Heterogeneous data networks : congestion or corruption? /
TCP is a popular transport protocol used in present-day Internet. TCP was designed and tuned for networks where losses are mainly due to congestion. When packet losses occur, TCP assumes that packet losses are due to congestion, arid responds by reducing its congestion window. When a TCP connection...
| Main Author: | |
|---|---|
| Format: | Thesis Book |
| Language: | English |
| Published: |
[Place of publication not identified] :
[publisher not identified] ;
1999.
|
| Subjects: | |
| Online Access: | http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=730318831&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD |
| Summary: | TCP is a popular transport protocol used in present-day Internet. TCP was designed and tuned for networks where losses are mainly due to congestion. When packet losses occur, TCP assumes that packet losses are due to congestion, arid responds by reducing its congestion window. When a TCP connection traverses a wireless link, a significant fraction of packet losses may occur due to transmission errors. TCP responds to such losses also by reducing congestion window. This results in unnecessary degradation in TCP performance. Our main objective is to design a loss discriminator which can diagnose with high accuracy the real cause of a loss : congestion or corruption. Depending on this determination, the TCP sender can take actions appropriate for the actual type of loss. This dissertation proposes three simple loss discriminatory based on congestion avoidance techniques (CATs) proposed previously. These sender-based discriminatory use simple statistics on round-trip times and/or throughput, to determine the cause of a packet loss. An objective of this dissertation is to investigate the ability of CAT-based discriminators to determine the cause of a packet loss. The ultimate objective is to design an efficient loss discriminator. This objective is hard to achieve. We prove that it is impossible to design a perfect discriminator based on congestion avoidance techniques. This dissertation analyzes traces of more than 14,000 real TCP connections over 737 paths on the internet. We show that there is a weak correlation between the number of packets in flight and the round trip time even for slow links. We designed a receiver-based loss discriminator which performs efficiently under certain conditions. Finally, this dissertation presents one simple scheme based on multiple acknowledgements. This scheme performs well when the last hop is wireless. |
|---|---|
| Item Description: | Vita. "Major Subject: Computer Science". |
| Physical Description: | xii, 124 leaves : illustrations ; 28 cm. Issued also on microfiche from University Microfilm Inc. |
| Bibliography: | Includes bibliographical references (leaves 119-123). |