Professional Documents
Culture Documents
w1 One
w1 One
CS 552
Badri Nath
Rutgers University
badri@cs.rutgers.edu
FALL 2013
1
amrutag@cs.rutgers.edu
Course info
l
http://www.cs.rutgers.edu/~badri/552.html
Course Goals
l
Course Materials
l
Research papers
Links to pdf on Web page
l Combination of classic and recent work
l ~50 papers
l Optional readings
l
Recommended textbooks
For students not familiar with networking
l Peterson & Davie (4th edition)
l Alternative: Kurose & Ross
l
Reading papers
l
l
l
l
Books
l
l
l
l
l
l
Grading
l
l
l
l
l
Class Coverage
Quick overview of undergraduate networking
l Pre requisite: 352 or equivalent
l Students expected to know
l
Protocol performance
l Investigate protocol trade-offs, cost models
l New Workloads, new technologies
l
10
Class etiquette
Cell phones in off position
l No FB status updates in class
l If you need to surf while in class (I prefer not),
do not disturb your neighbors
l Stop me anytime to ask questions
l
Lecture Topics
l
l
l
l
l
Traditional
Layering
Internet architecture
Routing (IP)
Transport (TCP)
Protocols
(HTTP,DHCP,DNS)
l
l
l
l
l
Recent Topics
Data center networking
Cloud computing
Mobility/wireless
Enterprise Networks
Software defined
networking
Energy considerations
12
What is a Network?
l
l
l
l
l
l
l
l
l
l
l
l
copper wire
Lasers (optic fibre)
Microwave
Cable (coax)
wireless
satellite link
13
Why Networks?
l
Availability of Resources
l
Load Sharing/utilization
l
l
High Reliability
l
Human-to-Human Communication
l
14
What is an internet?
l
Network of networks
A global internet based on the IP protocol
15
Electronic mail
WEB
File transfer, sharing
Social networks (FB, linkedin,
twitter)
On-line commerce
Search
Resource distribution (hosting)
Video conferencing
Games
16
Interactive entertainment
l
The good
l
The bad
l
The ugly
l
18
Impact on society
Net neutrality
l Laws and censorship (SOPA ---)
l
l
l
Google vs China
Wiki going black -- protest web censorship
Regulation
Internet Players
l
Designers
l
l
Service Providers
l
l
l
l
Internet Traffic
21
Internet Growth
22
23
24
25
26
Social Networking
27
28
Drpeering.net
WWW Growth
Log scale
10000000
1000000
100000
10000
WWW
1000
100
10
Jul-99
Jul-98
Jul-97
Jul-96
Jul-95
Jul-94
Jul-93
29
Source:www.isoc.org/guest/zakon/Internet/History/HIT.html
WWW growth
30
31
Facebook growth
32
33
34
35
36
37
Content Growth
38
39
40
41
42
43
44
45
46
Search
Connections
Real-time updates
Device State?
47
Whats hot?
Software Defined
Networking (SDN)
Mobile &Wireless
Embedded
Internet of connected things
Internet
Crowd Sourcing
Hybrid computing
Social Networking
48
Integrating Networks
Social Networks
Vehicular Networks
Computer Networks
Mobile Networks
49
Computer networks
l
Social Networks
l
Vehicular networks
l
Computing nodes
Automobiles
Mobile networks
l
Smart phones
50
Social computing
Can we use humans and computers?
l Exploit humans
l
Exploit computers
l
Exploit both
l
Socio-computational systems
51
Objective of networking
l
End-hosts to communicate
l
Different technologies
l Different protocols
l
52
How to communicate?
Circuit switching
l Establish a connection before communicating
l
l
l
Packet switching
l
l
l
l
l
l
IP
A set of rules with a well-defined interface
54
l
l
55
Corruption?
l
Reliability
l
Data lost?
Overload
l
Congestion control
Security
l
Encryption, authentication
56
57
Host B
Application
Layer
Application Protocol
Presentation
Layer
Presentation Protocol
Session
Layer
Session Protocol
Transport
Layer
Transport Protocol
Application
Layer
Presentation
Layer
Session
Layer
Transport
Layer
Network
Layer
Network
Layer
Network
Layer
Network
Layer
Data Link
Layer
Data Link
Layer
Data Link
Layer
Data Link
Layer
Physical
Layer
Physical
Layer
Physical
Layer
Physical
Layer
Router
Router
58
Problems
Seven layers not widely accepted
l Standardized before implemented
l Top three layers fuzzy
l Internet or TCP/IP layering widespread
l
59
Application
A simplified model
The network layer
l
Transport (TCP/UDP)
Internet/Network
(IP)
Host-to-Net
60
l
l
VOIP (SIP), IM
Cathedral vs Bazaar
David Clark, The design philosophy of the DARPA internet protocols, 1998
61
Implications (cathedral vs
bazaar)
l
Bazaar
l
l
Cathedral
l
l
l
Edge is programmable
Nimble, novel applications
Core elements still rigid
Standards, slow evolution
Cant do anything radical
# 2 Multiplexed utilization
l
l
TCP
Transport
IP
Network Interface
64
Host A
Host B
Application Protocol
http
TCP
Network
Layer
Host-toNet Layer
http
IP
Ethernet
Network
Layer
IP
Host-toNet Layer
Network
Layer
Host-toNet Layer
OC3
Ethernet
IP
Ethernet
Network
Layer
Host-toNet Layer
65
#4 distributed management
Need only know local information
l Information distributed over different nodes
l
l
l
Scalable
No single hot spot
Different entities manage different parts of the
system
Impact on naming, routing, addressing
Scale
l Protocols should work in networks of all sizes and
distances
Incremental deployment
l New protocols need to be deployed gradually
Heterogeneity
l Different technologies, autonomous organizations
End-to-end argument
l Networking functions should be delegated to the edges;
application knows best
67
End-to-end argument
l
l
l
68
Communication system
An end system connected by a
communication subsystem
l Questions?
l Who is responsible for a given function
l
Subsystem?
l End units?
l Or both (redundant) or jointly?
l
69
End-to-end argument
Functions placed at lower level implies
specific problems being solved in a general
way
l Best aim:
l Simple lower layer with smart end points
l
Gives flexibility
70
e2e argument
l
l
l
l
l
l
Performance
Duplicated effort
Should not impact
applications that do not use
that functionality
71
e2e tradeoffs
l
Network redirection
l
72
New metrics
l
l
System Performance/availability
l
l
Performance tools
l
74
75
The cost of a cloud: Research Problems in data center networks by Albert Greenberg et.al, CCR
Datacenter cost
Servers
45%
l
l
l
l
l
Infrastructure
25%
Network
15%
Power
15%
Power cost
Power to run the IT equipment
Power to run cooling, UPS etc Overhead
PUE=Total power /IT power
1.2 ideal -- 20% overhead
Typically 2 to 3 PUE Air conditioning costs enormous
76
77
Energy-proportional metric
l
78
Figure 1. Average CPU utilization of more than 5,000 servers during a six-month period. Servers
79
are rarely completely idle and seldom operate near their maximum utilization, instead operating79
most of the time at between 10 and 50 percent of their maximum
Energy Efficiency =
Utilization/Power
Figure 2. Server power usage and energy efficiency at varying utilization levels, from idle to
peak performance. Even an energy-efficient server still consumes about half its full power
when doing virtually no work.
80
80
Doing nothing
VERY well
Design for
wide dynamic
power range and
active low power
modes
Energy Efficiency =
Utilization/Power
Figure 4. Power usage and energy efficiency in a more energy-proportional server. This
server has a power efficiency of more than 80 percent of its peak value for utilizations of
30 percent and above, with efficiency remaining above 50 percent for utilization levels as
81
low as 10 percent.
81
If Cost/Green Conscious
Make network elements less power hungry
l
l
l
Approximate Number
Deployed
Total
AEC TW-h
Hubs
93.5 Million
1.6 TW-h
LAN
Switch
WAN
Switch
Router
95,000
3.2 TW-h
50,000
0.15TW-h
3,257
1.1 TW-h
84
More Numbers
l
85
Enterprise networks
l
86
87
l
l
88
Memory
Main CPU
Switch fabric or bus backplane
Line cards (designs ranging from simple to
complex with ASICs or network processors to
process packets)
Bang for the buck
89
Key Questions
l
should not trigger network reconfiguration in sleep state vs. failure state
90
More on sleeping.
Uncoordinated sleeping:
l
l
l
l
Coordinated sleeping:
l
l
l
l
l
l
Computation Placement
Energy cost varies by location
l Energy cost varies by hour of the dat
l Can we push computation to a
geographically distant place to save energy?
l Can we exploit time difference?
l Peak vs non-peak power rates
l
92
Coal vs nuclear
93
Research
l
l
The
Featur
e
Current Network
Featur
e
Operating
System
Specialized Packet
Forwarding Hardware
Million of
lines
of source
code
Billions of
gates
Bloated
Power Hungry
An
industry
with
a
mainframe-mentality,
reluctant
to
change
95
Slides
from Stanford Site: Nick Mckeown, Martin Casado, Scott Shenker et al,
Load balancer
Firewall
Wireless Router
Layer 2 Switch
NAT box
96
Featur
e
Feature
Network OS
1. Open interface to
hardwareOpenFlow
Simple
Packet
Forwarding
Hardware
Simple
Packet
Forwarding
Hardware
Simple
Packet
Forwarding
Hardware
Simple
Packet
Forwarding
Hardware
Simple
Packet
Forwarding
Hardware
98
Step 1:
Separate Control from Datapath
Network OS
OpenFlow
Switch
OpenFlow
Switch
OpenFlow
Switch
OpenFlow
Switch
99
Flow
OpenFlow
Table
Switch
OpenFlow
Switch
OpenFlow
Switch
OpenFlow
Switch
100
Action
Stats
Packet + byte counters
1.
2.
3.
4.
Switch
MAC
MAC
Port
src
dst
+ mask
IP
Src
IP
Dst
IP
Prot
TCP
sport
TCP
dport
Plumbing Primitives
1.
Header
Match: 1000x01xx0101001x
l
l
2.
Actions:
l
l
l
Examples
Switching
Switch
MAC
Port
src
*
MAC
Eth
dst
type
00:1f:..
*
VLAN
IP
ID
Src
IP
Dst
IP
Prot
TCP
TCP
Ac/on
sport
dport
IP
Dst
IP
Prot
TCP
TCP
Ac/on
sport
dport
port6
Flow
Switching
Switch
MAC
Port
src
MAC
Eth
dst
type
VLAN
IP
ID
Src
17264 80
port6
Firewall
Switch
MAC
Port
src
*
MAC
Eth
dst
type
*
VLAN
IP
ID
Src
IP
Dst
IP
Prot
TCP
TCP
Forward
sport
dport
22
drop
Examples
Rou/ng
Switch
MAC
Port
src
*
MAC
Eth
dst
type
VLAN
IP
ID
Src
IP
Dst
5.6.7.8 *
VLAN
IP
ID
Src
IP
Dst
IP
Prot
vlan1 *
TCP
TCP
Ac/on
sport
dport
port6,
port7,
*
*
port9
IP
Prot
TCP
TCP
Ac/on
sport
dport
*
port6
VLAN
Switching
Switch
MAC
Port
src
*
MAC
Eth
dst
type
00:1f..
*