Transcript Darwin: Customizable Resource Management for Value

20-755: The Internet
Lecture 5: Internetworking II
David O’Hallaron
School of Computer Science and
Department of Electrical and Computer Engineering
Carnegie Mellon University
Institute for eCommerce, Summer 1999
Lecture 5, 20-755: The Internet, Summer 1999
1
Today’s lecture
•
•
•
IP: Internetworking with routers (50 min)
Break (10 min)
UDP, TCP (35 min)
Lecture 5, 20-755: The Internet, Summer 1999
2
Typical computer system
Keyboard
Processor
Interrupt
controller
Mouse
Keyboard
controller
Modem
Serial port
controller
Printer
Parallel port
controller
Local/IO Bus
Memory
IDE disk
controller
SCSI
controller
Video
adapter
Network
adapter
Display
Network
SCSI bus
disk
disk
Lecture 5, 20-755: The Internet, Summer 1999
cdrom
3
IP: Internetworking with routers
•
•
IP is the most successful
protocol ever developed
Keys to success:
– simple enough to implement on top of
any physical network
» two tin cans and a string.
– rich enough to serve as the base for
implementations of more complicated
protocols and applications.
» The IP designers never dreamed
of something like the Web.
– “rough consensus and working code”
» solid implementable specs.
Many different kinds
of applications
and
higher-level
protocols
IP
Many different
kinds
of networks
The “Hourglass Model”,
Dave Clark, MIT
Lecture 5, 20-755: The Internet, Summer 1999
4
Internet protocol stack
Berkeley sockets interface
User application program (FTP, Telnet, WWW, email)
Unreliable
best effort
datagram
delivery
(processprocess)
Unreliable
best effort
datagram
delivery
(host-host)
User datagram protocol
(UDP)
Transmission control
protocol (TCP)
Internet Protocol (IP)
Reliable
byte stream
delivery
(processprocess)
Network interface (ethernet)
Lecture 5, 20-755: The Internet, Summer 1999
hardware
Physical
connection
5
IP service model
•
IP service model:
– Delivery model: IP provides best-effort delivery of datagram
(connectionless) packets between two hosts.
» IP tries but doesn’t guarantee that packets will arrive (best
effort)
» packets can be lost or duplicated (unreliable)
» ordering of datagrams not guaranteed (connectionless)
– Naming scheme: IP provides a unique address (name) for each
host in the Internet.
•
Why would such a limited delivery model be
useful?
– simple, so it runs on any kind of network
– provides a basis for building more sophisticated and userfriendly protocols like TCP and UDP
Lecture 5, 20-755: The Internet, Summer 1999
6
IP datagram delivery:
Example internet
Network 1 (Ethernet)
H1
H2
H3
H7
Network 2
(Ethernet)
R1
R2
R3
H8
Network 4
(Point-to-point)
Network 3 (FDDI)
H4
Lecture 5, 20-755: The Internet, Summer 1999
H5
H6
7
IP layering
Protocol layers used to connect host H1 to host H8 in example internet.
H1
R1
R2
R3
H8
TCP
TCP
IP
ETH
IP
ETH
IP
FDDI
FDDI
Lecture 5, 20-755: The Internet, Summer 1999
IP
P2P
P2P
IP
ETH
ETH
8
Encapsulating IP datagrams in Ethernet
IP datagram
header
IP datagram data
IP datagram
Ethernet frame IP datagram
header
header
IP datagram data
Ethernet frame
The same idea is used for other types of physical networks
Lecture 5, 20-755: The Internet, Summer 1999
9
IP packet format
0
4
8
Ver Hlen
16
TOS
31
Length
Datagram ID
TTL
19
Flags
Protocol
Offset
Checksum
Source IP address
Destination IP address
Options (variable)
VER
IP version
HL
Header length (in 32-bit words)
TOS
Type of service (unused)
Length Datagram length (max 64K B)
ID
Unique datagram identifier
Flags
xxM (more fragmented packets)
Offset
Fragment offset
TTL
Time to Live
Protocol Higher level protocol (e.g., TCP)
Data
Lecture 5, 20-755: The Internet, Summer 1999
10
Fragmentation and reassembly
•
•
Different networks types have different maximum
transfer units (MTU).
A problem can occur if packet is routed onto network
with a smaller MTU.
– e.g. FDDI (4,500B) onto Ethernet (1,500B)
•
Solution: break packet into smaller fragments.
– each fragment has identifier and sequence number
•
Destination reassembles packet before handing it up
in the stack.
– alternative would be to reassemble when entering network with
larger MTU
•
Sender can disable fragmentation using flag.
Lecture 5, 20-755: The Internet, Summer 1999
11
Fragmentation example
H1
R1
R2
R3
H8
TCP
TCP
IP
ETH
IP
ETH
ETH IP 1400
MTU=1500
IP
FDDI
FDDI
FDDI IP 1400
MTU=4500
Lecture 5, 20-755: The Internet, Summer 1999
IP
P2P
P2P
IP
ETH
ETH
P2P IP
512
ETH IP
512
P2P IP
512
ETH IP
512
P2P IP
376
ETH IP
376
MTU=532
MTU=1500
12
Fragmentation example (cont)
start of header
ident=x m=1
offset=0
rest of header
First packet
512 data bytes
start of header
ident=x m=1 offset=512
rest of header
Second packet
512 data bytes
start of header
ident=x m=0 offset=1024
rest of header
Third packet
376 data bytes
Lecture 5, 20-755: The Internet, Summer 1999
13
Internet addresses
•
•
Each host h has a physical address P(h) and
a unique IP address I(h).
IP addresses contain a network part and a
host part:
3 classes of addresses:
012
8
16
0 network(7)
10
110
24
31
Class A (128 nets, 16 M hosts/net)
host (24)
network (14)
host (16)
network (21)
Lecture 5, 20-755: The Internet, Summer 1999
host (8)
Class B (16 K nets, 65 K hosts/net)
Class C (2 M nets, 256 hosts/net)
14
Example Internet addresses
Host
IP Number
Class Network
cs.cmu.edu
128.2.222.173
cmu.edu
128.2.35.186
cs.stanford.edu 171.64.64.64
B
B
B
0x0002
0x0000
0x2640
att.com
C
0x008085
01234
0
192.128.133.151
8
16
network
10
110
24
31
host
network
Class A
host
network
Lecture 5, 20-755: The Internet, Summer 1999
Class B
host
Class C
15
IP Datagram Forwarding
•
•
•
Forwarding: the process of copying an input
packet from an input port to an output port.
Routing: the process of building the tables on
each router that allow the correct output port
to be determined (beyond our scope)
Key points
– Every IP datagram contains the IP address of the
destination.
– Network part of IP address uniquely identifies a single
physical network.
– All hosts and routers with same network field in address
are on the same physical network.
– Every physical network on the Internet has a router
connected to at least one other physical network.
Lecture 5, 20-755: The Internet, Summer 1999
16
IP Forwarding Algorithm
Algorithm for host S sending to host D:
if (NetworkNum(S) == NetworkNum(D)) {
deliver packet directly to D /* IP->physical mapping via ARP */
else
deliver packet to default router
Algorithm for router receiving packet for host D
NextHop = lookup(NetworkNum(D));
if (NextHop is an interface)
deliver packet directly to D using interface NextHop
else
if (NextHop != <undefined>)
deliver packet to NextHop (a router)
else
deliver packet to default router
Lecture 5, 20-755: The Internet, Summer 1999
Forwarding table
consists of
(NetworkNum,
NextHop) pairs
17
IP Forwarding example
H1
H2
Network 1 (Ethernet)
H3
H7
Network 2
(Ethernet)
R1
R2
R3
H8
Network 4
(Point-to-point)
Network 3 (FDDI)
H4
Router R2
forwarding
table
NetworkNum
1
2
3
4
Lecture 5, 20-755: The Internet, Summer 1999
H5
H6
NextHop
R3
R1
Interface 1
Interface 0
18
ARP: Address resolution protocol
•
Initially:
– Hosts S and D on the same network with IP
addresses I(S) and I(D) and physical
addresses P(S) and P(D).
•
Problem:
– Given I(D), host S wants to discover P(D).
•
(I(S), P(S), I(D), ???)
S
Solution:
– Host S broadcasts triple (I(S), P(S),
I(D),???) on network.
– Host D (and only host D) responds with
tuple (I(S), P(S), I(D), P(D))
– Both sender and receiver maintain a
software cache of IP to physical mappings.
– Time out old entries
Lecture 5, 20-755: The Internet, Summer 1999
D
(I(S), P(S), I(D), P(D))
S
D
19
Subnetting
•
•
Problem: IP addressing scheme makes
inefficient use of addresses
Partial solution: subnetting
– physical network part of address identifies a “virtual”
physical network to the external world.
– use some of the high order “host” bits to identify local
physical networks within the “virtual” physical network.
network number
host number
11111111 11111111 11111111 00000000
xxxxxxxx xxxxxxxx xxxxxxxx 00000000
Class B address
&
Subnet mask (255.255.255.0)
=
Subnet number
- All hosts on same physical network have same subnet number.
- There is exactly one subnet mask per subnet.
- All hosts on subnet configured with this mask (ifconfig)
Lecture 5, 20-755: The Internet, Summer 1999
20
IP forwarding with subnetting
Algorithm on a host:
D1 = SubnetMask & destination IP address
if (D1 == MySubnetNum)
deliver datagram directly to destination
else
deliver datagram to default router
Algorithm on a router:
for each forwarding table entry <SubnetNum,SubnetMask,NextHop>
D1 = SubnetMask & destination IP address
if (D1 == SubnetNum)
if (NextHop is an interface)
deliver datagram directly to destination
else
deliver datagram to NextHop (a router)
Lecture 5, 20-755: The Internet, Summer 1999
21
Subnetting example
subnet mask: 255.255.255.128
subnet number: 128.96.34.0
128.96.34.1
128.96.34.15
H1
R1
128.96.34.130
128.96.34.129
R2
subnet mask: 255.255.255.128
subnet number: 128.96.34.128
128.96.34.139
H2
128.96.33.1
128.96.33.14
H3
forwarding
table for R1
subnet mask: 255.255.255.0
subnet number: 128.96.33.0
SubnetNum
128.96.34.0
128.96.34.128
129.96.33.0
Lecture 5, 20-755: The Internet, Summer 1999
SubnetMask
255.255.255.128
255.255.255.128
255.255.255.0
NextHop
interface 0
interface 1
R2
22
IPv6
•
•
•
3
010
•
Also called Next Generation IP and IPng
Extends address space from 32 bits to 128
bits
Hierarchical address space:
registryID
providerID
SubscriberID
SubnetID
48
InterfaceID
neat feature
– embedded InterfaceID allows host to assign itself an IP address!
Lecture 5, 20-755: The Internet, Summer 1999
23
IPv6 packet format
4
8
Ver Pri
16
24
31
FlowLabel
PayloadLen
NextHdr HopLimit
Source Address
Ver
Pri/Flowlabel
PayloadLen
NextHdr
HopLimit
Source Address
Dest Address
IP version (6)
Quality of Service)
packet len (max 64KB)
optional/encapsulated
header type
same as TTL in IPv4
128-bit source addr
128-bit dest addr
Destination Address
Optional header examples:
fragmentation (44)
authentication (51)
TCP (6)
Next header/data
Lecture 5, 20-755: The Internet, Summer 1999
24
Converting from IPv4 to IPv6
•
•
Not possible to have a “flag day”
Must upgrade incrementally
– dual stack operation
» IPv6 nodes run both IPv4 and IPv6 protocol stacks
– IP tunneling
» IP packet sent as payload of another IP packet
» networking community’s version of indirection!
IPV6
IPv6
router
IPv6
router
IPv4 network
IPV6
IPV4
Lecture 5, 20-755: The Internet, Summer 1999
IPV6
IPV6
IPV4
25
Break time!
Lecture 5, 20-755: The Internet, Summer 1999
26
Today’s lecture
•
•
•
IP: Internetworking with routers (50 min)
Break (10 min)
UDP, TCP (35 min)
Lecture 5, 20-755: The Internet, Summer 1999
27
UDP: User datagram protocol
Berkeley sockets interface
User application program (FTP, Telnet, WWW, email)
Unreliable
best effort
datagram
delivery
(processprocess)
Unreliable
best effort
datagram
delivery
(host-host)
User datagram protocol
(UDP)
Transmission control
protocol (TCP)
Internet Protocol (IP)
Reliable
byte stream
delivery
(processprocess)
Network interface (ethernet)
Lecture 5, 20-755: The Internet, Summer 1999
hardware
Physical
connection
28
UDP: User datagram protocol
•
•
•
•
Extends IP to provide process-to-process
(end-to-end) datagram delivery
Mechanism for demultiplexing IP packets
Based on port abstraction
Process identified by <host, port> pair.
SrcPort
DstPort
CheckSum
Length
Data
Lecture 5, 20-755: The Internet, Summer 1999
29
TCP: Transmission Control Protocol
Berkeley sockets interface
User application program (FTP, Telnet, WWW, email)
Unreliable
best effort
datagram
delivery
(processprocess)
Unreliable
best effort
datagram
delivery
(host-host)
User datagram protocol
(UDP)
Transmission control
protocol (TCP)
Internet Protocol (IP)
Reliable
byte stream
delivery
(processprocess)
Network interface (ethernet)
Lecture 5, 20-755: The Internet, Summer 1999
hardware
Physical
connection
30
TCP: Transmission control protocol
•
Uses IP to provide reliable process-to-process byte
stream delivery.
– stream orientation
» sender transfers ordered stream of bytes; receiver gets identical
stream
– virtual circuit connection
» stream transfer analogous to placing phone call
» sender initiates connection which must be accepted by receiver.
– buffered data transfer
» protocol software free to use arbitrary size transfer units
– unstructured streams
» stream is a sequence of bytes, just like Unix files
– full duplex
» concurrent transfers in both directions along a connection
Lecture 5, 20-755: The Internet, Summer 1999
31
TCP functions
•
•
•
•
Connections
Sequence numbers
Sliding window protocol
Reliability and congestion control.
Source Port
Dest. Port
Sequence Number
Acknowledgment
Hlen/Flags
Window
D. Checksum
Urgent Pointer
Options..
Lecture 5, 20-755: The Internet, Summer 1999
32
Connections
•
Connection is a fundamental TCP communication
abstraction.
– data sent along a connection arrives in order
– implies allocation of resources (buffers) on hosts
•
The endpoint of a connection is a pair of integers:
– (IP address, port)
•
A connection is defined by a pair of endpoints:
– ((128.2.254.139, 1184), (128.10.2.3, 53))
(128.2.254.139, 1184)
Lecture 5, 20-755: The Internet, Summer 1999
connection
(128.10.2.3, 53)
33
Sequence space
•
•
Each stream split into a sequence of
segments which are encapsulated in IP
datagrams.
Each byte in the byte stream is numbered.
– 32 bit value
– wraps around
– initial values selected at runtime
•
Each segment has a sequence number.
– indicates the sequence number of its first byte
– Detects lost, duplicate or out of order segments
Lecture 5, 20-755: The Internet, Summer 1999
34
TCP flow control mechanism:
sliding window
•
•
The purpose of flow control is to keep
senders from flooding receivers with packets
and filling up their memories.
Often confused with congestion control,
which tries to keep the senders from flooding
the network with packets.
Lecture 5, 20-755: The Internet, Summer 1999
35
Sliding window protocol (sender)
•
Sender maintains a “window” of unacknowledged
bytes that it is allowed to send, and a pointer to the
last byte it sent:
current window
1 2 3 4 5 6 7 8 9 10 11 ...
left
curr
byte stream
right
Bytes through 2 have been sent and acknowledged (and thus can be discarded)
Bytes 3 -- 6 have been sent but not acknowledged (and thus must be buffered)
Bytes 7 -- 9 have been not been sent but will be sent without delay.
Bytes 10 and higher cannot be sent until the right edge of window moves.
Lecture 5, 20-755: The Internet, Summer 1999
36
Sliding window protocol (receiver)
•
Receiver acknowledges receipt of a segment with two
pieces of information:
– ACK: the sequence number of the next byte in the contiguous
stream it has already received
– WIN: amount of available buffer space.
•
ACK indicates that data was received correctly.
– sender can increment left edge of window
– sender can delete data to the left of the window.
•
WIN indicates that more buffer space was freed up.
– sender can increment the right edge of its window
– sender can transmit more data.
Lecture 5, 20-755: The Internet, Summer 1999
37
Sliding window protocol (example)
Sender
Receiver’s buffer
Receiver
Application
does 2K write
0
4K
2K, SEQ = 0
empty
ACK=2K, WIN = 2K
2K
Application
does 3K write
2K, SEQ =2K
ACK=4K, WIN = 0
4K
Sender
is blocked
Application
reads 2K
ACK=4K, WIN = 2K
2K
Sender may
send up to 2K
1K, SEQ =4K
1K
Lecture 5, 20-755: The Internet, Summer 1999
2K
38
Opening and closing connections
Host 1
The three way handshake
Application
does a connect to
a socket on Host 2
SYN, SEQ = J, WIN = 4K
ACK =J +1, SYN, SEQ = K, WIN = 4K
ACK = K+1,
Application
does a close on a
connection
FIN, SEQ = M
Host 2
J is the initial
sequence number
for messages from
Host 1 to Host 2.
K is the initial
sequence number
for messages from
Host 2 to Host 1.
SYN is the
“synchronize” flag
ACK = M+1
FIN, SEQ = N
ACK = N+1
Lecture 5, 20-755: The Internet, Summer 1999
Host 2 replies
with its own close.
FIN is the “finish”
flag
39
Reliability and congestion control
•
Reliability:
– sender
» saves segments inside its window
» uses timeouts and sequence numbers in ACKS to detect lost
segments.
» retransmit segments it thinks are lost
– receiver
» uses sequence numbers to assemble segments in order
» also to detect duplicate segments (how might this happen?)
•
Congestion control
– sender maintains separate separate congestion window
– uses smaller of the two windows
– uses “slow start” algorithm to adaptively set congestion window
size.
Lecture 5, 20-755: The Internet, Summer 1999
40
End-to-end data issues
•
Presentation formatting
– must account for different data formats on different
machines
» different byte orders
» different word sizes
•
Compression
– data can be compressed/decompressed on the endpoints
to save network bandwidth (beyond our scope)
•
Encryption
– sensitive data can be encrypted/unencrypted on the
endpoints.
•
Authentication
– Receivers may want to verify that messages really do
come from the sender.
Lecture 5, 20-755: The Internet, Summer 1999
41
Key themes in IP internetworking
•
Protocol layering
– Way to structure complex system
– Handle different concerns at different layers
•
•
•
Must cope with heterogeneous networks
Must cope with huge scale
Must cope with imperfect environment
– Packets get corrupted and lost
•
No one has complete routing table
– Too many hosts
– Hosts continually being added and removed
– In the future, they will start moving around (mobile
computing)
Lecture 5, 20-755: The Internet, Summer 1999
42
Next time: Programming the
global IP Internet
Berkeley sockets interface
User application program (FTP, Telnet, WWW, email)
Unreliable
best effort
datagram
delivery
(processprocess)
Unreliable
best effort
datagram
delivery
(host-host)
User datagram protocol
(UDP)
Transmission control
protocol (TCP)
Internet Protocol (IP)
Reliable
byte stream
delivery
(processprocess)
Network interface (ethernet)
Lecture 5, 20-755: The Internet, Summer 1999
hardware
Physical
connection
43