6.

888
Advanced Topics in Networking

Lecture 1: Introduction

Mohammad Alizadeh

Spring 2016

✧ Includes material from lectures by Nick McKeown (Stanford), Jennifer Rexford (Princeton),
and George Porter (UCSD)
 The Internet: An Exciting Time
One of the most influential inventions
– A research experiment that escaped from the lab
– … to be the global communications infrastructure

Ever wider reach

– Today: 2 billion users, 15 billion devices
– Tomorrow: more users, content, sensors, “things”,
40 billion devices by 2020

Constant innovation
– Web, P2P, video, online shopping, social networks, cloud, …

2
 Transforming Everything
The ways we do business
– E-commerce, advertising, cloud computing, ...
The way we have relationships
– E-mail, IM, Facebook friends, virtual worlds
The way we think about law and govern
– Interstate commerce, national boundaries?
– Censorship and wiretapping
The way we fight
– Cyber-attacks, including nation-state attacks

3
But what is networking?

4
A Plethora of Protocol Acronyms?
SNM WA IP
P SI PP MA
P FT P X
LLD P C
P UD
P HI
OSP RT P ICM IMA IGM P
F P P P P
BG
PI HTT EC
RE P AR
M P N
D I P TC
RI MPL RTC
P P P
SMT S BF CID P
P RTS
NNT P D R
SAC TL NA
P STU
K SS S T
DN
H DHC N
S VLA VT
PO LIS TFT P LD 5
A Heap of Header Formats?

6
TCP/IP Header Formats in Lego

7
 A Big Bunch of Boxes?
Label Load
Router Switch
Switched balancer
Router Scrubber
Repeater
Gateway Bridge
Intrusion Route
Deep Detection Reflector
Packet System
Inspection DHCP
Firewall server Packet
NAT Hub shaper
Packet
DNS sniffer
WAN Base Proxy
accelerator server station 8
An Application Domain?

9
 A place to apply theory?
Algorithms and data structures
Control theory
Queuing theory
Optimization theory
Game theory and mechanism design
Formal methods
Cryptography
Programming languages
Graph theory 10
 A place to build systems?
Distributed systems
Operating systems
Computer architecture
Software engineering
…

11
 So, Why is Networking Cool?
Relevant
– Can impact the real world
– Can measure/build things

Interdisciplinary
– Well-motivated problems + rigorous solution techniques

Widely-read papers
– Many of the most cited papers in CS are in networking
– Congestion control, distributed hash tables, resource
reservation, self-similar traffic, multimedia protocols
12
 So, Why is Networking Cool?
Young, relatively immature field
– Tremendous intellectual progress is still needed
– You can help decide what networking really is

Defining the problem is a big part of the challenge

– Recognizing a need, formulating well-defined problem
– … is at least as important as solving the problem.

Lots of platforms for building your ideas

– Testbeds: Emulab, PlanetLab, Orbit, GENI
– Programmability: Click, Mininet, NetFPGA, Switch chips
13
 This course
... is about the latest in networking research

Main goal:
Prepare for high quality research in this field

14
 We’ll focus mostly on…
Data Center Networking

Software Defined Networking

Two “hot” areas of research

Significant interest in both academia & industry
Lots of opportunities for impact
15
DCN/SDN Papers at SIGCOMM

50
45
40
35
# of Papers

30
25 Other
20 DCN / SDN
15
10
5
0
2010 2011 2012 2013 2014 2015

16
 Readings & Presentations
Each group of two students will present one paper on
12 November 2021 (offline class)
– Read paper and relevant references
– 15 minute talk
– https://forms.gle/TWcJm3ZjbrDdXJ7M7

17
 Projects
Research project of your choice

Project ideas
– Explore your own
– I will also suggest some ideas
– Can involve system implementation, algorithms, theory, …
– Can be related to your research (come talk to me)

18
 Project Timeline
Proposal (1 page) Oct 30

Paper presentation Nov 12

Project Report (4-6 pages) Jan 7

Project Presentation Jan 14 (tentative)

19
 Grading

Class participation 10%

Attitude 15%
Paper Presentation 25%
Midterm 25%
Project 25%

20
Data Center Networking

21
 What are Data Centers?
Large facilities with 10s of thousands of networked servers
– Compute, storage, and networking working in concert
– “Warehouse-Scale Computers”

22
Types of Data Centers
Specialized data centers built
for one big app
– Social networking: Facebook
– Web Search: Google, Bing

“Cloud” data centers

– Amazon EC2, Windows Azure
– Google App Engine

23
 Cloud Computing
On-demand
– Use resources when you need it; pay-as-you-go
Elastic
– Scale up & down based on demand
Multi-tenancy
– Multiple independent users share infrastructure
– Security and resource isolation
– SLAs on performance & reliability (sometimes)
Dynamic Management
– Resiliency: isolate failure of servers and storage
– Workload movement: move work to other locations

24
 Data Centers with 100,000+ Servers
Microsoft

Microsoft

25
Google Facebook
26
These things are really big
10-100K servers
100 billion
100s searches
of Petabytes per month
of storage

100s of Terabits/s of Bw
(more than core of Internet)
1.15 billion users
10-100MW of power
(1-2 % of global energy consumption)
120+ million users
100s of millions of dollars
27
 Datacenter Traffic Growth

✧ Source: “Jupiter Rising: A Decade of Clos Topologies and Centralized

Control in Google’s Datacenter Network”, SIGCOMM 2015. 28
 What’s Different about DCNs?
INTERNET

Fabric

Servers
29
 What’s Different about DCNs?
Single administrative domain
No need to be compatible with outside world
Tiny round trip times (microseconds)
Latency/tail latency critical
Massive multipath topologies
Shallow buffers
Backplane for large-scale parallel computation

30
 Example: Web Search
1.
TLA
Partition/Aggregate
Picasso Art is…
Deadline
2. Art is=a250ms
lie…

App Structure

…..
3.

Picasso
• Strict deadlines 1. MLA ……… MLA 1. Art is a lie…
Deadline = 50ms
2. 2. The chief…
• Tail Latency Matters

…..
…..

3. 3.

“Everything
“It is
“I'd
“Art
like
“Bad you
“Computers
your awork
isto liecan
live in
artists
that
as imagine
are
life
useless.
thatman
acopy.
makes
poor real.”
is the
us
“The
Deadline =They
10ms “Inspiration
chief enemy does
of creativity
exist, is
can
ultimate
but itwith only
Good
realize
must lots
good give
seduction.“
artists
the you
ofyou
find answers.”
money.“
truth.
steal.”
sense.“ working.”

Worker Nodes 31
 Data Center Challenges
Massive bisection bandwidth
– Topologies
– Load balancing
– Optics
Ultra-Low latency (<10 microseconds)
– Rate-control or packet scheduling?
– Centralized or distributed?

Managing resources across network & servers

– Multi-tenant performance isolation
– App-aware network scheduling (e.g. for big data)

Next-generation hardware
– RDMA, Rack-Scale Computing
32
Software Defined Networking

33
Software Defined Network (SDN)

Control Control Control

Program Program Program

Global Network Map

Control Plane

Control
Packet
Control
Packet
Forwardin
Forwardin Control
Packet
Control
Packet
gg Forwardin
Forwardin
gg
Control
Packet
Control
Packet
Control
Packet
Control
Packet Forwardin
Forwardin
Forwardin
Forwardin gg
gg Control
Packet
Control
Packet
Forwardin
Forwardin 34
gg
Software Defined Network
A network in which the control plane is
physically separate from the forwarding plane.

and

A single control plane controls

several forwarding devices.
(That’s it)

35
 Software Control Plane

SDN

Intended consequences...
1. Put network owners and operators in control.
2. Networks that are more reliable and more secure.
3. Networks that cost less: simpler, streamlined hardware.
4. Networks that cost less to operate (fewer features).

36
An example
Routing

37
 function Dijkstra(Graph, source):
for each vertex v in Graph:
dist[v] := infinity ;
previous[v] := undefined;
dist[source] := 0 ;
Q := the set of all nodes in Graph ;
while Q is not empty: // The main loop
u := vertex in Q with smallest distance in dist[] ;
remove u from Q ;
if dist[u] = infinity:
break ;

for each neighbor v of u:

alt := dist[u] + dist_between(u, v) ;
if alt < dist[v]:
dist[v] := alt ;
previous[v] := u ;
Edsger Dikjstra decrease-key v in Q;
return dist[], previous[];
end function
1930-2002

38
Photo: Hamilton Richards
A
B

39
The Opte P
40
 1. Figure out which routers and links are present.
B 2. Run Dijkstra’s algorithm to find shortest paths.

“If a packet is going to B,

then send it to output 3”

Data B 2
1 “If B , send to 3”

3 B

41
 95%
1. Figure out which routers and links are present.
2. Run Dijkstra’s algorithm to find shortest paths.
5%
50,000 lines of code
50,000 lines of code
50,000 lines of code

42
Dijkstra IS-IS BGP MPLS Firewall…

Global Network Map

Network OS

43
 OSPF
OSPF Dijkstra
Dijkstra 5%
Global Network Map

Network 95% Network OS

Map

OS
Specialized Packet
Packet
Forwardin
Forwardin
Hardware gg

Packet
Packet
Packet
Packet Forwardin
Forwardin
Forwardin
Forwardin gg
gg Packet
Packet
Forwardin
Forwardin
gg
44
Specialized Hundreds of protocols
Features 6,500 RFCs
Specialized
Control
Tens of millions of lines of code.
Plane Closed, proprietary, outdated.

Specialized Billions of gates.

Hardware Power hungry and bloated.

45
 AppAppAppAppAppAppAppAppAppAppApp

Specialized
Open Interface
Applications
Specialized Windows
or Linux or
Mac
(OS) OS
Operating
System
Open Interface
Specialized
Hardware Microprocessor

Vertically integrated Horizontal

Closed, proprietary Open interfaces
Slow innovation Rapid innovation
Small industry Huge industry 46
 AppAppAppAppAppAppAppAppAppAppApp

Specialized Open Interface

Features
Control Control Control
or or
Specialized Plane Plane Plane
Control
Plane Open Interface

Specialized Merchant
Hardware Switching Chips

Vertically integrated Horizontal

Closed, proprietary Open interfaces
Slow innovation Rapid innovation
47
A Major Trend in Networking

Entire backbone

runs on SDN
Bought for $1.2 billion
(mostly cash)
48
 An Opportunity to Rethink
How should future networks be
– Designed
– Managed
– Programmed

What are the right abstractions

– Simple
– Powerful
– Reusable

49
For next time…

50
51