulting in mathematically provable performance bounds for statistical and deterministic service guarantees [55] and finally (iv) an observation based approach, which uses measured behaviour (e.g. COMET s approach [56] and Clark s predictive service [13] of the aggregate traffic and the user ....

Keshav, S., "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experiences, Vol. 2, pp 157173, 1991.

....implementation that can be used to implement the above algorithm. Their implementation is apparently presented in [5] however, we are unable to obtain a copy of [5] at this time. It is possible that their implementation of fair queueing is analogous to the implementation summarized below. Keshav [17] also presents a heap based implementation of fair queueing. However, his fair queueing algorithm is somewhat different from that in [6] We maintain two binary heaps, HB and HC . Heap HB has item with smallest B i value, among all its items, at its root. HC has item with smallest C i value, ....

S. Keshav, "On the efficient implementation of fair queueing," Journal of Internetworking: Research and Experience, vol. 2, September 1991.

.... [Campbell,92a] In communications systems, protocol support such as end to end QoS negotiation, renegotiation and indication of QoS degradation are required [Boerjan,92] In networks, suitable resource reservation protocols [Zhang,93] and service disciplines in switch queues are needed [Keshav,91] Parekh,92] Hyman,93] The QoS A also provides a framework for the maintenance and management of QoS over all system layers. This includes management functions such as admission control for new connections and monitoring to ensure that QoS levels are being maintained by the service provider. ....

....parameter remains part of our delta service contract only since it is dependent upon the type of scheduling adopted. In our implementation we have selected the earliest deadline first (EDF) scheduling discipline for the end systems and are experimenting with EDF and weighted fair queueing [Keshav,91] in our ORL ATM swicthes. Promoting the delivery parameter to the generalised service contract is for further study in the QoS A project. 6. Related Work There is currently very little literature available on the integrated treatment of QoS across all architectural layers. One early contribution ....

[Article contains additional citation context not shown here]

Keshav, S., "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experiences, Vol. 2, pp 157-173, 1991

....a sorted queue require O(log(n) time, where n is the number of flows. While the BR guarantees fairness [GM90] the packet processing cost makes it hard to implement cheaply at high speeds. A naive FQ server would require O(log(m) where m is the number of packets in the router. However Keshav [Kes91] shows that only one entry per flow need be inserted into a sorted queue. This still results in O(log(n) overhead. Keshav s other implementation ideas [Kes91] take at least O(log(n) time in the worst case. Stochastic Fair Queuing (SFQ) SFQ was proposed by McKenney [McK91] to address the ....

....implement cheaply at high speeds. A naive FQ server would require O(log(m) where m is the number of packets in the router. However Keshav [Kes91] shows that only one entry per flow need be inserted into a sorted queue. This still results in O(log(n) overhead. Keshav s other implementation ideas [Kes91] take at least O(log(n) time in the worst case. Stochastic Fair Queuing (SFQ) SFQ was proposed by McKenney [McK91] to address the inefficiencies of Nagle s algorithm. McKenney uses hashing to map packets to corresponding queues. Normally, one would use hashing with chaining to map the flow ID in ....

S. Keshav. On the efficient implementation of fair queueing. In Internetworking: Research and Experience Vol.2, 157-173, September 1991.

.... a cell for transmission, the server picks the cell with the smallest virtual finishing time F , among the cells already queued for service (breaking ties arbitrarily) Computing R(t) is complex, and indeed is similar to simulating multiple GPS service events between consecutive cell departures [20,21]. Self clocked fair queueing (SCFQ) 8,9] uses the virtual finishing time F serv of the cell currently in service as an estimate of current virtual time, thereby avoiding the burden of computing R 0 (t) Analysis [9] and simulation [8] suggest that SCFQ retains the throughput fairness ....

....reschedules it for later service. As a result, the FLB shaper ranks cells by their virtual finishing times in the fair queueing system, instead of sorting cells by conformance times. Although fair queueing implementations typically require a priority queue to sort cells by virtual finishing times [20], the FLB shaper implements self clocked fair queueing using two sets of FIFO queues. 4.1 Per Connection Queueing Traffic shapers in high speed networks require efficient hardware mechanisms for arbitrating amongst competing cells. A shaper can reduce sorting complexity by considering only the ....

S. Keshav, "On the efficient implementation of fair queueing," J. Internetworking: Research and Experience, vol. 2, pp. 157--173, Sept. 1991.

No context found.

S. Keshav, "On the Efficient Implementation of Fair Queueing", Internetworking: Research and Experience, vol. 2, January 1991, pp. 113--131.

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