Content Distribution Networks

Design Space • Caching

Content Distribution Networks

– explicit – transparent (hijacking connections)

Outline

• Replication

Implementation Techniques Hashing Schemes Redirection Strategies

Spring 2002

CS 461

– server farms – geographically dispersed (CDN)

1

Spring 2002

CS 461

2

Denial of Service Attacks (DoS)

Story for CDNs • Traditional: Performance (Response Time) – move content closer to the clients – avoid server bottlenecks

client

client

• New: Flash Crowds & DDoS (System Throughput) – distribute load over massive resources – multiplicatively raise level of resources needed to attack server

attacker Spring 2002

CS 461

3

Spring 2002

CS 461

client 4

Distributed DoS (DDoS)

Redirection Overlay Geographically distributed server clusters

zombie attacker

R

R

client client

R R

Internet Backbone

server R slave attacker zombie

R

R client zombie

Spring 2002

R clients

zombie CS 461

5

Distributed request-redirectors

Spring 2002

CS 461

6

Techniques

CDN Components • DNS aaa.com

bbb.com

ccc.com

– one name maps onto many addresses – works for both servers and reverse proxies

Backend servers Cache

• HTTP – requires an extra round trip

Geographically distributed surrogate servers

• Router – one address, select a server (reverse proxy) – content-based routing (near client)

Redirectors

• URL Rewriting

Cl ients

Spring 2002

CS 461

– embedded links 7

Spring 2002

CS 461

8

Redirection: Which Replica?

Hashing Schemes: Modulo

• Balance Load • Cache Locality • Network Delay

svr0

URL (key)

%

• Easy to compute • Evenly distributed • Good for fixed number of servers • Many mapping changes after a single server change

svrN

Spring 2002

CS 461

9

Spring 2002

CS 461

Highest Random Weight (HRW)

Consistent Hashing (CHash)

high

url-0

• Hash server, then URL • Closest match svr1 • Only local mapping changes after adding or removing servers • Used by State-of-the-art CDNs

URL

weight0

svr0

svrN

url-1 svr2

Unit circle Spring 2002

10

CS 461

11

RH svr0 svr1 svr2 svr svr svrN Spring 2002

sort

• •

weight1 weight2

• •

Hash(url, svrAddr) Deterministic order of access set of servers Different order for different URLs Load evenly distributed after server changes

weight0 weightN

low CS 461

12

Redirection Strategies

Redirection Strategies (cont)

• Random (Rand)

• Coarse Dynamic Replication (CDR)

– Requests randomly sent to cooperating servers – Baseline case, no pathological behavior

– – – –

• Replicated Consistent Hashing (R-CHash) – Each URL hashed to a fixed # of server replicas – For each request, randomly select one replica

• Fine Dynamic Replication (FDR)

• Replicated Highest Random Weight (R-HRW)

– Bookkeeping min # of replicas of URL (popularity) – Let more popular URL use more replicas – Keep less popular URL from extra replication

– Similar to R-CHash, but use HRW hashing – Less likely two URLs have same set of replicas

Spring 2002

CS 461

Using HRW hashing to generate ordered server list Walk through server list to find a lightly loaded one # of replicas for each URL dynamically adjusted Coarse grained server load information

13

Spring 2002

CS 461

C

Network Topology

Simulation

C

C

WA

R

R

R

R

MA

IL

C

C

R

R

PA CA

Models network, application, and OS Built on NS + LARD simulators 100s of servers, 1000s of clients >60,000 req/s using full-TCP transport Measure capacity, latency, and scalability

R

R

S

R

MI C

• End-to-end network simulator prototype – – – – –

C

C

S

– Server overload, network congestion…

S

S

R

• Identifying bottlenecks

14

NE

C

R

CO R

S

DC R

S

R

SD CA

S

S

GA TX

R C

S

R C

R

C C

S – Server, C – Client, R - Router Spring 2002

CS 461

15

Spring 2002

CS 461

16

Capacity: 64 server case

Simulation Setup

Normal Operation 35000

• Workload Throughput req/s

– Static documents from Web Server trace, available at each cooperative server – Attackers from random places, repeat requesting a subset of random files

• Simulation process

30000 25000 20000 15000 10000 5000 0

– Gradually increase offered request load – End when servers very heavily overloaded

Rand CDR

R-CHash FDR

R-HRW FDR-Ideal

A single server can handle ~600 req/s in simulation Spring 2002

CS 461

17

Spring 2002

Under Attack (250 zombies, 10 files, avg 6KB)

Random’s Max: 11.2k req/s R-HRW CDR

40000

R-CHash Rand

R-CHash Max: 19.8k req/s R-HRW FDR

FDR

20000 10000 0 R-CHash FDR

R-HRW FDR-Ideal

CS 461

80 60 40 20

80 60 40 20

0

0 0.1

A single server can handle ~600 req/s in simulation 19

R-CHash CDR

100 CDF of Response

100

30000

CDF of Response

Throughput req/s

50000

Rand CDR

18

Latency: 64 Servers Under Attack

Capacity: 64 server case

Spring 2002

CS 461

1 10 100 Response Time in Logscale (Seconds)

Spring 2002

CS 461

0.1

1 10 100 Response Time in Logscale (Seconds) 20

Latency At CDR’s Max: 35.1k req/s

Normal Operation

CDR Throughput req/s

CDF of Response

100 80 60 40 20 0

45000 40000 35000 30000 25000 20000 15000 10000 5000 0 0

0.1

1 10 100 Response Time in Logscale (Seconds)

Spring 2002

CS 461

16

32 CDR

21

Spring 2002

Various Attacks (32 servers) 1 victim file, 1 KB

Throughput req/s

Throughput req/s

25000 20000 15000 10000 5000 0 400

500

600

700

800

Num of Zombies (slave attackers)

Rand Spring 2002

R-HRW

96 112 128

0

16

32

48

64

80

96 112 128

Num of Servers

Num of Servers R-HRW Rand

CDR CS 461

R-HRW

Rand 22

• Servers join DDoS protection overlay – Same story as Akamai – Get protection and performance

25000 20000 15000

• Clients use DDoS protection service

10000

– Same story as proxy caching – Incrementally deployable – Get faster response and help others

5000 0

300

64 80

Deployment Issues

30000

30000

200

48

50000 45000 40000 35000 30000 25000 20000 15000 10000 5000 0

10 victim files, avg 6KB

35000

100

Under Attack (250 zombies, 10 files)

Throughput req/s

FDR

Capacity Scalability

100

200

300

400

500

600

700

800

Num of Zombies (slave attackers)

CDR

Rand CS 461

R-HRW

CDR 23

Spring 2002

CS 461

24