Jacobson V., Problems with Arizona's Vegas, email to the end2end list, March 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... requirement on the minimum RTO, they do not reduce the timeout for the common case (i.e. RTT 200 msec) Besides the increased processing burden, proposals for increasing the granularity of the clock have been faced with arguments about stability, since the RTT in the Internet is highly variable [69, 103]. For example, the RTT estimate could be skewed by measurements from small packets, and would fail if a large packet is sent, due to the additional transmission delay that it sees on each hop, particularly over low speed links. This problem was recognized early on [127] Another problem would ....

.... Thus, packets are retransmitted earlier than for the other versions, such as on the receipt of 1 duplicate ACK, if the retransmit timer is found to have expired (i.e. without waiting for the clock) However, these changes have been controversial, since they may correspond to broken timer behavior [103]. Finally, Vegas makes some minor modi cations to some parameters, such as reducing the congestion window by 1 4 instead of 1 2 after a Fast Retransmit, or starting with a window of 2 segments even after a retransmit timeout [39, 2] There have been mixed reports about Vegas performance. ....

Jacobson V., Problems with Arizona's Vegas, email to the end2end list, March 1994.

....synchronous losses sharing a bottleneck node together with a constant exogenous traOEc, Vegas ff;fi has the same behavior as Reno and cannot estimate the available bandwidth if B Nff Gamma . This is the reason why Vegas performances decrease as the number of active connections increases [9]. 4. Conclusion In this paper we have studied the behavior of TCP Vegas and TCP Reno. We have considered a simple model, which allowed us to investigate dioeerent features of each algorithm, and to understand the behavior of Vegas in presence of multiple connections, and dioeerent network ....

V. Jacobson, L. Peterson, L. Brakmo, S. Floyd, Problems with Arizona's Vegas, mailing list end2end-tf (discussion on TCP Vegas), 1994.

....of the congestion window is larger than the amount of outstanding data. Reno are caused by multiple segment loss, therefore, changes which help to reduce the number of such timeouts prove very helpful. The results for TCP Vegas s fast retransmission policy (C) support Jacobson s argumentation [18] who claimed that the new policy most likely results in only a negligible performance gain. The fact that algorithm (E) has virtually no effect on throughput indicates that the multiple segment loss situations that cannot be remedied by algorithm (D) can hardly be survived without incurring a ....

V. Jacobson. problems with Arizona's vegas, March 1994. end2end-tf mailinglist.

....determine the timeout value 1 . Floyd also raised concerns about the fact that TCP Vegas does not wait for roughly half an RTT until sending segments following the fast retransmission to let congestion dissipate [13] In March 1994, Jacobson explained some concerns about TCP Vegas in more detail [21]. He pointed out that Vegas s new mechanism for retransmitting segments may either result in only a negligible gain or may work only if the retransmit timer was broken. Floyd later hinted at another possible weakness of TCP Vegas, that is, the possibility of persistent congestion [14] At ....

....help to avoid timeouts due to multiple segment loss. As shown in [5] the majority of the timeouts in TCP Reno are caused by multiple segment loss, therefore, changes which help to reduce the number of such timeouts, prove very helpful. Furthermore, these findings support Jacobson s argumentation [21] that TCP Vegas s new fast retransmission policy (A3) most likely results in only a negligible performance gain. 9 In the low background traffic case, the retransmissions caused by ACKs for new data (A4) have a small but the most dominant effect (about 5 ) among the changes in recovery. ....

V. Jacobson. problems with Arizona's vegas, March 1994. end2end-tf mailinglist.

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