Internet-Scale Interoperability

ICS 123

Internet-Scale Interoperability

ICS 123Richard N. Taylor and Eric M.

Dashofy*UC Irvine

http://www.isr.uci.edu/classes/ics123s02/

* With the usual thanks to David Rosenblum

ICS 123

2

Topic 15Internet-ScaleInteroperability

Distributed Applications on the Internet

A natural paradigm:Asynchronous Event-Driven Interaction

A natural paradigm:Asynchronous Event-Driven Interaction

The Internet as a platform for buildingdistributed systems

Multinational corporate workflow

Worldwide software deployment

Monitoring of global financial markets

ICS 123

3


Example:An Arbitrage Application

Yen Trader

in Zurich

Yen Traderin London

Arbitrageur in

New York

125.92 per $

125.90 per $

ICS 123

4


Technologies for Event-Based Interaction

Many relevant technologies for LANsTheoretical models

Communication andCollaboration Systems

Low-level eventmanagers

Tool IntegrationFrameworks

Implicit invocation

Software buses

Event-based specification,analysis, debugging

Active database systems

Software agent technology

Event-action systems

New technologies are needed thatoperate at Internet scale

New technologies are needed thatoperate at Internet scale

ICS 123

5


Some Statistics from Network Wizards (www.nw.com)

• January 2002:– 147,344,723 hosts in DNS– 236 active level 1 domains (e.g., .edu)

» Top 10: .com (24,863,331), .net (18,853,655), .edu (6,085,137), .jp, .uk, .us, .mil, .de, .ca, .au

– 2,867,326 level 2 domains (e.g., .uci.edu)– 35,967,238 level 3 domains (e.g., .ics.uci.edu)– This probably does not include all the “hidden” hosts on the

internet (I.e. those behind firewalls or with a 10.x.x.x address)

•These numbers are– Almost 100% higher than the January 2000 numbers, which

are:» nearly 75% higher than the January 1999 numbers» more than 2 times the January 1998 numbers» about 4.5 times the January 1997 numbers

ICS 123

6


Eight Fallacies of Distributed Computing (Peter Deutch)

•1. The network is reliable

•2. Latency is zero

•3. Bandwidth is infinite

•4. The network is secure

•5. Topology doesn't change

•6. There is one administrator

•7. Transport cost is zero

•8. The network is homogeneous

Discuss: Why were people able to make such assumptions with relatively little risk in the past?

ICS 123

7


Attributes of Internet Scale

Huge numbers of users

Vast numbers of events

Network latency

Autonomy

Resource accounting

Security

Mobility

ICS 123

8


Some Warning Signs That a Technology Is Not Scalable

Extrapolating the success of a LAN-basedtechnology

Objects that manage their own set of "eventlisteners"

JavaBeans

"... the server ..."

Roll-your-own frameworksCORBA Event Service

ICS 123

9


Event Observation and Notification Terminology

invoker

object of interest

@ London

object of interest@ Zurich

Event Service

pattern ofevents

invoker

notification

120.92 per $

120.90 per $

interested party

@ New York

ICS 123

10


Event Observation and Notification Activities

1. Making a class of events observable1. Making a class of events observable

2. Expressing interest in a pattern of events2. Expressing interest in a pattern of events

3. Occurrence of an event3. Occurrence of an event

4. Observing an event4. Observing an event

5. Relating observations with each other5. Relating observations with each other

6. Notifying an interested party6. Notifying an interested party

7. Receiving a notification7. Receiving a notification

8. Responding to a notification8. Responding to a notification

ICS 123

11


SIENA: Scalable Internet Event Notification Architectures

Event Notification

Service

Advertise

Publish Subscribe

Service Access Points

Notifications

ICS 123

12


Goals of SIENA

Research Project

•SIENA provides an event observation and notification service...

•Scalability– vast dimensions, scarce connectivity, heterogeneity,

openness, decentralization

•Expressiveness– flexible data modeling– accurate selection– aggregation of events

ICS 123

13


Interface of SIENA

•SIENA:– publish(notification n)– subscribe(URI subscriber, pattern p)– unsubscribe(URI subscriber, pattern p)– advertise(URI publisher, filter p)– unadvertise(URI publisher, filter p)

• Interested party:– notify(notification n)

ICS 123

14


Notification Model in SIENA

•A notification is a list of attributes

string class = travel/airlines/offersdate starts = Jun 99date expires = Aug 99string origin = DENstring destination = MXPstring carrier = BAfloat price = 850string currency = USD

string class = travel/airlines/offersdate starts = Jun 99date expires = Aug 99string origin = DENstring destination = MXPstring carrier = BAfloat price = 850string currency = USD

attribute=(type,name,value)

ICS 123

15


Filters

•A filter is a list of attribute filters

string class >* travel/airlines/date starts < Jul 99date expires > Jul 99string origin = DENstring destination = MXP

string class >* travel/airlines/date starts < Jul 99date expires > Jul 99string origin = DENstring destination = MXP

attribute filter=(type,name,operator,value)

ICS 123

16


Patterns

•A pattern is an algebraic expression whose basic elements are filters

string class >* system/files/writestring file = /etc/passwdint userid > 0

string class >* system/files/writestring file = /etc/passwdint userid > 0

string class >* system/files/writestring file = /var/adm/utmpint userid > 0

string class >* system/files/writestring file = /var/adm/utmpint userid > 0

and then

ICS 123

17


Some Design Choices

Advertisementsnone vs. required vs. optional

Notificationstyped vs. untyped

What to Broadcastnotifications, subscriptions and/or advertisements

Pattern Languagesequences vs. REs vs. REs with negation vs. ...

ICS 123

18


Simulation StudiesSimulation Variables

Routing Algorithms

Sites (of Service Access Points)

number, inter-site cost

Servers

site locations, topologies

Objects of Interest

number, message rates, site locations, server connections

Interested Parties

number, patterns, message rates, site locations, server connections

Event service simulated on top of generated Internettopologies

GT-ITM (Georgia Tech)

ICS 123

19


Distributed Servers

Maintain Local State InformationKnown Advertisements

Known Subscriptions

Maintain Links with Neighboring Servers inService Topology

Which neighbors provide notifications

Which neighbors need notifications

Perform Matching of Notifications to LocalSubscriptions

ICS 123

20


Centralized Server Architecture

ServerServer

ICS 123

21


Hierarchical Architecture

… ……

…

…

ICS 123

22


General Graph Architecture

ICS 123

23


Sample Simulation Outputs

•Cost per Service: total cost of all messages involved in a single service request

•Delay per Service: time delay between departure of first message and arrival of last one for a single service request

•Cost per Site: total cost of all messages handled by a site

•Cost per Link: total cost of all messages passing through one link

Both averages and totals computed for metrics

ICS 123

24


Hierarchical Architecture with Subscription Forwarding

…

…

…

All subscriptions propagate upwards

Notifications propagate only alongactively subscribed paths

ICS 123

25


Hierarchical Architecture with Advertisement Forwarding

…

…

…

All advertisements propagateupwards

Subscriptions propagate upwardsonly as far as necessary

Notifications propagate only alongactively subscribed paths

ICS 123

26


Side Topic: Delivery without Routing

•We have had large-scale information delivery without routing for ~100 years. Where? How?

ICS 123

27


Side Topic: Delivery without Routing

•We have had large-scale information delivery without routing for ~100 years. Where? How?

Rabbit Ears

ICS 123

28


What About Event Delivery Over Wireless Networks?

•Routing and distribution greatly simplified– No need for sophisticated routing algorithms– No need for carefully-designed server topology

•But reliability may become greatly complicated

– Simple wireless handheld devices, such as pagers– One-way communication– No guarantee of delivery– Need sophisticated broadcast algorithms

ICS 123

29


Approximate State Synchronization

ICS 123

30



ICS 123

31



ICS 123

32



ICS 123

33



ICS 123

34


Middleware vs.Network Infrastructure

• Wasted explicit addressing in IP

• Redundant, unused routing information in DNS

• Implicitly-addressed, content-based routing (CBR)

• Hardware routers with CBR routing tables and algorithms

Today:Siena over TCP/IP

Future:Siena alongside TCP/IP

SIENA

SIENA Client

TCP/IP

SIENA Client

TCP/IPCBR

Documents

Internet-Scale Interoperability