Presentation: Architectural Design for Distributed Systems Objektorienteret netværkskom

Presentation:

Architectural Design for Distributed Systems

Objektorienteret netværkskom.

Agenda

• Architectural models• OOA/OOD and distributed system diff.• Layering• Interface Partioning & Granularity• Adapting patterns for distributed usage

A few words of warning:Assumed students are knowledgeable about OOA/OODWe will only look at some aspects of distributed architectureNot much literature available on this subjectNo textbook solutions to this problemThis lecture is just for inspiration – not a dictate

Architectural models

• Architectural model: Description of the components of a system and the relationship between them.

• Describe the components of systems and their interaction; describe mapping of components to computers.

• Define useful patterns for the distribution of data and workload.

• Define the functional roles of components and the patterns of communication between them.

Basic architectural models

Server

Client

Client

invocation

result

Serverinvocation

result

Process:Key:

Computer:

Client-server: WWW, OO middleware, distributed systems in general

Others: service, peer-to-peer, proxy, mobile code, spontaneous network

Read more in Colourius et al.

Designing Distributed Systems

• Use OOA/OOD (or other method)• Same procedure as with stand-alone system design

• Use ”best-practices” aka ”design patterns”

• BUT: beware of the pit-falls of distributed design

• (Guerraoui & Fayad)

• Beware of “Gold Plating”• Do not use excessive amount of time on design

• “Death By Interfacing/Layering”

Basic OO Design – Use Case Driven

Use Case N

Actor 1

Use Casespec.

“Models” the domaine.g. an Account or

Sensor.

System/ActorInteraction

Use Case impl.Links Model & Boundry

«control»

«boundary»

«entity»

Domain Model for Use Case N

Logic DomainModel from the Analysis

OOA

Eckel’s ROPES ModelArchitecture design

Scope: nodes, packages (sub systems), components (e.g. a driver DLL), tasks

Mechanistic design

Scope: Group of collaboratingclasses Class

Class

Class

Class

Node

Package

ComponentActive object

Detailed designScope: Class

Class name

Attributes

Operations

Bd. s193

Use Sub-systems for Structuring

• At least use two sub-systems:• a client

• a server

• More sub-systems may be introduced as needed

Getting Distributed

• Until now – stand-alone & single process• Distributed Systems much more complex• We focus only on OO systems• Ensuring optimal design?• Client side / server side• Three things to consider:

• Layering• Granularity issues• Adapting Design Patterns

Design with Layers

• Client Presentation• Provides a user interface to the end-users. • Thin/Rich. MVC.

• Server Side Presentation• Building a response to the Client Presentation tier.

• Server Side Business Logic• Use Case implementation. Control classes. Business logic.

• Server Side Domain Model• Domain Model. Entity classes.

• Enterprise Integration / Communication• Legacy system. Web services.

• Persistence / Resource• Relational Database. File-system.

Interface Partitioning & Granularity

• OOA/OOD maps real world model to domain model• Granularity always an issue• Stand-alone & single process systems:

• Should we have a fine-grained model – with one or more control classes pr use case and a detailed domain model?

• Should we have a coarse-grained model with only a few classes?

• In distributed systems, this gets worse

Three Aspects of a Distributed Object System

Granularityof Model

Interface Design

SystemPartitioning

Three important aspects when producing a Distributed object Model which has impact on eachother

Anti-pattern: Round tripping vs. partition of resources vs. desired level of interface design resolution

(some might want to map system objects 1-1 with real world objects, accounts and transactions

Producing a real world

OO mapping is important

Fine Grained = Natural Mapping = Easy to Understand

Fine Grained = Slow Distributed performance

Coarse Grained = Better Distributed performance

Fine Grained = Easy to Partition = Good Scalability

Coarse Grained = Harder to Partition = Poorer Scalability

Using Design Patterns

• Provides guidelines, not actual implementation • Proven track record• Reusable• Help you communicate your design ideas to other

designers

Anti-pattern: Gold Plating – spending excessive amount of time on design

Design Patterns Examples

• ”Classic Design Patterns”:• Singleton• Observer• Iterator• Facade• Proxy (you have already seen this)• Factory• Broker• Many others

• All may be used, but some should be adjusted• Remember: patterns are only for inspiration

• NOT dictate• Lets look at a few

Façade Pattern (also GoF) + DTO’s

Used for encapsulation and decoupling – usually one pr sub-system- Session Façade pattern, Façade at the Distribution Boundary

Used for encapsulation and decoupling – usually one pr sub-system- Session Façade pattern, Façade at the Distribution Boundary

The entire Client sub-system is decoupled from the server sub-systemand a Client Proxy hides the complex network detail of a distributed system

By using Façade pattern, only a few objects needs to be made Remote

The entire Client sub-system is decoupled from the server sub-systemand a Client Proxy hides the complex network detail of a distributed system

By using Façade pattern, only a few objects needs to be made Remote

Replicating Objects / Layering

• Client Presentation tier• Provides a user interface to the end-users. • Thin/Rich. MVC.

• Server Side Presentation tier• Building a response to the Client Presentation

tier.• Server Side Business Logic tier

• Use Case implementation. Control classes. Business logic.

• Server Side Domain Model tier• Domain Model. Entity classes.

• Enterprise Integration tier• Legacy system. Web services.

• Persistence tier / Resource layer• Relational Database. File-system.

Using D

TO

’s + facades

Replicating Objects /Data Transfer Object PatternHow to do it in OOMI?

Replication Problems

• Concurrency Issues• Optimisitic Concurrency -> Best shot• Pessimistic -> Lock the objects• Intelligent Versioning -> Update clients• Huge complexity if framework does not support

The Observer Pattern (GoF)

Publisher

Subscriber

Subscriber

Subscriber

Generates a lot of network trafficeven though the three subscribersresides in the same process spaceAnti-pattern: Round-trip

Publish-Subscribe Channel (Distributed Observer)

channel

Subscriber

Subscriber

Subscriber Channel

Publisher

Only one notify message between thechannel objects as opposed to the naive Observer pattern.

channel

Subscriber

Subscriber

Subscriber Channel

Publisher

Only one notify message between thechannel objects

Iterator Pattern

Client Iterator Collection

1. Create iterator

2. Get next item3. Get next item4. …..

Generates a lot of network traffic

Distributed Iterator

Client Iterator Collec-tion

1. Create iterator

2. Query3. Get next item

4. Get next item

5. …….

Result of 2.

Only one call needed to transfer all objects. Design more than one “next” method to allow for different number of objects wanted

Factory Pattern

• How to create objects? Can not instantiate• We need a staging point – a Factory object

• Distributed version: Object Factory• For creating, finding & managing both DTO and remote

objects• Often seen in conjunction with façade, replicating objects

& proxy

Remember

• You have been presented with some basic input for the design of distributed systems

• This is only for inspiration not a dictate• Even though we have the ideals of transparency –

one must remember the differences that does exist