w3gtcp pres
Embed Size (px)
TRANSCRIPT
-
8/8/2019 w3gtcp Pres
1/21
TCP/IP over 3G Wireless
Links with Rate and DelayVariation
M. Chan and R.Ramjee
Bell Labs, Lucent technologies
-
8/8/2019 w3gtcp Pres
2/21
Content
Introduction Outline of a model used here
Proposed solution(Ack Regulator) Outline of the simulation result for
using AR Conclusion(contributions and some left)
-
8/8/2019 w3gtcp Pres
3/21
Introduction
Wireless link losses result in poor TCPthroughput.
Added mechanisms add the Delay andRate variation at the same time
-
8/8/2019 w3gtcp Pres
4/21
Introduction
This paper proposed a model to explainthe impact of variable rate and delay on
long lived TCPs performance It proposed a solution to improve the
performance in 3G wireless network
-
8/8/2019 w3gtcp Pres
5/21
Model
This paper model the performance of asingle long lived TCP flow over a network(rate)with a single bottleneck server, and a
wireless link(delay). The model used here is an extension of a
model in ideal environment and constant rate
and delay.
Here the rate and delay both arechanged to variable.
-
8/8/2019 w3gtcp Pres
6/21
Model
Here are the figures and table they got:
-
8/8/2019 w3gtcp Pres
7/21
Model
-
8/8/2019 w3gtcp Pres
8/21
Ack Regulator
NamedAck regulator since it regulates theflow of acks back to the TCP source.
The solution is designed for improving theperformance of TCP flows towards the mobilehost
Implemented at the wireless edge, RNC,
Need a per-TCP-flow queue
-
8/8/2019 w3gtcp Pres
9/21
Ack Regulator
The implementation:
Run on every transmission of data and arrival of ack.
-
8/8/2019 w3gtcp Pres
10/21
Ack Regulator
The intuition: To avoid any buffer overflow loss until the congestion window at theTCP source reaches a pre-determined
threshold and beyond that, allow only asingle buffer over flow loss.
Therere two important variable:1.ConservativeMode2.AckReleaseCount
-
8/8/2019 w3gtcp Pres
11/21
Ack Regulator
The Ack Regulator algorithm:
-
8/8/2019 w3gtcp Pres
12/21
Ack Regulator
Therere two algorithms forConservativeMode and AckReleaseCount
-
8/8/2019 w3gtcp Pres
13/21
Simulation Result
It used the Topology as this:
-
8/8/2019 w3gtcp Pres
14/21
Simulation Result
Summary of the comparison:(Reno ,Sack vsAR)
1.
43% Reno, 19% Sack
-
8/8/2019 w3gtcp Pres
15/21
Simulation Result
2. Variable rate:
Throughput 15% Reno, same as Sack
-
8/8/2019 w3gtcp Pres
16/21
Simulation Result
3. Variable delay and rate:
22% Reno and 10% Sack
-
8/8/2019 w3gtcp Pres
17/21
Simulation Result
4.HDR variable rate and delay:
5% to 33% Reno and 0.5% to 24% Sack
-
8/8/2019 w3gtcp Pres
18/21
Simulation Result
5. Multiple flows sharing HDR:
32% Reno and 36% Sack
-
8/8/2019 w3gtcp Pres
19/21
Conclusion
Contribution: 1. Evaluated the impact of variable rate and variable
delay on long-lived TCP performance.
2. Proposed a model to explain and predict TCPsthroughput over a link with variable rate and/ordelay.
3. Proposed a network-based solution called AckRegulator
-
8/8/2019 w3gtcp Pres
20/21
Conclusion
Question left: Queue: Two kinds(dqueue and aqueue) and per-
TCP flow queue, maybe need more bandwidth( badstate!)
Big window used: 500KB, never window limited
-
8/8/2019 w3gtcp Pres
21/21
Thank you very much
Nice Weekend