comparative study of content-centric vs. content delivery

15
Comparative Study of Content-Centric vs. Content Delivery Networks: Quantitative Viewpoints 2015.06.08 Seongik Hong 1

Upload: others

Post on 31-May-2022

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Comparative Study of Content-Centric vs. Content Delivery

Comparative Study of Content-Centric vs. Content Delivery Networks:

Quantitative Viewpoints

2015.06.08

Seongik Hong

1

Page 2: Comparative Study of Content-Centric vs. Content Delivery

Internet Traffic Breakdown

2 Source: ARS technica

• Bandwidth explosion: As Internet use soars, can bottlenecks be averted?

Bandwidth usage is soaring, driven by the proliferation of Internet-connected devices.

Exponential growth

Page 3: Comparative Study of Content-Centric vs. Content Delivery

Advent of Content-Centric Networking

3

Contents Server

Contents Requestor

Contents Server Address IP packet

Contents Name

IP-based Current Internet

CCN-based Future Internet

Network Router

CCN packet

Payload Security

Information

Payload

cache

What routers do (IP)

follow FIB

follow PIT

follow FIB

follow FIB

FIB: Forwarding Information Base, PIT: Pending Interest Table

What routers do (CCN)

PIT

Page 4: Comparative Study of Content-Centric vs. Content Delivery

Storage-based Networking

4

• Can storage costs fall to zero?

Storage cost drops sharply compared with network

Page 5: Comparative Study of Content-Centric vs. Content Delivery

Existing Approach: Content-Delivery Network

5

• Content-Delivery Network (CDN)

a large distributed system of servers deployed in multiple data centers across the Internet

3. IP Control Plane

(OSPF)

1. CDN Control Plane

2. CDN Data Plane

IP

IP

IP

IP

IP

IP

IP

IP

IP

4. IP Data Plane

Request Router (URL)

Edge Cache

Servers

IP Routers

(IP Packet)

Original Contents

Server

Carry Traffic

Page 6: Comparative Study of Content-Centric vs. Content Delivery

CCN vs. CDN

6

• Qualitative Comparison

Many studies

• Quantitative Comparison

Not many, why?

o no real CCN deployment with referenceable router implementation

o specific CDN configuration in commercial deployment is generally not available to the public

Page 7: Comparative Study of Content-Centric vs. Content Delivery

IP Routers

7

ge2/1

ge2/2

ge2/3

ge2/4

ge2/5

ge2/6

ge2/7

ge1/1

ge1/2

ge1/3

ge1/4

ge1/5

ge1/6

ge1/7

Ingress Packet Buffer

(Small: KB)

Switching Fabric

Packet Processor (Ingress Lookup/

Egress Scheduling)

Egress Packet Buffer

(Medium: MB)

Line Card #1 Line Card #2Switching Module

Control Module

MACTable

MAC Address Table

OSPF Process

LSDB (Link-State Data Base) SPF Calculation

10R1

R2

R3

R4

R5

R6

R7

R8

20 1020

1020

1020

30 10

201.1/16R9

10R1 R4

R710

101.1/16R9

RIB

FIBIP

PrefixNextHop

1.1/16...

R4...

Interface

ge2/2….

Ingress Packet Buffer

(Small: KB)

Packet Processor (Ingress Lookup/

Egress Scheduling)

Egress Packet Buffer

(Medium: MB)

MACTable

FIBIP

PrefixNextHop

1.1/16...

R4...

Interface

ge2/2….

OSPFNeighbors

OSPFNeighbors

Advertize Routes Advertize Routes

Destination Network PrefixDestination Network Prefix

Next HopNext Hop

1.1/16...

1.1/16... R4

...

R4...

R1 R4 Destination NetworkDestination Network

Next HopNext Hop

1.1/16...

1.1/16... R4

...

R4...

R1 R4 OIFOIF

ge2/2...

ge2/2...

Copy RIB to FIB in all Line Cards

• Typical IP router architecture

Base for CCN router architecture

Page 8: Comparative Study of Content-Centric vs. Content Delivery

CCN Routers

8

1

Content store

2

Supermicro X9SCM-F

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Hitachi 3TB Deskstar

Crucial m4 512GB

Raid Controller (Onboard)

Crucial m4 512GB

LSI SAS 9201-16i 16Port LSI SAS 9201-16i 16Port

8GB ECC 1333MHz Intel E3-1260L Supermicro AOC-STGN-i2S

10GbE 10GbE

SSDOS, Log, Popular Content

10GbE NIC

Memory Processor

Content Storage Content Storage (Mirror) Content Storage Content Storage (Mirror)

Mainboard

Redundancy

Netflix Cache Server

Additional parts for CCN

1. Forwarding strategy, PIT (ignore cost)

2. CS index & CS

A

• Processor & Memory

downsize since there is

no TCP overhead

• LAN card

Connector & interface

chip for backplane

connection

B

Page 9: Comparative Study of Content-Centric vs. Content Delivery

CDN Topology

9

CDN Edge Server

Request Router

Edge Server (H/W)

Edge Server (S/W)

PE Router

CDN

CCN

• Typical IPTV CDN carrier topology

Base for CCN network topology for comparison

Page 10: Comparative Study of Content-Centric vs. Content Delivery

Network Cost

10

• total number of subscribers : 3million • content encoding rate : 8Mbps (for

High Definition TV) • concurrent usage rate : 10% • number of CDN edge servers : 30 • Each edge server holds 20TB of

contents • The content popularity observes Zipf’s

law with the exponent s

The ratio represents the caching capacity, for example, 4:3:2:1 means each level i (i = 1,2,3 and 4), node can store 40%, 30%, 20% and 10% of the total contents.

Level 4

Level 0

Page 11: Comparative Study of Content-Centric vs. Content Delivery

H/W & S/W Cost

11

Product Product Description Discount Price*

CDN Edge Server

(CDS-TV)

CDS-TV (RTSP Streaming, RTSP

Session Management, Content Placement,

Server Heath-Report, etc.)

$30.4 per

Stream

Content Acquirer (CDS-CA)

CDS-CA (Reverse Proxy, Content Placement,

etc.)

-

H/W cost S/W cost

Level 4

Level 0

Page 12: Comparative Study of Content-Centric vs. Content Delivery

Total Cost

12

• As the ratio of the concurrent user increases, the costs for the CDN is doubled and tripled.

• On the contrary, with CCN, the increase rates are only 45% and 91%.

Page 13: Comparative Study of Content-Centric vs. Content Delivery

Protocol Overhead

13

Kernel

(TCP/IP)

Client

HTTP Request (GET)

TCP SYN

TCP SYN-ACK

ACK

Send HTTP Response

Read Application Data

Write Sock Buffer

TCB Read

TCP/IP Checksum

Calculation

NIC Driver call

ACK

HTTP Response

TCP FIN

TCP FIN ACK

Done

64KB

64KB

Kernel Space Memory

ClientCPU

Interest

CCN RouterCDN Edge Server

User Space

SG List

Network Interface Card (NIC)

64KB

TCB

TCB

Memory

SG List

(Header ptr &

Payload ptr)

Application

( )

ReceiveReceive HTTP Request

Parse the request (L7 Lookup)

→ Cache Hit

Generate HTTP Response

HeaderSend (memory pointer*, 64KB)

Locate the requested file

*

SG List

TCP, IP, Eth Headers

IP Payload

CCN Data

Packets

CS Lookup → Cache Hit

TCP Sequence number

IP Address

MAC addressWrite TCP, IP, ETH Header

Data Packet

Per-User (TCP

Session)User

Agnostic

User Agnostic

Locate the requested chunk

CPU

Application

Buffer

Socket Buffer

DMA

Web: 574

System Call: 499

TCP: 1372

IP: 380

Ethernet: 411

Driver: 530

DMA

Kernel (CCN)

< 574

NIC Driver call

Ethernet: 411

Driver: 530

Eth Headers

Write ETH Header

SG List

(Header ptr &

Payload ptr)

Create SG List

Network Interface Card (NIC)

3,776 instructions per packet <1,515 instructions per packet

CCN header

NIC NIC

Data Packets

Update TCB

Create SG List

User

Agnostic

(Web server)

No data at user agnostic data

packet

NetworkInterface

Card

NetworkInterface

Card

SG List(Header/payloadpointer

SG List(Header/payloadpointer

EthernetHeaders

• CDN: HTTP, TCP handling • CCN: Interest, PIT handling

Page 14: Comparative Study of Content-Centric vs. Content Delivery

Protocol Overhead

14

Page 15: Comparative Study of Content-Centric vs. Content Delivery

Conclusion

15

In this paper, we proposed • a typical nation-wide CDN topology, a corresponding CCN deployment • the first feasible CCN router architecture.

Based on them, we estimated • the total amount of network traffic, • corresponding H/W and S/W costs and protocol overhead for both CCN

and CDN.

We believe our work can be a design guide for how the CCN need to evolve to overcome IP-based services such as CDN.