Transport, Network, and Data-Link Layer Protocol Design Specifications that improve Near-Earth Data Communication Performance Paul D. Wiedemeier Computer Science Department University of Missouri Columbia Columbia, Missouri, 65211, USA WiedemeierP@missouri.edu Harry W. Tyrer Electrical and Computer Engineering Department University of Missouri Columbia Columbia, Missouri, 65211, USA TyrerH@missouri.edu MIZZOU High Performance Computer Networking Group (MU-HPCNG) v.2004.10.18 Problem The Transmission Control Protocol (TCP) transmits large files (data sets) over a GEO satellite reliably, but significant time is required. Approximately 6110 seconds (~102 minutes) to transmit a 20 MB data set over a GEO satellite with a 155.520 Mbps transmission rate, 280 ms transmission delay, and 1.0e- 04 BER using TCP Reno. Causes GEO satellite s large transmission delay TCP s small segment size and small maximum window size. TCP s congestion avoidance and control algorithms. Rationale Rural America is underserved by high bandwidth terrestrial infrastructure. Space missions using COTS hardware. 1
Our Solutions Multiple Segment Transmission (MST) with Majority Decoding (MD) Transport Layer Protocol MST/MD transmits large data sets over GEO satellites in less time than TCP Reno. Uses segment sizes and maximum window sizes larger than 64 KB. Provides reliable data transmission. IP Datagram Size Routing within Hybrid Networks A hybrid network consists of two earth stations each connected to a GEO satellite channel and a terrestrial channel. Route datagrams through network based on size. Round Robin Multiplexing 2 or more earth stations cooperate to share a GEO satellite s aggregate bandwidth. Similar to Token Bus. Research Assumptions Split Protocols at the Transport Layer File Spoofing at the Transport Layer Error Correction at the Transport Layer IPv6 Jumbograms at the Network Layer 2
Materials & Methods Personal Computer & Red Hat Linux Computer Network Protocol Simulator ns-2 version 2.1b9 GEOS Simulation Topology Hybrid Simulation Topology MST/MD Transport Layer Protocol Multiple Segment Transmission Function A source transmits each segment at least once, but at most n times (n odd). Maximum Window Size (MWS) = 2 X <= Transmission Rate * Round Trip Time. Segment Size (SS) = Maximum Window Size. Majority Decoding Function Based on binomial random variable theory. A destination determines the original message by inspecting the n bits at each position for all segments received with the same sequence number. The value for n, from above, is determined based on segment size. Assume a GEO satellite with a 155.520 Mbps transmission rate, 280 ms transmission delay, and 1.0e-04 BER... SS = MWS = 23 2 = 8 MB <= 155.520 Mbps * 560 ms = 10.3820 MB. Using our MST/MD transport layer protocol, if an 8 MB segment is transmitted 7 times (i.e. n), then the probability of successful transmission is 0.9999. 3
MST/MD TLP Variants Connection-Oriented Columbia version 1.0 (COCv1) A source transmits a segment n times, and then wait for ACKs or RTOs. A destination performs majority decoding on all multiply received segments. Connection-Oriented Columbia version 2.0 (COCv2) A source transmits a segment, and then wait for ACK or RTO, retransmits the same segment if no ACK or RTO, and transmits a segment at least once, but at most n times. A destination performs majority decoding only if all seven segments arrive in error. Connection-Less Columbia version 1.0 (CLCv1) A source transmits a segment n times, but does not wait for ACKS or RTOs. A destination performs majority decoding on all multiply received segments. MST/MD Results Zero Bit Error Rate and Varied Data Set Size Percent Decrease in Transmission Time versus TCP Reno Channel Utilization versus Data Set Size Data Set Size CO Columbia version 1 CO Columbia version 2 CL Columbia version 1 512 B 1 KB 7.6% 2 KB 30.7% 20 KB 60.4% 200 KB 74.8% 19.1% 2 MB 66.5% 93. 33.1% 20 MB 88.3% 98. 91.8% 200 MB 90.3% 98.5% 93. The table and graph indicate All MST/MD TLP variants perform better than TCP Reno when data set size >= 2 MB. Use CLCv2 when data set size >= 1 KB. 4
MST/MD Results Varied Bit Error Rate and 20 MB Data Set Percent Decrease in Transmission Time versus TCP Reno Channel Utilization versus Bit Error Rate Bit Error Rate CO Columbia version 1 CO Columbia version 2 CL Columbia version 1 0.0 88.3% 98. 91.8% 1.0e-08 88.9% 98.1% 92.2% 1.0e-07 91. 95.7% 93.7% 1.0e-06 93.2% 93.2% 96.7% 1.0e-05 97.8% 97.8% 98.9% 1.0e-04 99.4% 99.4% 99.7% The table and graph indicate Use COCv2 when bit error rate <= 1.0e-07. Use CLCv1 when bit error rate >= 1.0e-06. IP Datagram Size Routing The MST/MD segments are larger than their associated acknowledgements 8 MB versus 40 B 1. Acknowledgements also traverse GEO satellite channel. Inefficient channel throughput and utilization. Use Hybrid Simulated Topology GEO satellite channel Terrestrial channel Internet Protocol Datagram Size Routing (DGSR) Transmit 8 MB MST/MD segments over the GEO satellite channel. Transmit 40 B acknowledgements over the terrestrial channel. In this way, pure data transits the GEO satellite channel. 1. ns-2 simulator default 5
DGSR Results Zero Bit Error Rate and Varied Data Set Size Percent Decrease in Transmission Time versus GEOS+TCPReno Transmission Time versus Data Set Size Data Set Size 512 B 1 KB 2 KB 20 KB 200 KB 2 MB 20 MB 200 MB Hybrid+COCv2+DGSR 31.9% 48.9% 70.8% 81.4% 94.8% 98.5% 98.8% GEOS+COCv2 7.6% 30.7% 60.4% 74.8% 93. 98. 98.5% The table and graph indicate Both Hybrid+COCv2+DGSR and GEOS+COCv2 perform best when data set size >= 1 KB. Hybrid+COCv2+DGSR performs better than GEOS+COCv2 for all data set sizes. DGSR Results Varied Bit Error Rate and 20 MB Data Set Percent Decrease in Transmission Time versus GEOS+TCPReno Transmission Time versus Bit Error Rate Bit Error Rate 0.0 1.0e-08 1.0e-07 1.0e-06 1.0e-05 1.0e-04 Hybrid+COCv2+DGSR 98.5% 98.5% 87.1% 93.2% 97.8% 99.4% GEOS+COCv2 98. 98.1% 95.7% 93.2% 97.8% 99.4% The table and graph indicate Both Hybrid+COCv2+DGSR and GEOS+COCv2 perform better than GEOS+TCPReno. Use Hybrid+COCv2+DGSR when bit error rate <= 1.0e-08. Both Hybrid+COCv2+DGSR and GEOS+COCv2 perform equally when bit error rate >= 1.0e-06. 6
Round Robin Multiplexing By design, an earth station that uses a MST/MD transport layer protocol variant must secure the aggregate bandwidth of the communication channel, without transmission interference from the other earth stations. Our Round Robin Multiplexing (RRM) strategy functions similar to IEEE standard 802.4, Token Bus. Used by earth stations when transmitting data using a MST/MD transport layer protocol variant. Round Robin Multiplexing All earth stations are allocated a turn. An earth station transmits one data set during it s turn. An earth station passes to the next earth station when it does not have a data set to transmit. RRM Results Zero Bit Error Rate and Varied Data Set Size Transmission Time versus Data Set Size Varied Bit Error Rate and 20 MB Data Set Transmission Time versus Bit Error Rate The graph indicates GEOS+TCPReno+FDM performs better than GEOS+CLCv1+RRM for all data set sizes. Hybrid+COCv2+DGSR+RRM performs better than GEOS+TCPReno+FDM for data sets >= 200 MB. The graph indicates GEOS+CLCv1+RRM performs better than GEOS+TCPReno+FDM for bit error rates >= 1.0e-05. Hybrid+COCv2+DGSR+RRM performs better than GEOS+TCPReno+FDM for bit error rates >= 1.0e-04. 7
RRM Results Channel Utilization versus Earth Stations Fold Increase in Utilization versus GEOS+TCPReno+FDM Earth Stations Hybrid+COCv2 +DGSR+RRM GEOS+CLCv1 +RRM 2 198.40 408.9 4 198.42 409.0 8 198.47 409.1 16 198.5 409.3 32 198.7 409.7 64 199.1 410.5 128 199.9 412.1 1.0e-04 Bit Error Rate and 20 MB Data Set The table and graph indicate Both Hybrid+COCv2+DGSR+RRM and GEOS+CLCv1+RRM perform better than GEOS+TCPReno+FDM. Wiedemeier and Tyrer Core Technologies for Space Systems Conference November 8th 10th, 2004 RRM Results Channel Utilization versus Earth Stations 1.0e-04 Bit Error Rate and 20 MB Data Set The graphs indicate As the ratio of transmitting earth stations (ROTES) decrease from 1.0 to 0.25, the performance of GEOS+TCPReno+FDM also decreases. Wiedemeier and Tyrer Core Technologies for Space Systems Conference November 8th 10th, 2004 8
Conclusions & Additional Work Conclusions Our MST/MD transport layer protocols transmit large data sets (i.e. <= 20 MB) in less time and with better channel utilization than TCP Reno when the GEO satellite s bit error rate is large (i.e. 1.0e-04). Our datagram size routing strategy provides faster data transmission over a hybrid topology compared to a single GEO satellite channel. Our round robin multiplexing strategy allows organizations with large data transmission requirements, but few geographic locales, to use our MST/MD transport layer protocol variants. Additional Work Implement the MST/MD transport layer protocol variants, the IP datagram size routing strategy, and round robin multiplexing strategy from our designs, then emulate. Conduct tests within a test-bed, and/or test within a live environment. Compare the performance of CLCv1 to other connection-less protocols, including UDP, CFDP, MDP, and NORM. Publications & Presentations Wiedemeier and Tyrer, MST/MD: A Transport Layer Protocol that improves Large Data Set Transmission over Geo-Stationary Earth Orbit Satellites, 4 th NASA Space Internet Workshop, Hanover, Maryland, June 8 th 10 th, 2004, http://siw.gsfc.nasa.gov/. Wiedemeier and Tyrer, A Multiple Segment Transmission with Majority Decoding Transport Layer Protocol for Geo-Stationary Earth Orbit Satellites, Proceedings of the 2004 IEEE Aerospace Conference, Institute of Electrical and Electronics Engineers, Big Sky, Montana, March, 2004. Wiedemeier and Tyrer, Large File Transmission over High Bandwidth High Delay Communication Channels using TCP, Proceedings of the 2003 IEEE Aerospace Conference, Institute of Electrical and Electronics Engineers, Big Sky, Montana, March, 2003. 9
Thank you for your attention Questions 10