Content-oriented networking

J ne 10, 2011 u KT Taekyoung Kwon Seoul National University

outline
IP networking vs. Content networking CANA Others

Internet
Original motivation for Internet is to share computing resources
Remote login, file transfer

Hence the Internet communication model is host-to-host conversations

It has been working 40 years

TCP/IP has been so flexible for diverse applications

TCP/IP cannot be changed

Internet is ossifying

Why content-oriented networking (CON)?

Internet traffic is already content-oriented
CDN, Edge Caching Network, multimedia, P2P

Users/applications care what to receive

They dont care from where So, host-based comm. model is outdated

Storage cost is getting cheaper sharply

Compared to networking cost

Other advantages of CON coming up later

IP networking
Host-centric design Lookup-by-name
Indirection (from name to locator)
DNS Hosts only

Host/link availability concern

DNS is extendible and highly available

Distributed design Thanks to caching

Locators can be aggregated

Network prefix
Currently 350k+

Routing scalability is better than CON

Content-oriented networking (CON)

Route-by-name
No indirection, better availability Content name (or ID) is a routing entry Huge scalability concern

Global-scale and systematic CON is not feasible

At least billions of contents CCN at PARC (Van Jacobson)
Aggregation by using URLs

Other merits such as authentication

IP networking vs. CON

Network prefix
Destination
192.168.0.0/16

C o n te n t n a m e
Content Name
/a.com/b.jpg

Next Hop
Router C

Next Hop
Router C

/a.com/b.jpg

co /a. B:

m/ b

. jp

CANA: CONTENT-AWARE NETWORKING ARCHITECTURE

Joint work with Yanghee Choi

What is an IP address?

IP address
An IP address originally indicates the endpoint
End-to-end principle Serves as both locator and identifier

Current role of IP address

Not endpoint
NAT, tunneling, overlay,

Not identifier
Mobility, multi-homing,

 Just locator

Then what should be an IP address?

Not identifier

Locater of next transit point

NAT, tunneling, Some agents
E.g. mobility agent in mobile IP solutions

Transit-by-transit
Not end-to-end

Wait, where is the endpoint identifier?

How about using some other identifier?

General identifier requirements

Unique Routable/locatable Persistent
Location-independent

We choose uniqueness and routability

How about Content identifier (CID)?

CID will fill the lacking role of the IP address
Endpoint identifier

Globally routable and Unique

Domain name (or public IP address) + port number (or its hint)
Static content, e.g. http://www.nytimes.com/logo.jpg Dynamic content, e.g. 20.30.40.50:4000

Content-aware Network Architecture (CANA)

Application Global Transit Link Physical Content Transit header Global header

IP TCP Content More info. header header identifier on content

Network layer is renamed as transit layer Transport layer is extended to global layer
CID is added

CID is locatable and unique Additional content info (e.g. bit rate) helps other layers
Deep packet inspection is assumed for other layers

CANA: Host side

New model for IP subnet:
solicitor vs. agent
An access router becomes an agent Solicitor and its agent communicate in a content-oriented fashion

An agent contacts DNS

Solicitors cannot

solicitor cannot contact server directly

1 : I w n t a p a rti l r co n te n t a cu a ( e . g . HTTP URI ) internet 2 : H e re yo u a re agent

so l ci r i to

Other aspects of agents

Flash crowd can be dealt with by caching contents at agents

NAT does not matter

Agent failure may not matter

In wireless, there may be other agents Content identifier (CID) always forwards the content request to the correct endpoint

CANA: Publisher side

Registers its domain name with the DNS
Agents IP address (of the egress link)

Publisher and agent will communicate in a content-oriented fashion

1 : a re q u e st fo r yo u r co n te n t internet p u b l sh e r i 2 : h e re yo u a re agent

* Assume

that publication is already done

Intra-domain
Publishers agent will be contacted by the solicitors agent Publishers agent will receive the content from the publisher
Will relay the content to the host via the hosts agent

Agents can cache contents

host agent

Intra-domain

agent

publisher

Inter-domain
Gateway A requests the content to gateway B Gateway B will get the content from agent of publisher Then relay the content to gateway A Gateway A will relay the content to the agent of the host Gateways can cache contents ISP A
Gateway A Gateway B

ISP B

host

publisher

Content-aware routers (CARs)

Legacy routers look at IP address in transit header CARs also look at CID in global header CARs can participate in content relaying
CARs can cache contents

CANA operations: Content Request Message 1 se n d s a co n te n t re q u e st 1. H

m e ssa g e to A 1 , i src : d st i ts s H 1 :A 1 ( the content C H2 belongs to H 2 ) 2 . A 1 m a ke s a C I e n try ( C H2 , B H1) to deliver content data 3. The content request message now has A1:A2 as src:dst IP addresses 4. C1 makes a CIB entry (CH2 , A1) to deliver content data

..

As content request message traverses, a content info base (CIB) entry is set up backwards to relay content data

Internet security is a big threat

The key problem is that
Anyone can send any packet to anyone
E.g. DDoS

CANA: new service paradigm for ISPs

Not deliver packets But deliver contents

Policing at agent
Agent can perform proactive measures
To help prevent DDOS attack E.g. solicitors cannot generate too many requests
Especially to the same server

Agent will analyze the behaviors of its solicitors

Contents requests by solicitors are accountable Agents may collaborate

CANA: advantages
End hosts behind NATs are now first class citizens
Private addresses are fine Mitigates issues of IP address exhaustion and routing scalability

Ready to interwork with long disruption or delay networks

DTN, CNF, sensor, MANET

Security, accountability are enhanced Mobility and multicast are better handled

Mobility in CANA
The serving agent will take care of L2 handoff If a host moves to a new agent,
Host detects content delivery failure
It sends a new content request message

The new agent will continue the flow by relaying content request message toward the publisher

Multicast in CANA
R e ve rse p a th fo rw a rd i g n E a si r a d d re ss m a n a g e m e n t e A: source E,G,D multicast recipients First E joins
C has (CA,E), A has (CA,C)

G joins
C has (CA,E and G)

NetFPGA
CPU
Mem ory

PCI

PW-OSPF Java GUI Front Panel (Extensibl e)

Verilog EDA Tools (Xilinx, Mentor, etc.)

NetFPGA Driver
1GE 1.Design 1.Design 2.Simulate 2.Simulate L3 L2 In Q 3.Synthesize 1GE 3.Synthesize Par Par Mgm 4.Download 4.Download 1GE se se t Out My IP 1GE Q Bloc Look Mg k up 1GE mt Verilog modules interconnected by FIFO interfaces

FP G A
Memo ry

1GE 1GE 1GE

CANA implementation: caching

User-level software Cache Router
URL Parser Cache Table
Register IP to forward

HTTP-GET flow
Cache miss

Content flow
User-level software Cache Router
URL Parser Cache Table

TCP Layer IP Layer

TCP Layer IP Layer

Copy cached Content Destination IP matching

Web Traffic (port 80) Capture

Reference Router Logic

Web Traffic (port 80) Capture

Reference Router Logic

NetFPGA

NetFPGA

CANA: caching
User-level software Cache Router
URL Parser Cache Table
Registrate IP to forward Cache hit

Cache hit

TCP Layer IP Layer

Web Traffic (port 80) Capture

Reference Router Logic

NetFPGA

Experimental Scenarios
Content Router Legacy IP Router 1G Link
A

100M Link Cached Content

A A
A

Cache router performance

Throughput, latency

Increase the number of cache routers Measures

Experimental Scenarios
A
Content Router Legacy IP Router 1G Link 100M Link Cached Content

A
A A A

Content Delivery Efficiency

Throughput, delay Traffic Volume

Increase the number of clients Measures and compared with legacy IP network

Experimental Scenarios
Content Router Legacy IP Router 1G Link 100M Link

A1 A1
A1

A2

Cached Content

Mobility

Move client toward the new subnet Measures

Event though IP change is occurred, content downloading is still being operated

A2

Disruption time during handoff

CANA: Conclusions
IP address loses its identifier role Content identifier is introduced instead Combination of IP address and CID can solve or mitigate many problems in Internet Incremental deployment Performance issues for contentawareness

SCAN: SCALABLE CONTENT ROUTING FOR CONTENT-AWARE NETWORKING

Joint work with Kideok Cho, Munyoung Lee, Kunwoo Park, Yanghee Choi

IP networking
Lookup-by-name
DNS: Indirection from name to locator
DNS is extendible and highly available Distributed design, caching

Host/link availability concern

Delivery inefficiency concern

Locators can be aggregated

Network prefix
Currently 350k+

Routing scalability is better than CN

Content networking (CN)

Route-by-name
No indirection, better availability Content name (or ID) is a routing entry Huge scalability concern

Global-scale and systematic CN may not be feasible

NDN, TRIAD
Some aggregation by using URL-like names

DONA, PSIRP
Flat names for persistency

Better delivery performance

Exploiting multiple sources, multiple paths/interfaces

SCAN: A hybrid approach

Inefficiency in TCP/IP networking
Cannot know closer copies of the content Dont do parallel transmissions

Content networking
Scalability, reachability issues

Our Solution: A hybrid approach

IP routing: default routing for reachability Content routing: opportunistic routing for efficiency (closer & multiple copies)

SCAN Overview
Content routers (C-routers) do IP and content routing
Each content has a content identifier (CID) Local content table (LCT): cached content files Content routing table (CRT): CIDs of the content files in neighbor C-routers CIDs are advertised by Bloom Filter (BF)
C-router Bloom filter
0 1 0 1

ci d
0 1 0 1 0

C-router

Bloom Filter (BF) issues

More bits of the BF may be set to 1
As the number of content files increases

C-router will decay the bits of a BF probabilistically before exchanging content routing table (CRT) info
E.g. if decaying prob. is 0.5, around the half of the bits 1 will be randomly set to 0

If matched bits > threshold, scan the requested content

Threshold is decided by considering C-router distance

SCAN Operations (1/2)

Content Routing
Content A

(3) (3) (2) (2) (2) (2) (2)

Server
Content A

Legacy router C-router Content

(1 ) (2) (3 ) (4 ) (5 )

(2)

Content Request Scanning Request Content Response Range Request Content Delivery

(2) (2)
Content Request

(1 )

(3)
Content A

SCAN Operations (2/2)

Content Delivery (an example)
Server (5)
Content A

(5)

Content A

C-router Content
(1 ) (2) (3 )

Content Request Scanning Request Content Response Range Request Content Delivery

(4 )

(4 ) (4 ) (5)
Content A

(4 ) (5 )

41

Receives content request

Makes a copy & Does IP routing LCT lookup Can I afford scanning now? Yes

Forwarding Decision
Scanning
CID matched in LCT CRT lookup No Yes Forward the scan request to the matched interfaces Yes Respond to the end host

# of matched bits threshold No No

Done

42

Simulation Setup
GT-ITM: 1 transit and 5 stub domains
1*5+5*20 C-routers

100 servers among 1000 end hosts, Total 20,000 content files
10,000 different content files at 100 servers top 10% have avg. 10 copies at C-routers

Content file size: 1GB

Each C-router has 100 files

Each BF is 3,500 bit long with 14 hash functions

SCAN vs. IP routing, IP with caching, SCAN w/o BF (C-info of BF size), SCAN-full
43

Simulation Results (1/2)

44

Simulation Results (2/2)

45

SCAN: Conclusions
IP networking and content networking (CN) have pros and cons A hybrid approach is proposed
IP networking is fallback Opportunistic CN is exploited

Number of Contents will be huge

Bloom filter based advertisement Decaying is introduced

Spatial locality needs to be explored

others

WAVE (1/2)
Content Distribution
storage cost has been and will be rapidly decreasing routers can be co-located with storage modules How to cache?

WAVE (2/2)
Content Diffusion
Popular content is replicated Chunk-based, incremental caching of content

iCODE: ISP-centric Content Delivery

A content delivery architecture
An ISP can provide content delivery services to CPs

Merits
Reduced delivery latency ISP incentives
Inter-ISP traffic reduction Traffic engineering Customized content delivery services
Location Database C - router Contents

End Host

i Proxy

Original Server

i Tracker i DNS Server

R2
C1 C2

R1
C3

C1

C2

C3

R4
C2

R3
C1 C3

iCODE

Discussions

tkkw o n @ sn u . a c . kr

CANA: Some issues

Inter-domain performance Scalability (of solicitors) Scalability (of content files)
NetFPGA router

Agent compromise

Connectivity between host and agent

New communication model for subnet:
Host vs. agent Agent
Agent advertisement (AA) messages

Host
Receive AA message Private address (IPv4) and link local address (IPv6) No need to know destination address
Cannot contact DNS

Assume security association (SA) is set up