Effects of Advertised Receive Buffer Size and Timer Granularity on TCP Performance in a LEO...
-
Upload
vincent-turner -
Category
Documents
-
view
218 -
download
1
Transcript of Effects of Advertised Receive Buffer Size and Timer Granularity on TCP Performance in a LEO...
Effects of Advertised Receive Buffer Size and Timer Granularity on TCP Performance in a LEO Satellite Network
Effects of Advertised Receive Buffer Size and Timer
Granularity on TCP Performance in a LEO Satellite
Network
Marc Emmelmann, [email protected]
Fraunhofer Institute forOpen Communication Systems
IEEE Globecom 2002
Taipei, Taiwan
17-21November 2002
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Outline
Introduction - Project Framework- Target System Parameter- TCP Basics
Simulation Environment - Used Simulation Models- Simulation Parameters
Simulation Results - Effects of Buffer Size & Timer Granularity- Effects of Bit Error Rates- Overall TCP Performance
Conclusion & Outlook
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Project Framework
Partner - Tesat-Spacecom- German Aerospace Agency DLR- Fraunhofer FOKUS
Satellite Network - Support of multimedia applications- LEO orbits- ATM based
Project Aspects Target System Design
Antennas
Demonstrator
TCP Simulation (Reno)
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Target System DesignSystem Parameter
Satellite Orbit - 72 satellites (12 * 6)- Orbit altitude: 1350 km- Inclination 47 deg.- ISLs: on-board processing and switching
Bit Rates - Uplink: up to 2.048 Mbit/s- Downlink: up to 32.768 Mbit/s- 16 kbit/s granularity
Bit Error Rates - Target BER < 2*10e-8- Shadowing- Adaptive FEC
Data Transfer - Usage of TCP Reno
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
TCP Basics
Provides - reliable end-to-end transfer
Retransmissions caused by - expiration of retransmission timeout RTO- duplicate ACKs (Fast Retransmission)
Calculates the RTO - upon smoothed RTT- timer granularity G acts as low pass filter
RTO - may cause premature retransmissions(RTO < RTT)
- may “delay” the TCP transmission if segments are lost(RTO >> RTT)
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation Parameter
Lower Bound Of RTT in ms None 100 200 Default
Timer Granularity G
G=1 ms 2 7
G=50 ms 3 8
G=100 ms 4 9
G=250 ms 5 10
G=500 ms 6 11 12
Minimal G to avoid false retransmissions
(varying)1
Given:- TCP Flavor- Buffer Size- BER
More than 140 experiments
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation ResultsEffects of Buffer Size & Timer Granularity
Advertised Receive Buffer - Larger than BDP- Less than BDP- Equal to MSS
Timer Granularity - May cause false retransmissions
Advertised Receive
Buffer
64 kB
>> BDP
6.6 kB
= BDP
3.3 kB
1/2
BDP
1.46 kB
MSS
Min. Timer Granularity
(G)1 ms 1 ms 4 ms
26 ms (250
ms)
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation ResultsEffects of Bit Error Rates
BER = 2*10e-8 - Overlapping with BER = 0- No needless retransmissions
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation ResultsEffects of Bit Error Rates (cont.)
BER = 2 * 10e-5 - No resemblance with variable propagation delay- Prevention of false retransmissions
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation ResultsTCP Performance
Degradation of TCPthroughput BER = 2* 10e-8
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Simulation ResultsTCP Performance
Degradation of TCPthroughput BER 2*10e-5
Page [email protected] IEEE Globecom ‘02, Nov 17-21 2002, Taipei, Taiwan
Conclusion & Outlook
TCP - Sophisticated protocol- Minor influence of hand-tuned implementations- TCP flavor dominates
Demonstrator - Run experiments over “real” MAC- Combine simulation models and demonstrator
Further Information - http://www.fokus.fhg.de/cats/satellite- [email protected]