CSE390 – Advanced

Computer Networks
Lecture 22: Software designed
networking

Based on slides by J. Rexford @ Princeton & N. Mckeown @ Stanford

& S. Shenker @ Berkeley. Updated by P. Gill Fall 2014.
Data, Control, and Management
Planes

2
 Timescales
Data Control Management
Time- Packet Event (10 Human (min
scale (nsec) msec to sec) to hours)
Tasks Forwarding, Routing, Analysis,
buffering, circuit configuration
filtering, set-up
scheduling
Location Line-card Router Humans or
hardware software scripts
3
 Data and Control Planes
control plane
data plane Processor

Line card Line card

Line card
Switching
Line card
Fabric

Line card Line card

4
 Data Plane
• Streaming algorithms on packets
– Matching on some bits
– Perform some actions
• Wide range of functionality Processor
– Forwarding
– Access control
– Mapping header fields
– Traffic monitoring Switching
– Buffering and marking Fabric

– Shaping and scheduling

– Deep packet inspection
5
Switch: Match on Destination MAC

• MAC addresses are location independent

– Assigned by the vendor of the interface card
– Cannot be aggregated across hosts in LAN

mac1 mac2 mac3

host host ... host mac1
mac2

switch
mac3
host
mac4
mac5
host mac5
mac4
6
 Router: Match on IP Prefix
• IP addresses grouped into common subnets
– Allocated by ICANN, regional registries, ISPs,
and within individual organizations
– Variable-length prefix identified by a mask length
1.2.3.4 1.2.3.7 1.2.3.156 5.6.7.8 5.6.7.9 5.6.7.212
host host ... host host host ... host

LAN 1 LAN 2
router router router
WAN WAN

1.2.3.0/24
Prefixes may be nested.
5.6.7.0/24 Routers identify the
forwarding table longest matching prefix. 7
 Forwarding vs. Routing
• Forwarding: data plane
– Directing a data packet to an outgoing link
– Individual router using a forwarding table
• Routing: control plane
– Computing paths the packets will follow
– Routers talking amongst themselves
– Individual router creating a forwarding
table

8
 Example: Shortest-Path Routing
• Compute: path costs to all nodes
– From a source u to all other nodes
– Cost of the path through each link
link
– Next hop along least-cost path to s
v (u,v)
v 2 y
w (u,w)
1
3
1
x 4
x (u,w)
u z
2 y (u,v)
w 1
5 t
z (u,v)
6
4 3 s (u,w)
s
t (u,w) 9
 Distributed Control Plane
• Link-state routing: OSPF, IS-IS
– Flood the entire topology to all nodes
– Each node computes shortest paths link
– Dijkstra’s algorithm
v (u,v)
v 2 y w (u,w)
3 1
1 x (u,w)
x 4
u
2
z y (u,v)
1
5 t z (u,v)
w 4 3 s (u,w)
s
t (u,w) 10
 Distributed Control Plane
• Distance-vector routing: RIP, EIGRP
– Each node computes path cost
– … based on each neighbors’ path cost
– Bellman-Ford algorithm
v 2 y du(z) = min{c(u,v) + dv(z),
3 1
1 c(u,w) + dw(z)}
x 4
u z
2 1
5 t
w 4 3
s
11
 Traffic Engineering Problem
• Management plane: setting the weights
– Inversely proportional to link capacity?
– Proportional to propagation delay?
– Network-wide optimization based on traffic?

2
3 1
1
3
2 3
1 5

4 3
12
 Traffic Engineering: Optimization
• Inputs
– Network topology
2
– Link capacities 3 1
1
– Traffic matrix 3
2 1 5
• Output
– Link weights 4 3
• Objective
– Minimize max-utilized link
– Or, minimize a sum of link congestion

13
 Transient Routing Disruptions
• Topology changes
– Link weight change
– Node/link failure or recovery
• Routing convergence
– Nodes temporarily disagree how to route
– Leading to transient loops and blackholes

1 5 1 10 1 10

4 3 4 3 4 3
14
 Management Plane Challenges
• Indirect control
– Changing weights instead of paths
– Complex optimization problem
• Uncoordinated control
– Cannot control which router updates first
• Interacting protocols and mechanisms
– Routing and forwarding
– Naming and addressing
– Access control
– Quality of service
– …
15
Software Defined Networking
(high level view)

16
Control/Data Separation
decouple control and data planes
by providing open standard API

17
(Logically) Centralized Controller
Controller Platform

18
Protocols  Applications
Controller Application
Controller Platform

19
Outline
20

1. What are Software Defined

Networks?
2. Why SDN?
3. The Consequences
 For industry
 For research
 For standards and protocols
21

Open Interface

or or

Open Interface

Vertically integrated Horizontal

Closed, proprietary Open interfaces
Slow innovation Rapid innovation
Small industry Huge industry
22
22

Open Interface

or or

Open Interface

Vertically integrated Horizontal

Closed, proprietary Open interfaces
Slow innovation Rapid innovation
23

Routing, management, mobility management,

access control, VPNs, …

Million of lines 6,000 RFCs

of source code

Billions of gates Bloated Power Hungry

• Vertically integrated, complex, closed,

proprietary
• Networking industry with “mainframe” mind-set
The network is changing
24

Feature Feature

Feature Feature

Feature Feature

Feature Feature

Feature Feature
 Software Defined Network (SDN)
3. Consistent, up-to-date global network view 2. At least one Network OS
25 probably many.
Open- and closed-source

1. Open interface to packet forwarding

Packet
Forwarding Packet
Forwarding

Packet
Packet Forwarding
Forwarding Packet
Forwarding
Network OS
26
Network OS: distributed system that creates a
consistent, up-to-date network view
 Runs on servers (controllers) in the network
 NOX, ONIX, Trema, Beacon, Maestro, … + more

Uses forwarding abstraction to:

 Get state information from forwarding elements
 Give control directives to forwarding elements
Software Defined Network (SDN)

Packet
Forwarding Packet
Forwarding

Packet
Packet Forwarding
Forwarding
Packet
Forwarding
Control Program
28

 Control program operates on view of network

 Input:
global network view (graph/database)
 Output: configuration of each network device

 Control program is not a distributed system

 Abstraction hides details of distributed state
Forwarding Abstraction
29

Purpose: Abstract away forwarding hardware

 Flexible

 Behavior specified by control plane

 Built from basic set of forwarding primitives

 Minimal
 Streamlined for speed and low-power
 Control program not vendor-specific

 OpenFlow is an example of such an abstraction

 OpenFlow Basics
30

OpenFlow Protocol

Ethernet Switch
Control Path OpenFlow

Data Path (Hardware)

 OpenFlow Basics
31

“If header = p, send to port 4”

Packet “If header = q, overwrite header with r,
Forwarding add header s, and send to ports 5,6”
“If header = ?, send to me”

Flow
Packet Table(s)
Forwarding Packet
Forwarding
Plumbing Primitives
32
32

 Primitive is <Match, Action>

 Match arbitrary bits in headers:
Header Data

Match: 1000x01xx0101001x

 Match on any header, or new header

 Allows any flow granularity
 Action
 Forward to port(s), drop, send to controller
 Overwrite header with mask, push or pop
 Forward at specific bit-rate
General Forwarding Abstraction
33

Small set of primitives

“Forwarding instruction set”

Protocol independent
Backward compatible

Switches, routers, WiFi APs,

basestations, TDM/WDM
Example 1: OSPF and Dijkstra
34

 OSPF
 RFC 2328: 245 pages
 Distributed System
 Builds consistent, up-to-date map of
the network: 101 pages
 Dijkstra’s Algorithm
 Operates on map: 4 pages
 Example
35

Packet
Forwarding

Packet
Packet Forwarding
Forwarding Packet
Forwarding
Outline
36

1. What are Software Defined

Networks?
2. Why SDN?
3. The Consequences
 For industry
 For research
 For standards and protocols
GREAT TALK BY SCOTT
SHENKER
HTTP://WWW.YOUTUBE.COM/WATCH?V=WVS7
PC99S7W

(Story summarized here)

Networking
38

 Networking is
 “Intellectually
Weak”
 behind other fields

 about the mastery of complexity

Good abstractions tame complexity

 Interfaces are instances of those abstractions

No abstraction => increasing complexity

 We are now at the complexity limit
By comparison: Programming
39

 Machine languages: no abstractions

 Had to deal with low-level details
 Higher-level languages: OS and other abstractions
 File system, virtual memory, abstract data types, ...
 Modern languages: even more abstractions
 Object orientation, garbage collection,…
Programming Analogy
40

 What if programmers had to:

 Specify where each bit was stored
 Explicitly deal with internal communication errors

 Within a programming language with limited expressibility

 Programmers would redefine problem by:

 Defining higher level abstractions for memory
 Building on reliable communication primitives

 Using a more general language

Specification Abstraction
41

 Network OS eases implementation

 E.g., Helps manage distributed state
 Next step is to ease specification
 E.g., How do you specify what the system should do?

 Key goals
 Provide abstract view of network map
 Control program operates on abstract view

 Develop means to simplify specification

 Software Defined Network (SDN)
42

Abstract Network View

Global Network View

Packet
Forwarding Packet
Forwarding

Packet
Packet Forwarding
Forwarding Consequence:
Packet
Work on Nework Programming Languages Pyretic, Frenetic etc.
Forwarding
Outline
43

1. What are Software Defined

Networks?
2. Why SDN?
3. The Consequences
 For industry
 For research
 For standards and protocols
SDN in development
44

Domains Products
 Data centers  Switches, routers:
 Enterprise/campus About 15 vendors
 Cellular backhaul  Software: About 6

 Enterprise WiFi
vendors and startups
 WANs

New startups (6 so far). Lots of hiring in networking.

Cellular industry
46

 Recently made transition to IP

 Billions of mobile users
 Need to securely extract payments and hold users
accountable
 IP is bad at both, yet hard to change

SDN enables industry to customize their network

Telco Operators
47

 Global IP traffic growing 40-50% per year

 End-customer monthly bill remains unchanged
 Therefore, CAPEX and OPEX need to reduce 40-50%
per Gb/s per year
 But in practice, reduces by ~20% per year

SDN enables industry to reduce OPEX and CAPEX

…and to create new differentiating services
Example: New Data Center
48

Cost Control
200,000 servers
Fanout of 20  10,000 switches More flexible control
$5k vendor switch = $50M Tailor network for services
$1k commodity switch = $10M Quickly improve and innovate

Savings in 10 data centers = $400M

Consequences for research
49

Ease of trying new ideas

 Existingtools: NOX, Beacon, switches, Mininet
 More rapid technology transfer

 GENI, Ofelia and many more

A stronger foundation to build upon

 Provableproperties of forwarding
 New languages and specification tools
Consequences for standards
50

Standards will define the interfaces

The role of standards will change:

 Network owners will define network behavior
 Features will be adopted without standards

Programming world
 Good software is adopted, not standardized
Summary
51

 Networks becoming
 More programmatic
 Defined by owners and operators, not vendors

 Faster changing, to meet operator needs

 Lower opex, capex and power

 Abstractions
 Will shield programmers from complexity
 Make behavior more provable

 Will take us places we can’t yet imagine

Administravia …
52

 Assignment 4 due December 13

 Internet in the News (10% of final grade)
 Due next Monday Dec. 1 on Piazza
 Reading/commenting on others’ Internet in the News part of
participation mark
 Recent news: http://www.newsweek.com/china-could-shut-
down-us-power-grid-cyber-attack-says-nsa-chief-286119
 Lots of topics, pick something you find interesting

 No class Wednesday!
 Next Monday  Mobile networks!