P. McKenney, "Stochastic Fairness Queueing", Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....1. DRR with Longest Queue Drop Our first policy combined DRR with packet discard from the longest active queue. For the rest of the paper, we refer to this policy as plain DRR or DRR, since this packet discard policy is part of the original DRR algorithm [8] and was first proposed by McKenney in [9]. In [7] we present more details and motivation for developing the DRR variations presented below. 2. Throughput DRR (TDRR) In this algorithm, we store a throughput value associated with each DRR queue. The throughput parameter is maintained as an exponentially weighted average and is used in ....

P. McKenney, "Stochastic Fairness Queueing", Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.

....1. DRR with Longest Queue Drop Our first policy combined DRR with packet discard from the longest active queue. For the rest of the paper, we refer to this policy as plain DRR or DRR, since this packet discard policy is part of the original DRR algorithm [7] and was first proposed by McKenney in [8]. Through our simulation study, we found that plain DRR was not very effective in utilizing link bandwidth or providing fair sharing among competing TCP flows over a single bottleneck link. DRR did perform significantly better than RED and Blue when there were TCP flows with different RTTs or the ....

....Algorithms Several scheduling algorithms are known in the literature for bandwidth allocation and transmission scheduling. These include the packet by packet version of Generalized Processor Sharing [9] also known as Weighted Fair Queueing [10] VirtualClock [11] Stochastic Fairness Queueing [8], SelfClocked Fair Queueing [12] Weighted Round Robin [13] Deficit Round Robin (DRR) 7] and Frame based Fair Queueing [14] We chose DRR due to its simplicity and ease of implementation in hardware. 5.2 FRED One proposal for using RED mechanisms to provide fairness is Flow RED (FRED) 15] ....

P. McKenney, "Stochastic Fairness Queueing", Internetworking: Research and Experience, vol. 2, pp. 113-- 131, Jan. 1991.

....to be implemented in hardware [33] DRR with longest queue dropping is a well known algorithm that drops packets from the longest active queue. For the rest of the paper, we refer to it as plain DRR or DRR since longest queue dropping was part of the original algorithm first proposed by McKenney [34]. TDRR drops packets from the highest throughput active queue. Some of our results for QSDRR and RED are presented below to illustrate the effectiveness of QSDRR. 7.2. Simulation Environment We present some of the performance results below. Table 2 shows the algorithm independent parameters. See ....

P. McKenney, "Stochastic Fairness Queueing," Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.

....to be implemented in hardware [36] DRR with longest queue dropping is a wellknown algorithm that drops packets from the longest active queue. For the rest of the paper, we refer to it as plain DRR or DRR since longest queue dropping was part of the original algorithm first proposed by McKenney [37]. TDRR drops packets from the highest throughput active queue. Some of our results for QSDRR and RED are presented below to illustrate the effectiveness of QSDRR. 7.2. Simulation Environment We present some of the performance results below. Table 2 shows the algorithm independent parameters. See ....

P. McKenney, "Stochastic Fairness Queueing," Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.

....1. DRR with Longest Queue Drop Our first policy combined DRR with packet discard from the longest active queue. For the rest of the paper, we refer to this policy as plain DRR or DRR, since this packet discard policy is part of the original DRR algorithm [7] and was first proposed by McKenney in [8]. In [9] we present more details and motivation for developing the DRR variations presented below. 2. Throughput DRR (TDRR) In this algorithm, we store a throughput value associated with each DRR queue. The throughput parameter is maintained as an exponentially weighted average and is used in ....

P. McKenney, "Stochastic Fairness Queueing", Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.

....each associated with one level of the accounting bins. Each hash function maps a flow into one of the N accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [20] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability p m as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the N bins in each of the L levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

....ID (Source address, Destination address, Source port, Destination port, Protocol) into one of the N accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [24] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability pm as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the N bins in each of the L levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

....each associated with one level of the accounting bins. Each hash function maps a flow into one of the N accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [11] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability p m as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the N bins in each of the L levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

....capacity of all its attached links. MSS: Maximum Segment Size. MSS is flow specific and can be estimated by maintaining flow state where the segment sizes are averaged over the duration of the flow. Flow state can be maintained in a method that is similar to Stochastic F airness Queueing [8], whereby each flow is hashed onto a particular memory location. Hash collisions are ignored, thus keeping the time required for reading and updating flow state at 0(1) Alternatively, we can calculate this on a per packet basis assuming that most incoming packets of a flow are likely to be of the ....

P. McKenney. Stochastic Fairness Queueing. In Proceedings of IEEE INFOCOM, March 1990.

....each associated with one level of the accounting bins. Each hash function maps a flow into one of the N accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [11] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability pm as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the N bins in each of the L levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

....1. DRR with Longest Queue Drop Our first policy combined DRR with packet discard from the longest active queue. For the rest of the paper, we refer to this policy as plain DRR or DRR, since this packet discard policy is part of the original DRR algorithm [7] and was first proposed by McKenney in [8]. Through our simulation study, we found that plain DRR was not very effective in utilizing link bandwidth or providing fair sharing among competing TCP flows over a single bottleneck link. DRR did perform significantly better than RED and Blue when there were TCP flows with different RTTs or the ....

....Algorithms Several scheduling algorithms are known in the literature for bandwidth allocation and transmission scheduling. These include the packet by packet version of Generalized Processor Sharing [9] also known as Weighted Fair Queueing [10] VirtualClock [11] Stochastic Fairness Queueing [8], SelfClocked Fair Queueing [12] Weighted Round Robin [13] Deficit Round Robin (DRR) 7] Framebased Fair Queueing [14] We chose DRR due to its simplicity and ease of implementation in hardware. 5.2. FRED One proposal for using RED mechanisms to provide fairness is Flow RED (FRED) 15] The ....

P. McKenney, "Stochastic Fairness Queueing", Internetworking: Research and Experience, vol. 2, pp. 113--131, Jan. 1991.

....discipline that assigns an independent queue for each flow. WFQ can provide fair bandwidth allocation in times of congestion, and protects a flow from other flows. A weight can be assigned to each queue to give a different proportion of the network capacity. SFQ SFQ (Stochastic Fairness Queueing) [10] is an approximation of WFQ. WFQ is difficult to implement because a large number of queues are required as the number of flows increases. In SFQ, a hash function is used to map a flow to one of a fixed set of queues, and thus, it is possible for two different flows to be mapped into the same ....

P. E. McKenney. Stochastic fairness queueing. In Proceedings of INFOCOM, San Francisco, California, June 1990.

....Queues and schedulers are indistinguishable. We can exploit this property to build virtual queues, compound elements that act like queues but implement more complex behavior than FIFO queueing. The compound element in Figure 4. 4 is a virtual queue that implements stochastic fairness queueing [26]: packets are hashed into one of several queues that are scheduled round robin, providing some isolation between competing flows. It has one push input and one pull output, just like a Queue, so it can generally be used as a drop in replacement for Queue. 4.4 dropping policies The Queue element ....

P. E. McKenney. Stochastic fairness queueing. In Proc. IEEE Infocom, volume 2, pages 733--740, June 1990.

....the GPS discipline, which has some desirable characteristics but, since it uses an idealised fluid model, it cannot be implemented in the real world 4 . WFQ ALTQ is not a true implementation of the Weighted Fair Queuing discipline. It is, in fact, closer to a variant named Sthocastic Fair Queing [McKenney90]. Most important for this analysis is the fact that the discipline used in WFQ ALTQ, as well as in the WFQ 4 WF2Q is presented in [Bennett96] has having better characteristics than WFQ. The only difference is that packets are chosen not from all the queued packets but, instead, from those ....

P. McKenney, Stochastic Fairness Queueing, in Proceedings of IEEE INFOCOM, San Francisco, California, June 1990

....large source destination populations, since the state in a Fair Queueing implementation scales with the number of active end to end paths, which will be high in backbone gateways, for example. 3.4.2. Stochastic Fairness Queuing Stochastic Fairness Queueing (SFQ) has been suggested as a technique [McK90] to address the implementation issues relating to Fair Queueing. The first overhead that is reduced is that of looking up the source destination address pair in an incoming packet and determining which queue that packet will have to be placed in. SFQ does not require as many memory accesses as ....

....address pair in an incoming packet and determining which queue that packet will have to be placed in. SFQ does not require as many memory accesses as Fair Queueing to place the packet in the appropriate queue. SFQ is thus claimed to be more amenable to implementation for high speed networks [McK90]. SFQ uses a simple hash function to map from the source destination address pair to a fixed set of queues. Since the assignment of an address pair to a queue is probabilistic, there is the likelihood of multiple address pairs colliding and mapping to the same queue. This would potentially degrade ....

[Article contains additional citation context not shown here]

P. McKenney, Stochastic Fairness Queueing. To appear in Proceedings of INFOCOM '90.

....way to the global satisfaction of IP service consumers and providers. The multiple class best effort proposal requires an integrated approach to scheduling and queue management. Although significant work exists in each of these fields (for instance, WFQ, W2FQ or Stochastic Fair Queuing [4] [19], in the area of packet scheduling, and RED and BLUE [12] 11] for queue management) these approaches do not work in an integrated way, in addition to having several known limitations and problems [18] 6] 27] At the Laboratory of Communications and Telematics of UC (LCT UC) the authors are ....

P. McKenney, Stochastic Fairness Queueing, in Proceedings of IEEE INFOCOM, San Francisco, California, June 1990.

....incentives for end to end congestion control. This middle range would also include FCFS scheduling with differential dropping for flows using a disproportionate share of the bandwidth [LM96] or scheduling mechanisms such as ClassBased Queueing (CBQ) FJ95] or Stochastic Fair Queueing (SFQ) McK90] that can operate on levels of granularity between the two extremes of either a single flow or the entire aggregate of best effort traffic. The differential treatment of unresponsive flows can consist of preferentially dropping packets from unresponsive flows while keeping those packets in the ....

P. McKenney. "Stochastic Fairness Queueing". Proceedings of the Conference on Computer Communications (IEEE Infocom), 1990.

....each associated with one level of the accounting bins. Each hash function maps a flow into one of the N accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [47] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability p m as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the N bins in each of the L levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

.... for flow isolation: 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 ....

McKenney, P., "Stochastic Fairness Queueing", INFOCOM '90

....downstream, Queues and schedulers are indistinguishable. We can exploit this property to build virtual queues, compound elements that act like queues but implement more complex behavior than FIFO queueing. Figure 10 shows a virtual queue that implements a version of stochastic fairness queueing [McKenney 1990]: packets are hashed into one of several queues that are scheduled round robin, providing some isolation between competing flows. The IP router can be extended to use stochastic fairness queueing by replacing its Queues with copies of Figure 10. 4.2 Dropping policies The Queue element implements ....

McKenney, P. E. 1990. Stochastic fairness queueing. In Proc. IEEE Infocom, Volume 2 (June 1990), pp. 733--740.

....each associated with one level of the accounting bins. Each hash function maps a flow into one of the W accounting bins in that level. The accounting bins are used to keep track of queue occupancy statistics of packets belonging to a particular bin. This is in contrast to Stochastic Fair Queueing [20] (SFQ) where the hash function maps flows into separate queues. Each bin in SFB keeps a marking dropping probability M N as in BLUE, which is updated based on bin occupancy. As a packet arrives at the queue, it is hashed into one of the W bins in each of the = levels. If the number of packets ....

P. McKenney. Stochastic Fairness Queueing. In Proc. IEEE INFOCOM, March 1990.

....requires storage of per flow state and a flow lookup in ingress and egress routers and not in core routers. Another idea to identify unfriendly flows has been proposed in [19] A alternative approach to identification of unfriendly flows is to allocate a fairshare to each flow (see [1] 12] 13][18][22] and others) However, merely restricting unfriendly flows to their fair share does not necessarily create an incentive for users to implement end to end congestion control. Additionally, many unresponsive best effort flows restricted to their fair share may still cause congestion collapse as ....

McKenney, P., "Stochastic Fairness Queueing", Proc. of IEEE INFOCOM, 1990

....schedulers are indistinguishable. We can exploit this property to build virtual queues, compound elements that look exactly like queues from the outside but implement more complex behavior than FIFO queueing. Figure 10 shows a virtual queue that implements a version of Stochastic Fairness Queueing [15]: packets are hashed by flow identifier into one of several queues that are scheduled round robin, providing some isolation between competing flows. 4.2 Dropping policies The Queue element implements a simple dropping policy, namely a configurable maximum length beyond which all packets are ....

P. E. McKenney. Stochastic fairness queueing. In Proc. IEEE Infocom, volume 2, pages 733--740, June 1990.

....Keshav s 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 ....

....may only be implemented considering the network topology, for instance in gateways located at the edge of an isolated LAN. The use of a hash function is suggested for address lookup (i.e. matching of packets to their corresponding perconversation queues) Efficient hashing schemes are proposed in [10] and [18] An interaction of RFCN and the BUC algorithm with the address lookup scheme has to be considered: Some hashing schemes may generate more than one queue per conversation at an output port in case of collusions. If this happens, the BUC algorithm has to calculate only one window based on ....

McKenney, P., "Stochastic Fairness Queueing", INFOCOM '90

No context found.

- Paul McKenney. Stochastic Fairness Queueing. In Proceedings of the INFOCOM90 Conference, IEEE, pp. 733-740, San Francisco, USA, June 1990.

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