| R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, January 1999. |
.... are multiplexed on the same link (right) Figure 4(right) shows the evolution of the TCP window of selected flows out of 600 TCP flows: 200 long flows whose size is uniformly distributed between 900 and 1100 packets, and 400 short flows whose size is uniformly distributed between 5 and 20 packets [20]. A short (long) flow that terminates is replaced by another short (long) flow. The packet size is 576 bytes and the maximum window size is 256 packets. All 600 TCP flows share a common bottleneck 10 Mbps link with 100 msec propagation delay and a buffer size of 1000 packets. In this experiment, ....
R.T. Morris. Scalable TCP Congestion Control. PhD thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, January 1999.
....other once a flow length threshold is crossed. Figure 3(b) shows the evolution of the TCP window of selected flows out of 600 TCP flows: 200 long flows whose size is uniformly distributed between 900 and 1100 packets, and 400 short flows whose size is uniformly distributed between 5 and 20 packets [35]. A short (long) flow that terminates is replaced by another short (long) flow. The packet size is 576 bytes and the maximum window size is 256 packets. All 600 TCP flows share a common bottleneck 10 Mbps link with 100 mseconds propagation delay (this gives a bandwidth delay product of about 434 ....
....only consider the reduced overhead due to switching, but not performance of flows. Other studies have attempted to improve TCP fairness by either modifying TCP itself or by employing non tail drop buffer management at routers [44] 45] 46] 47] 48] 49] 50] 51] 30] 52] 53] 54] [35], 37] 25] In particular, Morris [35] 37] proposes solutions that require per flow information at all routers. In this paper, we advocate the use of a (less costly) classbased solution. Bonald et al.: 38] and Nandy et al.: 55] have shown that the well known RED buffer management policy may ....
[Article contains additional citation context not shown here]
R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, January 1999.
No context found.
R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, January 1999.
....distribution that is typical of flow lengths in the Internet [1, 4] the majority of flows are found to be very short, many are long, and some are very long. However, a number of studies have shown that interesting scaling properties can arise even when flow lengths are not highly variable [6, 7]. In particular, recent work by Veres et al. 7] shows that even without any variability in terms of flow lengths or network delays, TCP itself can sometimes exhibit chaotic behavior and produce traffic series that shows properties similar to those of self similar traffic generated by synthetic ....
R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, Januray 1999.
....NSF grants CAREER ANI0096045 and MRI EIA 9871022. in the Internet [1, 4] the majority of flows are found to be very short, many are long, and some are very long. However, a number of studies have shown that interesting scaling properties can arise even when flow lengths are not highly variable [6, 7]. Furthermore, although heavy tailed flow lengths are commonly associated with heavy tailed file sizes, the authors in [8] and [9] find no strong correlation between file sizes and transmission times (even though both show heavy tails) In this paper, we study the relationship between TCP s ....
....network while still maintaining network usage efficiency. The TCP control algorithm is simple but effective. However, this simplicity does not come for free. The behavior of TCP becomes less predictable or even This high variability in file transmission times is consistent with that observed in [6]. BUCS TR 2000 017 11 chaotic when the network condition is out of TCP s control. In this paper, we have demonstrated that when a TCP connection is going through a highly lossy channel and the loss condition is not affected by this single TCP connection s behavior, TCP starts to produce packet ....
R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, Januray 1999.
....NSF grants CAREER ANI0096045 and MRI EIA 9871022. in the Internet [1, 4] the majority of flows are found to be very short, many are long, and some are very long. However, a number of studies have shown that interesting scaling properties can arise even when flow lengths are not highly variable [6, 7]. Furthermore, although heavy tailed flow lengths are commonly associated with heavy tailed file sizes, the authors in [8] and [9] find no strong correlation between file sizes and transmission times (even though both show heavy tails) In this paper, we study the relationship between TCP s ....
....while still maintaining network usage efficiency. The TCP control algorithm is simple but effective. However, this simplicity does not come for free. The behavior of TCP becomes less predictable or even 10 This high variability in file transmission times is consistent with that observed in [6]. BUCS TR 2000 017 11 chaotic when the network condition is out of TCP s control. In this paper, we have demonstrated that when a TCP connection is going through a highly lossy channel and the loss condition is not affected by this single TCP connection s behavior, TCP starts to produce packet ....
R.T. Morris, Scalable TCP Congestion Control, Ph.D. thesis, Harvard University, Cambridge MA, The Divison of Engineering and Applied Sciences, Januray 1999.
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