TANENBAUMComputer Networks II1 Chapter 6-2 The Transport Layer TCP, UDP and Performance Issues.

TANENBAUM Computer Networks II 1

Chapter 6-2The Transport Layer

TCP, UDP and Performance Issues

TANENBAUM Computer Networks II 2

The Internet Transport Protocols: UDP

• Introduction to UDP

• Remote Procedure Call

• The Real-Time Transport Protocol

TANENBAUM Computer Networks II 3

UDP (User Data Protocol)• Provides a way to send IP datagrams without establishing

a connection• A UDP segment has 8-byte header followed by data• UDP length includes header and data

• UDP checksum includes the same format pseudoheader as in TCP

TANENBAUM Computer Networks II 4

Remote Procedure Call

• Allows programs to call procedures located on remote hosts

• Looks as much possible as local call• Two library procedures called client stub and

server stub hide the fact that call is remote– The client program makes a local call to client stub to

handle the RPC

• Information is transported in parameters– For this purpose, client stub packs parameters into a

message (parameter marshalling)

TANENBAUM Computer Networks II 5

Remote Procedure CallSteps in making a remote procedure call. The stubs are

shaded.

TANENBAUM Computer Networks II 6

Problems with RPC

• Passing pointers is impossible

• Array size must be specified and fixed

• It may not be possible to deduce the types of the parameters

• Global variables can not be used

• RPC commonly uses UDP

TANENBAUM Computer Networks II 7

Real Time Transport Protocol (RTP)

• Used to transport multimedia streams• Can multiplex different streams such as the audio and video into a

single UDP socket• Each packet is given a unique increasing number• No flow control, no error control, no acks,no retransmission

mechanism• Contains a payload type field to specify encoding that allows

changing quality of the multimedia data when available bandwidth drops

• Contains timestamps that are used to synchronize streams. However, these timestamps are used to compute the relative time from the start rather than absolute time

• Has a sister protocol, Real Time Control Protocol (RTCP) that carries delay, jitter, bandwidth and congestion information for feedback purposes

TANENBAUM Computer Networks II 8

The Real-Time Transport Protocol(a) The position of RTP in the protocol stack. (b)

Packet nesting.

TANENBAUM Computer Networks II 9

The Real-Time Transport Protocol (2)

The RTP header.

TANENBAUM Computer Networks II 10

The Internet Transport Protocols: TCP• Introduction to TCP• The TCP Service Model• The TCP Protocol• The TCP Segment Header• TCP Connection Establishment• TCP Connection Release• TCP Connection Management Modeling• TCP Transmission Policy• TCP Congestion Control• TCP Timer Management• Wireless TCP and UDP• Transactional TCP

TANENBAUM Computer Networks II 11

TCP Basic Functionality• Provides reliable end-to-end byte stream over an

unreliable internetwork• Transport entity

– A piece of software that is either• A user process or• Part of kernel

– Accepts user data streams from local processes– Breaks them into pieces not exceeding 64K bytes– Send each piece as an IP datagram– At the receiver, it reconstructs original byte streams

TANENBAUM Computer Networks II 12

The TCP Service Model• The sender and receiver create end points

called sockets

• Each socket has– IP address– Port number (16 bit and local)

• A socket may be used for multiple connections at the same time

• Some port numbers are reserved– FTP: 21– Telnet: 23

TANENBAUM Computer Networks II 13

The TCP Service Model

Some assigned ports.Port Protocol Use21 FTP File transfer23 Telnet Remote login25 SMTP E-mail69 TFTP Trivial File Transfer Protocol79 Finger Lookup info about a user80 HTTP World Wide Web

110 POP-3 Remote e-mail access119 NNTP USENET news

TANENBAUM Computer Networks II 14

The TCP Service Model (2)

(a) Four 512-byte segments sent as separate IP datagrams.

(b) The 2048 bytes of data delivered to the application in a single READ CALL.

A TCP connection is a byte stream, not a message streamMessage boundaries are not preserved end to end

TANENBAUM Computer Networks II 15

The TCP Service Model (cont’d)• When an application passes data to TCP, TCP

may send it immediately or buffer it at its discretion

• To force data out, applications can use PUSH flag, which tells TCP not to delay transmission

• At the receiver side, the same effect can be obtained by using URGENT flag

• However, at the receiver side, on the contrary of PUSH flag, URGENT flag causes an interrupt to the application (provides signaling)

TANENBAUM Computer Networks II 16

The TCP Protocol• Every byte on a TCP connection has its own 32-

bit sequence number to be used for acks and sliding window sequencing

• Window mechanism has its own 32-bit header• Data is exchanged in form of segments

– A segment consists of a 20-byte header and zero or more bytes

– A segment must fit into a IP payload (65635 bytes) or Maximum Transfer Unit, MTU of the underlying network (further segmentation is a major problem)

– Sliding window acks next expected sequence number

TANENBAUM Computer Networks II 17

The TCP Segment Header

TCP Header.

TANENBAUM Computer Networks II 18

The TCP Segment Header (2)

The pseudoheader included in the TCP checksum.

TANENBAUM Computer Networks II 19

The TCP Segment Header (cont’d)• Every segment begins with a fixed format

20-byte header• Segments without any data are legal

– Ack messages or other control messages

• Ack: Next byte expected• Options

– Specifies beginning of data

• Urgent pointer specifies offset of urgent data when URG flag is set

• ACK bit indicates that the ack field is valid

TANENBAUM Computer Networks II 20

The TCP Segment Header (cont’d)• PSH flag indicates pushed data• RST flag indicates problem with connection• SYN flag indicates connection request or

connection accepted together with ACK flag• FIN flag is used to release a connection• Window Size field indicates how many bytes can

be accepted starting at the byte acknowledged.• Checksum is a control field on header, data and

pseudo header that indicates portions of IP header

TANENBAUM Computer Networks II 21

The TCP segment Header (cont’d)

• Options– A host may specify the maximum TCP

payload it is willing to accept• If not specified, the default is 536 bytes

– Window scale option• In general 16 bit window size is not sufficient for

current systems• This option scales up the window size

– Selective Reject instead of go back n can be indicated via options field

TANENBAUM Computer Networks II 22

TCP Connection Management• Server waits for an incoming connection by executing

the LISTEN and ACCEPT primitives• Client executes CONNECT primitive

– IP address, port number and options specified• Destination checks to see if a process has done a

LISTEN on the indicated port– If not,request is rejected

• If there is a process which has done a LISTEN, it is asked if it would accept the connection– If process rejects, destination sends a segment with RST set. If

process accepts, request is acked

TANENBAUM Computer Networks II 23

TCP Connection Establishment

(a) TCP connection establishment in the normal case.

(b) Call collision.

6-31

TANENBAUM Computer Networks II 24

TCP Connection Management (cont’d)

• If simultaneous connection request happens only one connection is established

• TCP connections are full duplex• However, they are considered as two simplex

connections and each of them is released independently

• To release, a node sends a segment with FIN bit set and the other side acks the release

• Same procedure is repeated for the other simplex connection.

TANENBAUM Computer Networks II 25

TCP Connection Management (cont’d)

• To avoid two-army problem, timers are used in release process

• That is, if ack for release does not arrive on time, timer goes off and client releases the connection

• After release, both sides wait for maximum packet lifetime to erase tables

TANENBAUM Computer Networks II 26

TCP Connection Management Modeling

The states used in the TCP connection management finite state machine.

TANENBAUM Computer Networks II 27

TCP Connection Management Modeling (2)

TCP connection management finite state machine. The heavy solid line is the normal path for a client. The heavy dashed line is the normal path for a server. The light lines are unusual events. Each transition is labeled by the event causing it and the action resulting from it, separated by a slash.

TANENBAUM Computer Networks II 28

TCP Transmission Policy• Window management is not directly tied to

acks as in data link protocols

• Exclusive buffer messages manage the transmission

• If no buffer at receiver, then no transmission by sender except– Urgent data may be sent– One byte segment can be sent to ask receiver

to renounce its buffer status (to prevent deadlock)

TANENBAUM Computer Networks II 29

TCP Transmission

Policy

Window management in TCP.

TANENBAUM Computer Networks II 30

TCP Transmission Policy (cont’d)

• Sender and receiver are not forced to transmit or receive as soon as they receive data from the application. This improves performance as follows:– If one byte messages are sent (like in

TELNET) then use NAGLE’s algorithm• Send first byte and keep the rest until ack comes

back• As ack comes in send the rest and keep the further

incoming bytes until ack is received

TANENBAUM Computer Networks II 31

TCP Transmission Policy (2)

• Silly Window Syndrome: Receive bytes one by one and send window messaging accordingly

TANENBAUM Computer Networks II 32

TCP Transmission Policy (cont’d)

• Clark’s Solution to Silly Window Syndrome– Prevent receiver from sending one byte

updates and make it wait until decent amount of space available before it sends buffer messages

– Sender may also pospone sending messages

• Nagle’s Algorithm and Clark’s solution are complementary and they can be used at the same time

TANENBAUM Computer Networks II 33

TCP Congestion Control• When congestion occurs, IP has limited

effect on managing congestion

• Most of the congestion control is done by TCP by cutting down the data rate

• Indication of congestion– Timeouts– Packet discards– In fiber optic cable transmission errors are

minimized so timeouts mainly due to congestion

TANENBAUM Computer Networks II 34

Congestion Window

• In addition to receiver’s buffer information, the sender also maintains a congestion window.

• This is mainly due to the fact that even if receiver may have space for fast data transfer, network may not carry it due to congestion.

• The number of bytes to be sent is the minimum of the two windows

TANENBAUM Computer Networks II 35

TCP Congestion Control(a) A fast network feeding a low capacity receiver.(b) A slow network feeding a high-capacity receiver.

TANENBAUM Computer Networks II 36

Slow Start Algorithm

• Initialize the size of the congestion window to the maximum segment size

• If no timeout occurs increase the size of the window by one segment and send again.

• Repeat above n times. If no timeout, then increase the window by n segments.

• Internet uses threshold in addition to the above algorithm

TANENBAUM Computer Networks II 37

TCP Congestion Control (2)

An example of the Internet congestion algorithm.

TANENBAUM Computer Networks II 38

TCP Timer Management:Retransmission Timer

• When goes off, segment is retransmitted

• Setting the timer value is much more difficult than that of data-link layer, beacuse the delay is not easily predictable

• TCP uses a dynamic algorithm to solve this problem

• For each connection, TCP maintains a variable called RTT (round trip time)

TANENBAUM Computer Networks II 39

TCP Timer Management

(a) Probability density of ACK arrival times in the data link layer.

(b) Probability density of ACK arrival times for TCP.

TANENBAUM Computer Networks II 40

Retransmission Timer (cont’d)• M: The time to receive ack:Smoothing factor, usually 7/8

• RTT=RTT+(1-)M

• D=D+(1-)|RTT-M|

• Timeout=RTT+4D

• Karn’s Algorithm– Do not update RTT on any segments that

have been retransmitted (difficult to figure out for what segment the ack is)

TANENBAUM Computer Networks II 41

TCP Timers (cont’d)

• Persistence Timer– Used to prevent deadlock when buffer

management messages are lost

• Keepalive Timer– Check the other side if the line is idle for a

long time

• Timed Wait– Used to close the connection

TANENBAUM Computer Networks II 42

Wireless TCP and UDP• In theory, transport protocol should be

independent of the technology of the underlying network layer

• However, if the transmission is wireless, performance of TCP degrades drastically– Wired connections

• Timeouts are considered to be caused by congestion not by lost packets

– Wireless Connections• Timeouts are primarily due to lost packets

TANENBAUM Computer Networks II 43

Wireless TCP and UDP (cont’d)

• Wired TCP Timeout– Slowdown to alleviate congestion

• Wireless TCP Timeout– Retransmit as soon as possible

• Indirect TCP– Split TCP connection into two separate

connections• Sender-to-base station• Base station to mobile host

TANENBAUM Computer Networks II 44

Wireless TCP and UDP

Splitting a TCP connection into two connections.

TANENBAUM Computer Networks II 45

Indirect TCP• Snooping Agent

– Observes and caches TCP segments going out to the mobile host and acknowledges coming back from it

– When an ack for a segment to mobile host is late, it retransmits the segment to mobile host without telling anything to the source

– Congestion control algorithm will never start unless there is congestion on the wired part

• Wireless UDP– Stations should not expect reliability

TANENBAUM Computer Networks II 46

Translational TCP

• If RPC requires long replies, then UDP is not suitable and TCP becomes the choice

• However, TCP requires nine packets and this not efficient

• A variant of TCP called T/TCP (Translational TCP) solves this problem– Allows transfer of data during setup

TANENBAUM Computer Networks II 47

Transitional TCP(a) RPC using normal TCP.(b) RPC using T/TCP.

TANENBAUM Computer Networks II 48

Performance Issues

• Performance Problems in Computer Networks

• Network Performance Measurement

• System Design for Better Performance

• Fast TPDU Processing

• Protocols for Gigabit Networks

TANENBAUM Computer Networks II 49

PERFORMANCE ISSUES:Performance Problems

• Fast line-slow CPU

• Broadcast storm

• Power failure (RARP density when machines are up)

• Lack of memory allocation

• Timeouts set incorrectly

• Bandwidth-delay product (BDP)– The receiver window should be at least BDP

• Jitter

TANENBAUM Computer Networks II 50

Performance Problems in Computer Networks

The state of transmitting one megabit from San Diego to Boston(a) At t = 0, (b) After 500 μsec, (c) After 20 msec, (d) after 40 msec.

TANENBAUM Computer Networks II 51

Measuring Network Performance

• Make sure that the sample size is large enough• Make sure that the samples are representative• Be careful when using a coarse-grained clock• Be sure that nothing unexpected is going on

during your tests• Caching can wreak havoc with measurements• Understand what you are measuring• Be careful about extrapolating the results

TANENBAUM Computer Networks II 52

Network Performance Measurement

The basic loop for improving network performance.

• Measure relevant network parameters, performance.

• Try to understand what is going on.

• Change one parameter.

TANENBAUM Computer Networks II 53

System Design for Better Performance

Rules:• CPU speed is more important than network speed.• Reduce packet count to reduce software overhead.• Minimize context switches.• Minimize copying.• You can buy more bandwidth but not lower delay.• Avoiding congestion is better than recovering from it.• Avoid timeouts.

TANENBAUM Computer Networks II 54

System Design for Better Performance (2)

Response as a function of load.

TANENBAUM Computer Networks II 55

System Design for Better Performance (3)

Four context switches to handle one packet with a user-space network manager.

TANENBAUM Computer Networks II 56

Fast TPDU Processing• Main obstacle is protocol software

• Processing overhead components– Overhead per TPDU– Overhead per byte

• Optimization for overhead per TPDU– Minimize tests if in ESTABLISHED state– Headers of consecutive data TPDU’s are

almost the same• Store a prototype header and overwrite the

changing values

TANENBAUM Computer Networks II 57

Fast TPDU Processing

The fast path from sender to receiver is shown with a heavy line.The processing steps on this path are shaded.

TANENBAUM Computer Networks II 58

Fast TPDU Processing (2)

(a) TCP header. (b) IP header. In both cases, the shaded fields are taken from the prototype without change.

TANENBAUM Computer Networks II 59

Optimizations for high speed networks

• Buffer management– Avoid unnecessary copying

• Timer Management– Use linked list of timer events sorted by expiry

time• At every clock tick, the counter in the head entry is

decremented

– Timer wheel• More efficient if timer interval is known in advance

TANENBAUM Computer Networks II 60

Timer Wheel

TANENBAUM Computer Networks II 61

Protocols for Gegabit Networks• 32 bit sequence number is small

• Communication speed is much faster than computing speed

• Go back n is not good when bandwidth-delay product is high

• Gegabit lines are delay limited rather than bandwidth limited

• Variance of packet arrival time may be very important

TANENBAUM Computer Networks II 62

Protocols for Gigabit Networks

Time to transfer and acknowledge a 1-megabit file over a 4000-km line.

TANENBAUM Computer Networks II 63

Gegabit (cont’d)• Design for speed not for bandwidth optimization• Make protocols simple and let the main CPU do

the work• Reserve necessary sources at the connection

set up time rather than implementing e.g. Slow start algorithm

• Minimize header length and copy time• Checksum the header and data separately• Concentrate on succesful cases rather than

failures

TANENBAUM Computer Networks II 64

Required Reading

• Tanenbaum Sections– 6.4– 6.5– 6.6