L. Zhang, S. Shenker, D.D. Clark, Observations and dynamics of a congestion control algorithm: the e#ects of two-way tra#c, in: ACM SIGCOMM 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....one may think that intermingling of packets belonging to di#erent flows, with the consequent separation of packets belonging to the same batch, would mitigate the strong correlations that we have found in the tra#c produced by the TCP sources. On the other hand, the ack compression phenomenon [22] is expected to exacerbate the burstiness of packet transmissions. Finally, heterogenous link capacities should be taken into account. Whatever happens, it is intuitive to understand that a queue model with instantaneous batch arrivals will always produce a conservative analysis, provided that ....

L. Zhang, S. Shenker and D. D. Clark, "Observations and Dynamics of a Congestion Control Algorithm: The E#ects of Two-Way Tra#c," In ACM SIGCOMM '91, pp. 133--147, 1991

....is in the network. 2) Senders always have data to send and will send as many packets as their windows allow. 3) There is only one bottleneck link that causes queue buildup. 4) Receivers acknowledge every packet received and there are no delayed acknowledgments. 5) There is no ACK compression [15]. 6) The queue length is measured in packets and all packets have the same size. 3 Queue Dynamics with TCP Window Control In this section, we study the relationship between the window size at the source and the queue size at the congested router. The purpose is to show the difference between ....

L. Zhang, S. Shenker, and D. D. Clark, Observations and dynamics of a congestion control algorithm: the effects of two-way traffic, Proc. ACM SIGCOMM '91, pages 133-147, 1991. 17

....from the receiver do not reach the sender in a timely manner, resulting in large transmission bursts. Thus, instead of observing the smooth ACK clocked transmissions characteristic of ideal TCP transfers where data packets are interspersed with ACKs, we observe a severe form of A CK com pression [16]. Here, all the ACKs for a window ar rive in closely spaced bursts, in response to which the sender transmits several data packets in close succession. This transmission pattern degrades performance for two reasons. First, bursts of data tend to couse losses becouse bottleneck routers usuolly do ....

ZHAN, L., SHENKER, S., AND CLARK, D. D. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91 (1991), pp. 133-147. 13

....breaks down in wireless networks where packets are often corrupted by bit errors. Second, the design of TCP algorithms heretofore has primarily focussed on the network path characteristics on the direction of data transfer. The flow of TCP acknowledgments has received only limited attention (e.g. [23]) However, in the presence of bandwidth and other types of asymmetry, the constrained flow of acks can have a significant adverse effect on performance. Finally, TCP s window growth and data driven loss recovery algorithms have been designed with long data transfers in mind. Their effectiveness ....

L. Zhang, S. Shenker, and D. D. Clark. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91, pages 133--

....from the receiver do not reach the sender in a timely manner, resulting in large transmission bursts. Thus, instead of observing the smooth ACK clocked transmissions characteristic of ideal TCP transfers where data packets are interspersed with ACKs, we observe a severe form of ACK compression [16]. Here, all the ACKs for a window arrive in closely spaced bursts, in response to which the sender transmits several data packets in close succession. This transmission pattern degrades performance for two reasons. First, bursts of data tend 2 to cause losses because bottleneck routers usually do ....

Zhang, L., Shenker, S., and Clark, D. D. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Trac. In Proc. ACM SIGCOMM '91 (1991), pp. 133-147. 13

....the link can support (i.e. when k 1, assuming every data packet is acknowledged) they get queued behind one another. The spacing between them when they emerge from the link is dilated with respect to their original spacing. This is in contrast to the widely studied ack compression phenomenon ([110], 73] which happens when acks get queued at a fast link, i.e. k 1) Thus, the sender clocks out new data at a slower rate than if there had been no queuing of acks. Another consequence is that the growth of the sender s window size slows down. This is part of the reason why the downstream ....

L. Zhang, S. Shenker, and D. D. Clark. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91, pages 133--147, 1991.

....heterogeneity in network technologies and characteristics. The specific measurements reported in this paper were taken over a wireless cable modem network and a packet radio network. There has been some previous work on understanding the effects of two way traffic on TCP performance. In [20], the authors demonstrate how two way traffic can lead to ack compression, where closely bunched acknowledgments disrupt the smooth ack clocked transmission at the sender. More recently, there has been interest in how asymmetric bandwidth networks exacerbate this problem. In [16] the authors ....

L. Zhang, S. Shenker, and D. D. Clark. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91, pages 133--147, 1991.

....Connection Behavior: 1. Loss recovery: How effective are TCP s fast retransmission and recovery mechanism in avoiding timeouts 2. Receiver bottleneck: How often is a TCP receiver s advertised window size the limiting factor for performance 3. Ack compression: How often does ack compression [11,18] occur and how does it impact the packet loss rate Parallel Connection Behavior: 1. Throughput: How does the throughput seen by a client vary with the number of connections it has open to a server 2. Congestion control: How does a set of parallel connections as a whole respond to ....

....a larger window) From this analysis, we make the following recommendation. Future network implementations should increase their default socket buffer size to avoid the receiver window from becoming a bottleneck. Default values of 4 KB are often too small. 3.1. 3 Ack Compression Ack compression [18] occurs when the spacing between successive acknowledgments is compressed while they are in transit between the receiver and sender. The acknowledgments then arrive at the sender at a higher rate than they were generated by the receiver. This disturbs the ack clocking nature of TCP [10] causing ....

L. Zhang, S. Shenker, and D. D. Clark. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91, pages 133--147, 1991.

....restrict the performance of a connection Are typical receiver advertised window sizes large enough to utilize the more sophisticated loss recovery mechanisms Are the effective window sizes available to senders limited by small receiver advertised window sizes 3. Ack compression: Ack compression [11,20] occurs when the spacing of acknowledgments is compressed, disrupting the ack clocking nature of TCP. How often does this phenomenon occur How is it correlated with congestion related phenomena such as packet losses Parallel Connection Behavior: 1. Throughput: How does the throughput seen by a ....

....a larger window) From this analysis, we make the following recommendation. Future network implementations should increase their default socket buffer size to avoid the receiver window from becoming a bottleneck. Default values of 4 KB are often too small. 3.1. 3 Ack Compression Ack compression [20] occurs when the spacing between successive acknowledgments is compressed while they are in transit between the receiver and sender. The acknowledgments then arrive at the sender at a higher rate than they were generated by the receiver. This disturbs the ack clocking nature of TCP [9] causing ....

L. Zhang, S. Shenker, and D. D. Clark. Observations and Dynamics of a Congestion Control Algorithm: The Effects of Two-Way Traffic. In Proc. ACM SIGCOMM '91, pages 133--147, 1991.

No context found.

L. Zhang, S. Shenker, D.D. Clark, Observations and dynamics of a congestion control algorithm: the e#ects of two-way tra#c, in: ACM SIGCOMM 1991.

No context found.

L. Zhang, S. Shenker, and D. D. Clark, Observations and dynamics of a congestion control algorithm: the effects of two-way traffic, Proc. ACM SIGCOMM '91, pages 133-147, 1991. 17

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