1 / 32

Lect3. - 09/12/04

CIS 4100 Systems Performance and Evaluation. Lecture 3 by Zornitza Genova Prodanoff. Lect3..ppt - 09/12/04. Lecture Outline. Chapter 2: Types of networks Network protocols Client/Server Model Peer-to-Peer Model Web services protocols. ZGP00 2. Communication Protocols.

toki
Télécharger la présentation

Lect3. - 09/12/04

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. CIS 4100 Systems Performance and Evaluation Lecture 3 by Zornitza Genova Prodanoff Lect3..ppt - 09/12/04

  2. Lecture Outline • Chapter 2: • Types of networks • Network protocols • Client/Server Model • Peer-to-Peer Model • Web services protocols ZGP002

  3. Communication Protocols • Communication is between two remote processes • Protocol is a set of rules • Addressing • Routing • Error detection • Error recovery • Sequence control • Flow control ZGP003

  4. Communication Protocols (Continued) • Connection oriented protocols (e.g. TCP) • Brake the message into fragments • Once connection is established no fragment routing occurs • Ensure that all fragments are processed in original order • Used for applications communicating through large messages • Connectionless protocols (e.g. UDP) • Each message is routed independently • Used when messages are smaller than the max. fragment size • Order of arrival is not important - a single fragment is sent ZGP004

  5. Communication Protocols (Continued) • Protocol specification • Syntax: message format and fields (bits) designation • Semantics: actions taken by each entity when events occur • E.g. fragment loss, timeout, message arrival • Complex design • – Solution: use layered approach • - Abstract from details (at each layer) and let the protocols at other layers handle them ZGP005

  6. Type of Networks (Continued) ZGP006

  7. Client/Server Performance (continued) ZGP007

  8. Communication Protocols: Internet Protocols • The two most important protocols in the Internet are the Internet Protocol (IP) and the Transmission Control Protocol (TCP) • - Suite of protocols known as TCP/IP • Figure 2.7 shows the layering of important TCP/IP-based protocols • IP is a network layer protocol on top of which we find two transport layer protocols: TCP-a connection-oriented protocol-and User Datagram Protocol (UDP)-a connectionless protocol • Simple Mail Transfer Protocol (used to send e-mail), and Telnet (an interactive login protocol) use TCP • Network File System (NFS) [8], Domain Name Server (DNS), and Simple Network Management Protocol (SNMP) are built on top of UDP ZGP008

  9. Client/Server Performance (continued) ZGP009

  10. Communication Protocols: IP • Internet Protocol: • Specifies formats of packets sent across the Internet • Connectionless protocol • Source to destination forwarding of packets (called datagrams) • Every host connected to the Internet has a unique address, called IP Address • Address representation is dotted decimal notation, e.g. 129.2.0.37 • Prefix (identifies a network) and suffix (identifies a host) address bits ZGP0010

  11. ZGP0011

  12. Client/Server Performance (continued) ZGP0012

  13. Communication Protocols: TCP • Transport Control Protocol: • Connection oriented protocol • TCP guarantees that all the packets of a message are received in their original sequence • TCP guarantees that data is delivered without loss or modification of bits • Packets are called segments • Connection establishment through three-way-handshake • Full-duplex communication ZGP0013

  14. Client/Server Performance (continued) ZGP0014

  15. Communication Protocols: TCP • TCP uses ACKs (acknowledgement messages), timeouts, and retransmissions for error control • ACK are cumulative • Flow control is implemented by TCP through a sliding window mechanism • The window size is the maximum number of bytes that can be sent before an acknowledgement is received • Window size is limited by • the buffer size at the receiver • by network congestion as estimated by the sender ZGP0015

  16. Communication Protocols: TCP • B is limiting bandwidth • RTT is round trip time • MSS is maximum segment size • Wc is the current window size • Wm is the maximum window size (prevents buffer overflow) • Wc doubles every RTT but cannot exceed the receivers window Wm ZGP0016

  17. Communication Protocols: TCP • “Sliding window” flow control • The rate with which sender can send data is determined by the rate of incoming ACKs for previous segments • The rate of incoming ACKs is determined by the bottleneck in the round trip path (destination or the network) • Wm is advertised by the receiver • If network congestion is detected, the sender will reduce its window size Wc to reduce its transmission rate ZGP0017

  18. Client/Server Performance (continued) ZGP0018

  19. Communication Protocols: TCP (continued) • Slow start phase and Congestion avoidance phase • Consider Figure 2.10 • During slow start Wc is initialized to one segment and is increased by one for every ACK received • Once segments are sent an ACK is received after one RTT unit • Wc doubles every RTT unit ZGP0019

  20. Communication Protocols: TCP (continued) • Network congestion can be detected by TCP through • the receipt of a duplicate ACK, which indicates that the receiver received an out-of-sequence segment • or by a timeout at the sender • The current value of the window size is saved in a variable called slow start threshold window size, Wssthr, and Wc, is reduced • TCP Reno: If a duplicate ACK is received, W, is divided by two and TCP enters congestion avoidance phase (described in Figure 2.11) • TCP Reno: If a timeout occurs, Wc is set to one and TCO goes back to slow start ZGP0020

  21. Communication Protocols: TCP (continued) • When Wc reaches Wssthr, TCP enters congestion avoidance • During congestion avoidance the window size Wc is incremented by one every RTT unit ZGP0021

  22. Communication Protocols: TCP (continued) • In the example of Fig. 2.11, a duplicate ACK is received at time 5. Then, half of the current value of Wc, i.e., 8, is stored in Wssthr and the TCP connection enters a congestion avoidance phase. • Another duplicate ACK is received at time 20. The value of Wssthris set to 7 (14 / 2). If a timeout occurs, then Wc, is set to one and TCP goes back to slow start ZGP0022

  23. Communication Protocols: TCP (continued) • In Fig. 2.11, a timeout occurred at time 10 and the TCP connection enters a slow start phase • When Wc, reaches Wsethr during slow start, TCP switches to congestion avoidance. For example, at time 13 in Fig. 2.11, Wc, reaches the current value of Wssthr, which is 8, and the connection enters a congestion avoidance phase. ZGP0023

  24. Communication Protocols: TCP (continued) • TCP and congestion control • The throughput XTCP of a TCP connection, measured in segments per second • decreases with RTT, • decreases with the probability p that packets are dropped • increases with the receiver advertised window size Wr measured in segments • decreases with the value of the TCP timeout To • ( Based on Padhye’s expression for XTCP for a TCP Reno connection) • A Model is implemented in the TCPModel. XLS ZGP0025

  25. Communication Protocols: TCP (Continued) • Fig. 2.12 shows variation of the TCP throughput in KB/sec as a function of p and for four values of Wm: 10, 20, 30, and 40 segments • TCP timeout (To) = 2 seconds • Maximum segment size of 1,460 bytes • Limiting bandwidth of the connection between sender and receiver equal to 12,500 KB/sec • RTT = 0.04 seconds ZGP0026

  26. Client/Server Performance (continued) ZGP0027

  27. Communication Protocols: TCP (Continued) • As Wm increases, the throughput increases • The benefits of a larger window decrease as the packet loss probability increases because more data has to be retransmitted in case of packet losses ZGP0028

  28. Communication Protocols: TCP (Continued) • The best-case throughput of a TCP connection is obtained when the window size Wc, reaches the receiver maximum window size Wm and no congestion occurs • The throughput is given by segments per second • or bytes/sec where MSSis the • maximum segment size • The throughput of a TCP connection cannot exceed the limiting bandwidth B, in bytes/sec, of the network connecting the sender and receiver => ZGP0029

  29. Communication Protocols: TCP (Continued) • Then, the maximum throughput is achieved when the windows size given in segments , is equal to • The numerator is called the bandwidth-delay product. • During slow start, TCP operates with windows that are below the window size that would be achieved during congestion avoidance. • problem for short-lived connections that may never achieve the maximum throughput under an optimal window size ZGP0030

  30. Communication Protocols: TCP (Continued) • Table 2.1 shows the bandwidth-delay product, and the optimum receiver window size, , in segments, assuming a segment size of 1,460 bytes, for various types of networks. • As shown by the table, high-bandwidth, low-latency networks require large window sizes for the throughput of the TCP connection to use as much of the network bandwidth as possible. ZGP0031

  31. Client/Server Communication ZGP0032

More Related