| L. S. Brakmo and L. L. Peterson. Performance problems in bsd4.4 tcp. Computer Communication Review, 25(5):69--86, Oct. 1995. |
....and furthermore, automatic optimization appears to be a less error prone pro cess. TCP Lite was shipped with a minor bug in its header prediction code. The code as shipped includes one case in its fast path which requires re calculation of the congestion window which is not done in the fast path [5]. HIPPCO optimizations are general tree to tree translations which can easily be shown not to change the semantics of the overall program. It also needs to be driven home that a process intensive specication does not require a process intensive implemen tation. The Esterel front end takes a ....
L. Brakmo and L. Peterson. Performance problems in bsd4.4 tcp. Computer Communication Review, (5):6986, October 1995.
....be computed for connections that experience RTTs considerably smaller than 0. 5 seconds (in such cases the measured RTT of the connection is often 0 clock ticks) The solution is either to use a clock with a better granularity or to enhance the precision of the smoothed RTT (srtt) as proposed in [2]. A much simpler, though less accurate, solution is to use a pre defined T fact , e.g. 1 or 2, for connections whose RTT is smaller than 0.5 seconds. Note that in general, enhancing clock granularity has been shown to improve TCP performance (e.g. 6] Acknowledgments This research was ....
L. S. Brakmo and L. L. Peterson. Performance problems in bsd4.4 tcp. Computer Communication Review, 25(5):69--86, Oct. 1995.
....by x sim since its implementation of TCP Reno is directly derived from the BSD implementation of TCP Reno. We made two changes to the original implementations of TCP Reno and TCP Vegas provided in the x kernel. Both of these changes were proposed in a paper from the inventors of TCP Vegas [7] and have also been applied to the current TCP Reno releases of FreeBSD and NetBSD. The changes include a modification to the algorithm for computing the retransmission timeout value (see discussion below) and a fix of the check to reduce the congestion window upon leaving fast recovery. ae oe ....
L.S. Brakmo and L.L. Peterson. Performance Problems in BSD4.4 TCP. Computer Communication Review, 25(5):69--86, Oct 1995.
....oe R2 ae oe R1 200 Kbytes sec 50 ms Ethernet Ethernet H1 H2 H3 H4 Fig. 1. Network topology for simulations. We made two changes to the original implementations of TCP Reno and TCP Vegas provided in the x kernel. Both of these changes were proposed in a paper from the inventors of TCP Vegas [8] and have also been applied to the current TCP Reno releases of FreeBSD and NetBSD. The changes include a modification to the algorithm for computing the retransmission timeout value (see discussion below) and a fix of the check to reduce the congestion window upon leaving fast recovery. B. ....
L.S. Brakmo and L.L. Peterson. Performance Problems in BSD4.4 TCP. Computer Communication Review, 25(5):69--86, Oct 1995.
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