Z. Wang and J. Crowcroft. Eliminating periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communications Review, Apr. 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....an unreachable, down or unresponsive server, and cannot exploit the existence of an alternate server. In addition, loss of a packet triggers TCP s response to congestion using two well known mechanisms: fast retransmit and slow start [15] Research e orts have focused on improving these mechanisms [8, 12, 14, 19, 21, 29, 30], without altering their underlying assumption, i.e. that packet loss is caused by and is an indication of congestion in the core network. These observations point to the fact that TCP is overly limiting, as the end user may be more concerned with availability and or quality of the service rather ....

Z. Wang and J. Crowcroft. Eliminating periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communications Review, Apr. 1992.

....the flow of credits to a value of Wc, e.g. when processing the data takes longer than transmitting them, the transmitting maximum window size will be the minimum of both W and Wc. The idea of using the delay gradient as a feedback signal was also proposed by Zheng and Crowcroft who described in [21] an algorithm similar to the DA: a delay threshold Di = D q Q, axM t is defined, where D is the round trip propagation delay i.e. without queueing) Q,ax is the maximum queue length, M the packet size and the bottleneck capacity. Di can be rewritten as (1 oe)Dmi,zq oeDmax where Dmi,z = Dp and ....

Wang Z., Crowcroft J., Eliminating Periodic Packet Losses in the 4.3 Tahoe BSD TCP Congestion Control Algorithm, CCR, Volume 22, Number 2, April 1992.

....(this is done after a slow start or while recovering from a packet loss) These mechanisms are some of the salient and noteworthy features of TCP that has led to its wide application. Since these algorithms and mechanisms were introduced there have been several efforts to improve their behavior [13, 14, 15, 16, 17, 18, 19]. The important thing to note is that in all the cases above a packet loss is being treated as a loss of the packet in the network due to congestion, whereas the timeout for a packet could have occurred because the packet was never accepted due to server overload or because of a DoS attack. The ....

Z. Wang and J. Crowcroft. Eliminating periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communications Review, Apr. 1992.

....AEow of credits to a value of W c , e.g. when processing the data takes longer than transmitting them, the transmitting maximum window size will be the minimum of both W and W c . The idea of using the delay gradient as a feedback signal was also proposed by Zheng and Crowcroft who described in [26] an algorithm similar to the DA: a delay 1 A modied Delay Algorithm is described in chapter 5 4 which handles such changes. 8 threshold D i = D p ffQ max M= is dened, where D p is the round trip propagation delay i.e. without queuing) Q max is the maximum queue length, M the packet size and ....

Wang Z., Crowcroft J., Eliminating Periodic Losses in the 4.3 Tahoe BSD TCP Congestion Control Algorithm, Draft paper, zwang@cs.ucl.ac.uk.

....[10] Explicit Congestion Notication (ECN) 11] ATM ABR service class) signals from the network or on network responses to variations in source behavior. In the second case, mainly two adjustments methods are used to control the amount of data sent into the network : rate based and window based [15, 20, 6, 21]. The rate based scheme regulates the inter packet sending time, while the window based scheme operates by increasing or decreasing the window size. Two approaches can be distinguished for the latter scheme : delay based (based on the estimation of the RTT) and throughput based (based on the ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22:9 16, April 1992.

....development is given in [7] These changes can be seen in Figure 5 in the time interval from 4.5 to 7.5 seconds. One change is the increased queue size in the intermediate nodes of the connection, resulting in an increase of the RTT for each successive segment. Wang and Crowcroft s DUAL algorithm [12] is based on this increase of the round trip delay. The congestion window normally increases as in Reno, but every two round trip delays it checks to see if the current RTT is greater than the average of the minimum and maximum RTTs seen so far. If it is, then it decreases the congestion window by ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22(2):9--16, Apr. 1992.

....is given in [16] These changes can be seen in Figure 4.1 in the time interval from 4.5 to 7.5 seconds. One change is the increased queue size in the intermediate nodes of the connection, resulting in an increase of the RTT for each successive segment. Wang and Crowcroft s DUAL algorithm [28] is based on reacting to this increase of the round trip delay. The congestion window normally increases as in Reno, but every two round trip times the algorithm checks to see if the current RTT is greater than the average of the minimum and maximum RTTs seen so far. If it is, then the algorithm ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22(2):9--16, Apr. 1992.

....In current incarnations of TCP, the congestion window follows a sawtooth like pattern where the congestion window is continually increased until packet loss occurs. While this allows TCP to probe for additional bandwidth, such behavior eventually induces packet loss. The idea behind the Tri S [66, 67] and Vegas [7] modifications is to change the congestion avoidance phase so that it only performs its linear increase when the network is not congested. In both algorithms, if the round trip times indicate an increase in delay due to queues being built up in the network, the TCP source either ....

....the equation also provides some insight on how to improve loss rates given a fixed number of connections. One way is to alter the congestion avoidance mechanism itself so that it does not continually increase its sending rate beyond the network s capacity. For example, schemes such as Tri S [66, 67] or TCP Vegas [7] can be used to reduce the amount of packet losses observed. Another way is to increase the bottleneck link bandwidth L. By increasing the link bandwidth, the effective congestion windows of individual connections increases, thus decreasing the frequency of congestion notification ....

[Article contains additional citation context not shown here]

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. Computer Communication Review, 22(2):9--16, April 1992.

....bottleneck link. Therefore, we obtain the maximum allowable transfer rate for a new connection. Once established, STTP uses a proactive congestion avoidance algorithm, which is sensitive to variation in a connection s Round Trip Time (RTT) Similar methods are employed by Tri S [WC91] DUAL [WC92] and TCP Vegas [BOP94] TCP Vegas s congestion control consists of a bounded window within which a connection s throughput is allowed to vary. If the actual throughput falls outside this predicted 2 window, then the transmission rate is adjusted accordingly. For example if the Round Trip Time ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22:9-16, April 1992.

....this relative delay is the foundation of our congestion control algorithm. Our approach is significantly different from rate controlled congestion control approaches, e.g. TCP Vegas [2] as well as those that use an increasing round trip time (RTT) estimate as the primary indication of congestion [22, 18, 21], in that TCP Santa Cruz does not use RTT estimates in any way for congestion control. This represents a fundamental improvement over the latter approaches, because RTT measurements do not permit the sender to differentiate between delay variations due to increases or decreases in the forward or ....

....TCP implementations. Finally, Section 5 summarizes our results. 2 Previous Work Congestion control for TCP is an area of active research; solutions to congestion control for TCP address the problem either at the intermediate routers in the network [8, 13, 6] or at the endpoints of the connection [2, 7, 20, 21, 22]. Router based support for TCP congestion control can be provided through RED gateways [6] a solution in which packets are dropped in a fair manner (based upon probabilities) once the router buffer reaches a predetermined size. As an alternative to dropping packets, an Explicit Congestion ....

[Article contains additional citation context not shown here]

Z. Wang and J. Crowcroft. Eliminating periodic packet losses in the 4.3-Tahoe BSD TCP congestion control algorithm. In Computer Communication Review, volume 22 No. 2, pages 9 -- 16, April, 1992.

....causes back to back packet spikes in Reno s sending process. Vegas tries to enforce uniform spacing between the packets transmitted at a particular window size. Essentially, Vegas limits the peak rate that packets can be sent. Congestion Avoidance: As mentioned above, much current research [7, 19, 18] addresses congestion avoidance. Jain s CARD (Congestion Avoidance using Roundtrip Delay) scheme [7] suggests using throughput over round trip delay curve, which Jain denotes Power, to forecast congestion. When the power curve approaches its peak, it indicates the network is approaching the ....

Zheng Wang and Jon Crowcroft. Eliminating periodic packet losses in the 4.3-Tahoe BSD TCP congestion control algorithm. Computer Communication Rview, 22(2):9--16, April 1992.

....algorithm increases the congestion window by 1 cwnd for each in sequence acknowledgment that is received, resulting in an overall linear increase. Since the introduction and wide scale deployment of these algorithms, there has been considerable effort devoted to improving on their behavior [11, 3, 35, 36, 6, 17, 1, 15]. Most, if not all, of the problems addressed by these extensions can be attributed to TCP s ignorance of prevailing network conditions. In the following section, we describe how this lack of information interacts poorly with today s Internet. 3 Short Flows, Fat Pipes, Frequent Losses Numerous ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in the 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communications Review, 22(2), Apr. 1992.

....development is given in [10] These changes can be seen in Fig. 5 in the time interval from 4.5 to 7.5 seconds. One change is the increased queue size in the intermediate nodes of the connection, resulting in an increase of the RTT for each successive segment. Wang and Crowcroft s DUAL algorithm [17] is based on reacting to this increase of the round trip delay. The congestion window normally increases as in Reno, but every two round trip delays the algorithm checks to see if the current RTT is greater than the average of the minimum and maximum RTTs seen so far. If it is, then the algorithm ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22(2):9--16, Apr. 1992.

....development is given in [7] These changes can be seen in Figure 6 in the time interval from 4.5 to 7.5 seconds. One change is the increased queue size in the intermediate nodes of the connection, resulting in an increase of the RTT for each successive segment. Wang and Crowcroft s DUAL algorithm [11] is based on this increase of the round trip delay. The congestion window normally increases as in Reno, but every two round trip delays the algorithm checks to see if the current RTT is greater than the average of the minimum and maximum RTTs seen so far. If it is, then the algorithm decreases ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 22(2):9--16, Apr. 1992.

....in the case of persistent congestion, where BaseRTT cannot be derived at any stage of the connection. The congestion avoidance scheme of Vegas belongs to the class of delaybased congestion avoidance (DCA) schemes, rst investigated by Jain [19] while Wang and Crowcroft had proposed TCP Dual [32] and Tri S [31] 9 Recently, Martin et al. 23] studied the underlying assumption of DCA algorithms: that there is strong correlation between an increase in delay due to queueing events and packet loss. It was shown that there is only weak correlation, at the level of 7 18 . One interpretation ....

Zheng Wang and Jon Crowcroft. Eliminating Periodic Packet Losses in the 4.3 Tahoe BSD TCP Congestion Control Algorithm. ACM Computer Communication Review, 1992. 21

....congestion control algorithms. While bandwidth based increases provide end hosts with an upper bound on how aggressively they can ramp up their transmission rates, it is often desirable for a source to change its transmission rate more slowly or not at all when nearing the congestion point [2, 25, 26] in order to avoid oscillations inherent in tcp s windowing algorithm. We are currently investigating how to incorporate such techniques into the bandwidth based increase algorithm. We are also exploring additional mechanisms to improve the fairness of the windowing algorithms. One of the ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. Computer Communication Review, 22(2):9--16, April 1992.

....end host congestion control algorithms. While bandwidth based increases provide end hosts with an upper bound on how aggressively they can ramp up their sending rates, it is often desirable for a source to change its sending rate more slowly or not at all when nearing the congestion point [2, 25, 26] in order to avoid oscillations inherent in tcp s windowing algorithm. We are currently investigating incorporating such techniques into the bandwidth based increase algorithm. We are also exploring additional mechanisms to improve the DRAFT fairness of the windowing algorithms. One of the ....

Z. Wang and J. Crowcroft. Eliminating Periodic Packet Losses in 4.3-Tahoe BSD TCP Congestion Control Algorithm. Computer Communication Review, 22(2):9--16, April 1992.

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