Overlay and P2P Networks Applications continued and Summary

Prof. Sasu Tarkoma

18.2.2013

Contents

Today CDNs Some advanced topics Summary and conclusions

Content Delivery Networks (CDN)

Geographically distributed network of Web servers around the globe (by an individual provider, E.g. Akamai).

Improve the performance and scalability of content retrieval. Allow several content providers to replicate their content in a

network of servers.

Motivation

Network cost Huge cost involved in setting up clusters of servers around the globe and corresponding increase in network traffic

Economic cost

Higher cost per service rate making them inaccessible to lower and medium level customers

Social cost

Monopolization of revenue

CDN Technology

Intelligent wide area traffic management Direct clients’ requests to optimal site based on topological proximity

Two types of redirection: DNS redirection or URL rewriting Cache

Saves useful contents in cache nodes.

Two cache policies: least frequently used standard and least recently used standard.

CDN Types (Skeletal)

CDNs

Hosting CDN Relaying CDN

Partial Site Content Delivery

Full Site Content Delivery

URL Rewriting

DNS based

Request Routing Techniques

7

CDN

Replicate content on many servers Challenges

How to replicate content Where to replicate content How to find replicated content How to choose among known replicas How to direct clients towards replica

DNS, HTTP redirect, anycast, etc. Akamai

8

Server Selection

Service and content is replicated in many places in network How to direct clients to a particular server?

As part of routing à anycast, cluster load balancing As part of application à HTTP redirect As part of naming à DNS

Which server to use?

Best performance à to improve client performance Based on Geography? RTT? Throughput? Load?

Lowest load à to balance load on servers Any active node à to provide availability

CDN Architecture

Surrogate Surrogate

Request Routing

Infrastructure Distribution

and Accounting

Infrastructure

CDN

Origin Server

Client Client

Surrogate server placement problem

Given N possible locations at edge of the Internet, we are able to place K (K<N) surrogate servers, how to place them to minimize the total cost?

This is the: Minimum K-Median Problem Given N points, we must select K (centers), and then assign

each input point j to the selected center that is closest to it. The goal is to select K centers so as to minimize the sum of the assignment costs.

This is NP-Hard Many heuristics algorithms to solve this (tree, greedy ,…)

DNS

CDN DNS

CDN

Client ISP Clients

1 6

2 3

5

4

redirection Content Provider

Client DNS

(Local DNS

server for client)

CDN Type Coverage Solutions Akamai Commercial

CDN service including streaming data

Market leader Edge platform for handling static and dynamic content, DNS-based request-routing

Limelight Networks

Commercial On-demand distribution, live video, music, games, …

Surrogate servers in over 70 locations in the world

Edge-based solutions for content delivery, streaming support, custom CDN for custom delivery solutions, DNS-based request-routing

Coral Academic Content replication based on popularity (on demand), addresses flash crowds

Experimental, hosted on PlanetLab

Uses a DHT algorithm (Kademlia), support for static content, DNS-based request-routing

CoDeeN Academic testbed Caching of content and redirection of HTTP requests

Experimental, hosted on PlanetLab, collaborative CDN

Support for static content, HTTP direction Consistent hashing for mapping data to servers

Globule Academic Replication of content, server monitoring, redirection to available replicas

Apache extension, Open Source collaborative CDN

Support for static content, monitoring services, DNS-based request-routing

Akamai

Clients fetch html document from primary server URLs for replicated content are replaced in html Client resolves aXYZ.g.akamaitech.net hostname Akamai.net name server returns NS record for

g.akamaitech.net G.akamaitech.net nameserver choses server

in region Should try to choose server that has file in

cache - How to choose? Uses aXYZ name and consistent hash

14

How Akamai Works

End-user

cnn.com (content provider) DNS root server Akamai server

1 2 3

4

Akamai high-level DNS server

Akamai low-level DNS server

Akamai server

10

67

8

9

12

Get index.html

Get /cnn.com/foo.jpg

11

Get foo.jpg

5

Source: www.cs.cmu.edu/~srini/15-744/S08/lectures/17-DNS.ppt

Question

Consistent hashing is extensively used in CDN and DHTs Are there alternatives? Simply taking hashval mod N is not good due to remapping

when N changes Linear hashing predates consistent hashing and it can be

used to solve this

Linear Hashing

Use a family of hash functions h0, h1, h2, ... Each function’s range is twice that of its

predecessor When all the pages at one level (the current

hash function) have been split, a new level is applied

Splitting occurs gradually

Current hash function, then you know if a bucket has been split from a split counter

Pages are split when overflows occur – but not

necessarily the page with the overflow (use overflow bucket as a temporary measure)

Splitting in a round robin fashion

Linear Hashing II

Use a family of hash functions h0, h1, h2, ...

hi(key) = h(key) mod(2iN) N = initial number of buckets

h is some hash function

hi+1 doubles the range of hi

Keep track of the next bucket to split and the current level: half of a split bucket is moved to the new bucket

When all buckets split, go to the next level

Answer

Yes, linear hashing can be used but it is not so convenient as consistent hashing

Consistent hashing allows buckets to be added in any

order, whereas Litwin’s Linear Hashing (LH*) scheme requires buckets to be added one at a time in sequence

Coral: An Open CDN

Implement an open CDN Allow anybody to contribute Works with unmodified clients CDN only fetches once from origin server Runs in PlanetLab (http://www.coralcdn.org) Based on NSDI 2004 presentation and paper

Origin Server

Coral httpprx dnssrv

Coral httpprx dnssrv Coral

httpprx dnssrv

Coral httpprx dnssrv

Coral httpprx dnssrv

Coral httpprx dnssrv

Browser

Browser

Browser

Pool resources to dissipate flash crowds

19

Using CoralCDN

Rewrite URLs into “Coralized” URLs www.x.com → www.x.com.nyud.net:8090

Coral distributes the load

Who might “Coralize” URLs? Web server operators Coralize URLs Coralized URLs posted to portals, mailing lists Users explicitly Coralize URLs

20

Resolver Browser

Coral dns srv http prx Coral

dns srv http prx

Coral dns srv http prx

Coral dns srv http prx

Coral dns srv http prx

4 (get proxies) 4

2 5

3 probe client)

9 Found

8,11 looking up URL

1 6: proxy list

7 use proxy 10 cache and return www.x.com

.nyud.net

Coral dns srv http prx Coral

dns srv http prx

DNS Redirection

Return proxy, preferably one near client

Cooperative Web Caching

Coral Server Discovery

1. Each Coral server inserts its IP network prefix as key, its IP address as value

2. DNS server does DHT lookup on client IP prefix to find nearby Coral server

Details (multiple prefixes): Each Coral server uses traceroute to find nearby

routers Registers itself under IP of each nearby router Coral DNS server traceroutes to client Looks up each router IP address in mapping

Hierarchical DHT

A hierarchy of DHTs, with clustering at lower levels DHT based on XOR metric

Nearby (< 20 ms) Coral nodes form an L2 DHT L1: 60 ms L0: global Search in L2 DHT first If nearby copy exists, will find it first Only search L1, L0 if miss in lower level

Finding URLs

Look up the URL in a DHT key=URL, value=IP addr of Coral cache that has the URL

Coral cache fetches the page from that other cache

If DHT had more than one value for key, fetch

page from more than one In case one is down or slow

Coral lacks… Central management A priori knowledge of network topology

Anybody can join system Any special tools (e.g., BGP feeds)

Coral has… Large number of vantage points to probe topology Distributed index in which to store network hints Each Coral node maps nearby networks to self

Challenges for DNS Redirection

Internet Indirection Infrastructure (i3)

•  A DHT - based overlay network –  Based on Chord

•  Aims to provide more flexible communication model than current IP addressing

•  Also a forwarding infrastructure –  i3 packets are sent to identifiers –  each identifier is routed to the i3 node responsible for

that identifier –  the node maintains triggers that are installed by

receivers –  when a matching trigger is found the packet is

forwarded to the receiver

i3 II

•  An i3 identifier may be bound to a host, object, or a session

•  i3 has been extended with ROAM –  Robust Overlay Architecture for Mobility –  Allows end hosts to control the placement of

rendezvous-points (indirection points) for efficient routing and handovers

–  Legacy application support •  user level proxy for encapsulating IP packets to i3

packets

Source: http://i3.cs.berkeley.edu/

R inserts a trigger (id, R) and receives all packets with identifier id.

the host changes its address from R1 to R2, it updates its trigger from (id, R1) to (id, R2).

Mobility is transparent for the sender

Source: http://i3.cs.berkeley.edu/

A multicast tree using a hierarchy of triggers

Source: http://i3.cs.berkeley.edu/

Anycast using the longest matching prefix rule.

Source: http://i3.cs.berkeley.edu/

Sender-driven service composition using a stack of identifiers

Receiver-driven service composition using a stack of identifiers

Summary

Key applications Kademlia and Mainline DHT (XOR geometry) PAST and SCRIBE (Pastry) Akamai (consistent hashing)

Amazon (Dynamo, consistent hashing, ring geometry) Coral (XOR geometry)

Synthesis

We have covered building blocks for decentralized distributed systems Unstructured: Bittorrent, flooding, limited flooding, Bloom filter

summaries, depth-first search with backtracking, layered Structured: consistent hashing and DHTs, small world and location

swapping, network proximity, content placement, rendezvous Security: asymmetric and symmetric techniques, Sybil attack Additional techniques: gossip, anti-entropy, vector clocks, … Applications: Search, file exchange, secure and high performance

storage, multicast, DNS replacement, CDN, Future Internet?

Internet (TCP/IP)

Unstructured

Structured

Small world

Random

N/A

BT

DHTs

Freenet

Flooding

Limited flooding

Layered

Bloom Filt.

Gnutella

Tracker

Tit-for-tat

Depth first

Location swapping

Node id clustering

Consistent hashing

XOR

Kademlia

Coral

Ring

Trees/cubes

Rendezvous, replication, multicast, …

Proximity

Hybrids

Internet (TCP/IP)

Cluster

Wide-area

(unstructured)

Search

Storage

Rendezvous

Search

Storage

Rendezvous

Consistent hash (O(1) DHT)

Selective flooding

Replication, Gossip, etc.

Example: Dynamo

Limited flooding / depth

first / Bloom filters

Examples: Gnutella and

Freenet

Tracker

Example: BitTorrent

Wide-area (structured)

Search

Storage

Rendezvous

DHT

Examples: Lookup: Chord, CAN,

Kademlia Storage: PAST

Rendezvous: Scribe (for multicast), i3

Good for name/value data, note flat address space, one node is responsible, churn is

a concern Good for arbitrary data and

search functions, can aggregate routing info,

structure improves scalability

Consistent hashing alleviates network problems and

eventual consistency can be achieved through replication

and synchronization

Current research involving overlays

Virtualized cloud systems and data centers Cluster-based data analytics and processing Social networks Data dissemination (live video, subscriptions) Control plane for Future Internet proposals

The Cloud

… Google Microsoft

Amazon Yahoo …

Grading

Course grading will be based on the final exam and the assignments.

Assignments give bonus points for the exam

Max exam points 60, max bonus points 12 Course exam 27.2.2013 16:00 CK112 Final exam on 19.4.2013 16:00 in A111

Exam

Four essay questions, answer three, 18 max points, max 6 points per answer

Example questions: Q1: A) Compare Gnutella and Kademlia. What are the key

differences and similarities of the systems. B) Consider the structure and routing functions of the

systems. C) How is content placed on the overlay nodes in these two systems?

Q2: A) What problem does BitTorrent solve?

B) Compare BitTorrent to HTTP. C) What are the key metrics in BitTorrent data delivery? Briefly consider the significance of each metric.

Main theme Prerequisites Approaches learning goals Meets learning goals Deepens learning goals

Overlay and peer-to-peer networks: definitions and systems

Basics of data communications and distributed systems (Introduction to Data Communications, Distributed Systems)

Knowledge of how to define the concepts of overlay and peer-to-peer networks, and state their central features Ability to describe at least one system in detail

Ability of being able to compare different overlay and p2p networks in a qualitative manner Ability to assess the suitability of different systems to different use cases

Ability to give one’s own definition of the central concepts and discuss the key design and deployment issues

Distributed hash tables

Basics of data communications and distributed systems (Introduction to Data Communications, Distributed Systems) Big-O-notation and basics of algorithmic complexity

Knowledge of the concepts of structured and unstructured networks and the ability to classify solutions into these two categories Knowledge of the basics of distributed hash tables Ability to describe at least one distributed hash table algorithm in detail

Ability of being able to compare different distributed hash table algorithms Ability of designing distributed hash table-based applications Knowledge of key performance issues of distributed hash table systems and the ability to analyze these systems

The knowledge of choosing a suitable distributed hash table design for a problem Familiarity with the state of the art

Reliability and performance modelling

Basics of probability theory Basics of reliability in distributed systems

Ability to model and assess the reliability of overlay and peer-to-peer networks by using probability theory Knowledge of the most important factors pertaining to reliability

Ability of analytically analyzing the reliability and performance of overlay and peer-to-peer networks Understanding of the design issues that are pertinent for reliable systems

Familiarity with the state of the art

Content distribution Introduction to Data Communications

Knowledge of the basic content distribution solutions Ability to describe at least one overlay and p2p network based content distribution solution

Knowledge of different content distribution systes and the ability to compare them in detail Knowledge of several content distribution techniques

Familiarity with the state of the art

Security Basics of computer security Knowledge of the basic security issues with overlay and p2p networks Knowledge of the sybil attack concept

Ability to discuss how security problems and limitations can be solved Knowledge of how to prevent sybil attacks

Knowledge of how to prevent sybil attacks Familiarity with the state of the art

After this course

Seminars on overlays & related topics M.Sc. Thesis

Many topics in this area

Course development ideas

More focus design of systems Instructed group work on designing a large-scale system Design & partial implementation

Have a nice Spring!

Remember course feedback