Cycle 7 / Lecture 14

Cycle 7 / Lecture 14:
Software-Defined Networking,
End-host Applications
CS 498: Systems and Networking Lab
Matthew Caesar
April 25, 2012
1
Innovation in the network
• Things service providers would like to do
–
–
–
–
Mobility management
Network-wide energy management
New naming/addressing schemes
Network access control
• Problem with our network
–
–
–
–
Paths are fixed (by the network)
IP-only
Addresses dictated by DNS, DHCP, etc
No means to add our own processing
2
What routers look like today
•?
App
App
Routing, management, mobility
management,
access control, VPNs, …
App
Operating
System
Specialized Packet
Forwarding
Hardware
Million of
5400 RFCs Barrier to entry
lines
of source
code
500M gates Bloated Power Hungry
10Gbytes
RAM
3
IPSec
Firewall
Router
Software
Control
OSPF-TE
HELLO
HELLO
RSVP-TE
HELLO
Hardware
Datapath
Many complex functions baked into the infrastructure
OSPF, BGP, multicast, differentiated services,
Traffic Engineering, NAT, firewalls, MPLS, redundant layers, …
An industry with a “mainframe-mentality”
4
Glacial process of innovation made
worse by captive standards process
Deployment
Idea
Standardize
Wait 10 years
• Driven by vendors
• Consumers largely locked out
• Glacial innovation
5
Experimenter’s Dream
(Vendor’s Nightmare)
Standard
sw Network
hw Processing
Userdefined
Processing
Experimenter writes
experimental code
on switch/router
No obvious way to do this
Commercial vendor won’t open software and
hardware development environment
Complexity of support
Market protection and barrier to entry
Hard to build my own
Prototypes are flakey
Software only: Too slow
Hardware/software: Fanout too small
(need >100 ports for wiring closet)
Furthermore, we want…
• Isolation: Regular production traffic untouched
• Virtualized and programmable: Different flows
processed in different ways
• Equipment we can trust in our wiring closet
• Open development environment for all
researchers (e.g. Linux, Verilog, etc).
• Flexible definitions of a flow
Individual application traffic
Aggregated flows
Alternatives to IP running side-by-side
…
The solution: OpenFlow
• An open standard to run experimental
protocols in campus networks
– Added as a feature to commercial switches,
routers, and access points
• Widescale recent industry adoption
– Supported by Cisco, IBM, HP, NEC, etc
– Deployed at Microsoft, Facebook, Google,
Verizon, Yahoo!, etc
• 100% of Google’s backbone is OpenFlowenabled
– www.openflow.org
9
OpenFlow Switching
Controller
OpenFlow Switch specification
OpenFlow Switch
sw Secure
Channel
hw
Flow
Table
PC
Flow Table Entry
“Type 0” OpenFlow Switch
Rule
Action
Stats
Packet + byte counters
1.
2.
3.
4.
Forward packet to port(s)
Encapsulate and forward to controller
Drop packet
Send to normal processing pipeline
Switch MAC MAC Eth VLAN IP
Port
src
dst
type
ID
Src
+ mask
IP
Dst
IP
Prot
TCP TCP
sport dport
OpenFlow “Type 1”
• Spec available on www.openflow.org
• Additional actions
Rewrite headers
Map to queue/class
Encrypt
• More flexible header
Allow arbitrary matching of first few bytes
• Support multiple controllers
Load-balancing and reliability
Secure Channel
•
•
•
•
SSL Connection, site-specific key
Controller discovery protocol
Encapsulate packets for controller
Send link/port state to controller
OpenFlow Usage Models
1. Experiments at the flow level
User-defined routing protocols
Admission control
Network access control
Network management
Energy management
VOIP mobility and handoff
…
• Experiment-specific controllers
• Static or dynamic flow-entries
2. Experiments at the packet level
Slow: Controller handles packet processing
Fast: Redirect flows through programmable hardware
Modified routers, firewalls, NAT, congestion control…
3. Alternatives to IP
Lab 7: Creating an Ethernet
Switch
• In Lab 7, you’ll create a learning switch
using OpenFlow
– Unlike traditional Ethernet, all functions
done at the controller
– Broadcast, spanning tree, learning
• Next: a refresher of how learning
switches work
15
Learning Bridges
• Suppose Source sends a frame to Dest
– Which LANs should a frame be forwarded on?
• Trivial algorithm
– Forward all frames on all (other) LAN’s
– Potentially heavy traffic and processing overhead
• Optimize by using address information
–
–
–
–
“Learn” which hosts live on which LAN
Maintain forwarding table
Only forward when necessary
Reduces bridge workload
dest
A
B
source
Learning Bridges
• Bridge learns table entries based on source address
– When receive frame from A on port 1
add A to list of hosts on port 1
– Time out entries to allow movement of hosts
• Table is an “optimization”, meaning it helps
performance but is not mandatory
• Always forward broadcast frames
Port 1
A
B
C
Bridge
Port 2
X
Y
Z
Host
Port
A
1
B
1
C
1
X
2
Y
2
Z
2
Learning Bridges
• Examples
– Frame for A received on port 1: do nothing
– Frame for C received on port 2: forward to port 1
– Frame for S received on port 2: forward to port 1
Port 1
A
B
C
Bridge
Port 2
X
Y
Z
Networked Games
19
Today: Networked Games
• What are networked games?
• How are they designed and hosted?
• Security/Cheating in networked games
20
Overview of Networked Games
21
Online Games
• Computer games are a fast-growing
business
– Revenues already twice that of movies
– MMORPG revenues alone are >$3 billion
• Spurred by popularity of Flash and Java
• Increasingly associated with online
communities and social networks
• Varying business models
– Pay up front, advertising based, pay for
unlocking content, virtual currency
22
Genres of Online Games
• FPS (First-person shooters)
– You control a gun/crosshair
– You shoot and kill other players doing the same
• MMORPG (Massively multi-player on-line roleplaying games)
– You control an avatar
– You kill other avatars to gain loot and power
• RTS (Real-time strategy)
– You control an army
– You go head-to-head against another player’s
army
23
Popular FPS Games
• Half-Life/Counter-Strike (1/2), Battlefield (2 &
2142)
• Wolfenstein: Enemy Territory, Call of Duty (1/2)
24
Popular MMORPG Games
• World of Warcraft, Lineage (1 & 2)
• Runescape, Final Fantasy XI, EverQuest
(1 & 2)
25
Popular RTS Games
• Warcraft 3/Starcraft, Age of Empires
• Warhammer 40000, Command &
Conquer 3
26
Challenges in Network Game
Design
• Need for consistent views across players
– Player A thinks I got shot, but I didn’t see the bullet yet
• Large Scale: Extremely large numbers of clients
– Apple’s gaming platform has >67 million active users
• Clients can be malicious
– People sometimes cheat to win the game or make money
• Bandwidth-constrained clients
– Insufficient capacity, especially in upload direction
– Heterogeneous capacities across clients
27
Gaming architectures
• Hosted
– Racks of servers in data centers
– Example: OnLive (games on the cloud)
– Advantages: easier to manage and secure game, easier to
guarantee resources, privacy of client data, compatible with
advertising and content-unlocking business models
• Unhosted
– P2P sessions formed autonomously between clients
– Advantages: reduced hosting costs, avoids bandwidth
bottlenecks, no need to trust/coordinate with game provider
• Hybrid
– Host only subsets of the game
– Best (or worst?) of both worlds
28
Idea #1: Constrain updates to
Area of Interest (AOI)
• Observation: players only
interact with other players
within “Area of Interest” (AOI)
AOIs
– Players within local area or field
of view
• Idea:
– Restrict communication updates
to other players within AOI
– Reduces bandwidth/computation
overheads
– Problem: might not scale to large
battles
29
Idea #2: Reduce update rate
with Dopplegangers
• Observation: some players are
more interesting than other
players
Doppelgangers
Interest Sets
– More important to get updates
about guy about to shoot me than
some guy I’m not even looking at
– Limitation of human cognition
• Humans focus on small set of
players
• Idea: do lower-bandwidth
updates for non-interesting
players
– High-fidelity updates only for
interesting players
30
How to measure player interest?
• Lots of ways to do this
• One example approach (Donnybrook):
• Aij = Player i’s Attention value to Player j
aij
– Weighted function of
• Proximity: dist(i,j)
• Aim : aij
• Recency of interaction: tij
• Render top-5 most
interesting players with
high fidelity
Sniping
dist(i,j)
31
Idea #3: Avoid imbalanced load
by multicasting updates
“Very”
interesting
player
• Interest sets significantly reduce
bandwidth requirements
– But have heterogeneity in
interest
• Problem: player “very
interesting”
Forwarding pool
– Upload bandwidth limitations
• Solution:
– Forward through a subset of
players
• E.g., randomly chosen from
forwarding pool
– Can advertise membership
information through control
messages
Interested
players
Capture The Flag
32
Idea #4: Client-side lag
compensation
• Observation: we can’t let clients decide their
own position
– They could cheat
– Hence, server should be responsible for computing
position
• Problem: delay between server updates
– Makes movement jumpy
• Idea: Client compensates via interpolation
– Client computes a smooth change in position
between updates
33
Idea #5: Dead reckoning
• Problem: updates may get delayed or lost
in the network
• Observation: we can predict the player’s
position based on current position and
velocity
• Idea: Dead reckoning
– Calculating player’s position based on
previously observed position, estimated
speed and course, and time since last
update
– Client project into future , or server rewinds
into past, using linear physics
– Advantage: can hide delay/loss
– Disadvantage: errors/uncertainties are
cumulative, grow with time
34
Cheating in Networked Games
35
Cheating
• Achilles heel of the PC gaming platform
– Must be fixed to compete with consoles
– Causes legitimate, paying players to quit
– Creates bad word-of-mouth to discourage
new players
– Wrecks virtual economies in MMORPGs
• Why study cheating?
– Important problem in gaming
– Gives insights on how to secure large
distributed systems
36
Types of cheats
• Information exposure
– Wallhacks (OGC), Maphacks (Warcraft 3), Chest
hacks (showEQ)
• Automation
– Aimbot (OGC), Troop command macros (Warcraft
3), Auto-looting (WoW QuickLoot), AFK bots
• Protocol
– Reset cheat (Half-Life), Unit fabrication (Warcraft
3), Item duping (MMO), Speed hack (Half-Life),
Hit point hack (Diablo), Disconnect cheat
• Game bugs
– Game-specific coding errors that lead to
unintended behavior
Information exposure cheats
• Server or peer sends complete information to other client
– Cheat reveals information that should be hidden
• Wallhack
– Quake 4 – released 10/18/2005
– Call of Duty 2 – released 10/25/2005 (Server boycott due to
cheats)
Information exposure cheats
• Maphack (reveal map and enemy units)
– Warcraft3 without Maphack
Information exposure cheats
• Maphack (reveal map and enemy units)
– Warcraft3 with Maphack
Information exposure cheats
• Chest hacks
– Information about what loot is available in
map exposed
– Player goes straight to the locations with
the best loot
Automation cheats
• Automate game activities via Bots
• Aimbots
– OGC
– Automate aiming in FPS
• Macros and game bot farming
– MacroQuest for EQ2
– Automate wealth acquisition via
programs
Protocol cheats
• Hit point cheating
– Diablo protocol messages indicating
damage done to enemy
– Inject messages with inflated damage to
instantly kill opponent
• Item duping
– Disconnect while dropping item
– Ambiguity in whether event happened
globally
• Speed hack
– Inject movement messages to make your
character move or fire “faster” than normal
Types of cheats
• Information exposure
– Wallhacks: lets player see through opaque objects
– Maphacks: lets player see more of a level than
intended by developer
– Chesthacks: information about what loot is
available on map is exposed
• Automation
– Aimbot: helps target acquisition and automatically
shoots opponent
– Troop command macros: replicates commands
across units
– Auto-looting: helps grab items in area
44
– AFK bot: keeps session active or repeats task
while “away from keyboard”
Types of cheats
• Protocol
–
–
–
–
Reset cheat: return character to earlier state
Unit fabrication: create new units/weapons
Item duping: create a copy of an existing item
Speed hack: adjust speed of game (eg slow down,
run faster)
– Hit point hack: prevent HP from decrementing
– Disconnect cheat: guy about to lose can
disconnect
• Game bugs
– Game-specific coding errors that lead to
unintended behavior
45
Software methodology of cheats
• How they do it
– Proxy
• Use separate machine to modify network
packets (aimproxies)
– Program external to game
• Separate process running at higher privilege
level
• In-kernel modules
• Graphics/IO drivers (see-through drivers)
• Additional layer between game and
Windows/DirectX
– Library that hijacks game calls
• Runs in address space of game
Software methodology of cheats
• How they hide from anti-cheats
– Ability to disassemble signatures being
checked
• Polymorphism to thwart file and memory
signatures
– Run in privileged mode or in-kernel to
prevent anti-cheat from accessing it
– Automatic disable when anti-cheat code is
about to run
– Automatic disable when new anti-cheat
distributed
Anti-cheats
•
•
•
•
•
HLGuard (United admins)
Cheating Death (United admins)
PunkBuster
Warden
Intel AMT
Anti-cheats
• Scanners
– Continuously scan memory and filesystem for
foreign libraries and cheats
• Randomize to keep cheats honest
• Steam and VAC, PunkBuster
• Heuristics not perfect: Steam and modified OpenGL
drivers
• Remote screenshot
– Provide a facility for dumping
a player’s screen remotely
• PunkBuster
Anti-cheats
• Authentic peripherals
– Trusted keyboard/mouse clicks
• Hardware signing its movement and clicks
– Trusted network output
• Cryptographic timestamping/ordering
• Prevent look-ahead cheats
• Continuous player performance monitoring
– HLGuard
• Machine learning of reasonable human reaction time
• Ban those who react too fast
• Prone to false positives
– Cal-I (Cyberathlete league) players
Cheating links
• General
– http://rpgexploits.com
– http://msxsecurity.com
– http://zerogamers.com
• WoW
– WoW Glider
• http://wowglider.com
– WoW radar, WoW Sharp, ByteBot, GALB
– WardenNet, ISXWarden (anti-anti-cheats)
• http://ismods.com/warden
• http://edgeofnowhere.cc/viewtopic.php?t=311208
• http://www.rootkit.com/newsread.php?newsid==360
– ISXWoW
• http://ismods.com/downloads.php
Cheating links
• Half-Life
– OGC
• http://mpcdownloads.com
• http://www.mpcforum.com/showthread.php?t=
31409
• EverQuest 2
– MacroQuest
• http://sourceforge.net/projects/macroquest
Anti-cheat links
• WoW Warden
– http://www.ismods.com/warden
• PunkBuster
– http://punkbuster.com
• Valve Anti-Cheat (VAC)
– http://server.counterstrike.net/server.php?cmd=VAC
• HLGuard, Cheating-Death
– http://unitedadmins.com
• Intel’s AMT
– http://www.intel.com/go/iamt/
Course Recap
54
Cycle
Topic
Description
1.
Network
administration
Network defense, troubleshooting, and
design
2.
ISP networks
Routing protocols, ISP systems, how to
run your own ISP
3.
Equipment and
Devices
Router, switch, server, and network
architectures
4.
Server Software
Architectures
Socket and network programming, server
software architectures
5.
Distributed Services
Content distribution networks, wide-area
deployment, coordination protocols
6.
OS Network Stacks
Linux kernel internals, network stacks,
service-customized OS
7.
Networked
applications
Software-defined networking
Networked games
Overview of Cycles
55
Network Administration and
ISP Operations
• Jobs: ISP Operator, Network Admin
• Responsibilities:
– Select and purchase network
equipment
– Design, configure, and secure
networks
– Debug and troubleshoot problems
• Companies:
– ISPs, large enterprises, cloud and data
center operators, hosting facilities
– AT&T, Verizon, Sprint, Google,
Microsoft, UIUC …
• Resume keywords:
– BGP, OSPF, router configuration,
network administration, Cisco IOS
Equipment and Devices
• Jobs: Device architect, systems
architect, network architect
• Responsibilities:
– Design network equipment (routers,
switches, etc)
• Internal software and hardware architectures
• Testing and management equipment
– Network layout, provisioning, and planning
• Companies: Cisco, HP, NEC, Huawei,
IBM
• Resume keywords: performance
analysis, network analysis, switch
design, router internals
57
Networked Operating Systems
• Jobs: OS designer (traditional PC OSes and
equipment/device OSes),
• Responsibilities
– Implementation, design, and performance
tuning of network stacks
– OS kernel and driver levels, I/O and network
stacks
• Companies: OS companies (Microsoft, Google,
Apple), companies that manage their own inhouse OS (Akamai, Yahoo!), companies that
have device-customized OSes (Cisco, HP,
Avici)
• Resume keywords: Linux kernel,
kernel programming, Linux
TCP/IP network stack
58
Server Software and Distributed
Service Architectures
• Jobs: Service architect (Web, Cloud, Database,
etc), Distributed system architect
• Responsibilities: Building, designing, and tuning:
– Infrastructure: middleware, DHTs, locking systems, storage,
checkpointing, logging, monitoring, indexing
– Applications: social networks, search engines
• Companies: Akamai, Facebook, Twitter, Amazon, Microsoft,
Google, VMWare,
• Resume keywords: DHTs, Planetlab, Emulab,
sockets programming, content distribution
59
Software-defined Networking
• Jobs: SDN application designer, Network operating
system architect
• Responsibilities: Deploying and tuning SDN,
migrating existing setups to SDN, implementing SDN
applications, implementing new SDN infrastructure
• Companies: SDN startups (Nicira), companies using
SDN (Google, Verizon, Microsoft, Yahoo, Facebook)
• Resume keywords:
– OpenFlow, NOX, Software-defined Networking
60
Wrapping up
• Course feedback?
• Questions?
61