cs-495 distributed systems fabián e. bustamante, winter 2004 introduction distributed systems goals...

CS-495 Distributed Systems

Fabián E. Bustamante, Winter 2004

Introduction

Distributed Systems

Goals of Distributed Systems

Hardware and Software Concepts

Client-Server Model

CS-495 Distributed Systems2

Distributed System

A distributed system is a collection of (1) independent computers that (2) appears to its users as a single coherent system.

e.g.NOW

Workflow information system

WWW


Goals

To make them worth the effort– Connecting users and resources

• make easy to access remote resources &• share them in a controlled way

– Transparency – hide the fact that resources are distributed– Openness – so that they can be

• Applications can be portable• Different implementations can interoperate• DS are flexible and extensible

– Scalability – world-wide connectivity• Scalable in

– Size - # of users/resources– Geographic span– Administratively


Transparency

Kinds of transparency

Degree of transparency

Transparency Description

AccessHide differences in data representation and how a resource is accessed

Location Hide where a resource is located

Migration Hide that a resource may move to another location

RelocationHide that a resource may be moved to another location while in use

ReplicationHide that a resource may be shared by several competitive users

ConcurrencyHide that a resource may be shared by several competitive users

Failure Hide the failure and recovery of a resource

Persistence Hide whether a (software) resource is in memory or on disk


Scalability

Impediments to scalability

Concept Example

Centralized services A single server for all users

Centralized data A single on-line telephone book

Centralized algorithmsDoing routing based on complete information


Scaling Techniques

Hiding communication latency – avoid waiting for responses to remote service request when possible– Asynchronous communication– Reduce overall communication

Distribution – split a component into parts & distribute them across the system

Replication– Increase availability– Helps balance the load– May hide communication latency


Hardware Concepts

Multiprocessors MulticomputersSingle address space

shared by all


Multiprocessors

Single address space shared by all– Coherency is easy but scalability a problem

Solution – cache – now coherency is an issue


Multiprocessors – switch-based

Competition for the bus is still an issue

Switch-based systems?– Crossbar switch & Omega switching network


Multicomputers

Easier to build – problem is interconnection– Homogenous – System Area Network

• Homogeneous nodes rack mounted and• Interconnected by a high-performance network (e.g. grid,

hypercube)

– Heterogeneous – both in the nodes and the interconnection network

• Problem – hiding heterogeneity is hard


Software Concepts

OS for multicomputers/multiprocessors can be divided in two categories– Distributed OS – tightly-coupled systems– Network OS – loosely-coupled systems

From a NOW, to get a distributed system you need middleware

System Description Main Goal

DOSTightly-coupled operating system for multi-processors and homogeneous multicomputers

Hide and manage hardware resources

NOSLoosely-coupled operating system for heterogeneous multicomputers (LAN and WAN)

Offer local services to remote clients

MiddlewareAdditional layer atop of NOS implementing general-purpose services

Provide distribution transparency


Uniprocessor Operating Systems

OS for distributed systems and uniprocessors are very much alike– They act as resource managers for the underlying hardware

– Provide a virtual machine that hide the intricacies of the underlying hardware

Part of enabling resource sharing is protecting apps from each other– Dual mode of operation – monolithic OSs

Microkernels– Pros and cons


Multiprocessors Operating Systems

From uniprocessors to multiprocessors – a natural step

Protect OS data structures against concurrent access to guarantee consistency– Semaphores– Monitors


Multicomputer Operating Systems

No shared memory – communication through message passing

You can – Exposed message passing to application – Hide it through distributed shared memory


Distributed Shared Memory

Since MP seems to be more difficult to program for

Distributed Shared Memory– AS divided into pages– Pages distributed among

machines

Performance improvements– Replicate read-only pages

– Replicate all pages

– Weaker than strict consistency

– Larger pages sizes → potential problem of false sharing

CPU 1 references page 10

Page 10 is read only and replication is used


Network Operating System

Various machines interconnected; users can access remote resources

No assumptions of– Homogeneous underlying hardware– Managed as a single system

Pros and Cons– Harder to use– Harder to manage– Very easy to scale


Middleware

Remember the definition of a distributed system– Distributed OS – not intended to handle independent

machines– Network OS – does not provide a view of a single coherent

system

Most distributed systems are built from NOS + middleware


Middleware Models and Services

To make development and integration simple – adopt a paradigm– Everything is a file

– Remote Procedure Calls

– Distributed Objects

– Distributed Documents

Services common to many middleware systems– Communication

– Naming

– Persistence

– Distributed Transaction

– Security

Middleware and openness


Comparison between Systems

Item

Distributed OS

Network OSMiddleware-based OSMultiproc. Multicomp.

a) Degree of transparency

b) Same OS on all nodes

c) Number of copies of OS

d) Basis for communication

e) Resource management

f) Scalability

g) Opennessa) High/Low

b) Yes/No

c) 1/N

d) Shared memory/messages/files/model specific

e) Global central/global distributed/per node

f) No/Yes/So-So

g) Closed/Open


Comparison between Systems

Item

Distributed OS

Network OSMiddleware-based OSMultiproc. Multicomp.

Degree of transparency Very High High Low High

Same OS on all nodes Yes Yes No No

Number of copies of OS 1 N N N

Basis for communication Shared memory Messages Files Model specific

Resource management Global, centralGlobal, distributed

Per node Per node

Scalability No Moderately Yes Varies

Openness Closed Closed Open Open


Client-Server Model

Clients and Servers as a model for distributed applications

Layering: general organization of an Internet search engine into 3 different layers


Multitiered Architectures

The three logical levels → many physical distributions and potentially multiple tiers

Alternative client-server organizations


Modern Architectures

Distribution can be – Vertical – what we have seen so far– Horizontal – distributing the clients and servers

themselves• Distributed web server.

• P2P system


Hints for Computer System DesignWhy it helps?/

Where it helps?

Functionality Speed Fault-tolerance

Completeness Separate normal & worst case

Safety first

Shed load

End-to-End End-to-End

Interface Do one thing well

Don’t generalize

Get it right

Don’t hide power

Use procedure arguments

Leave it to the client

Keep basic interface stable

Keep a place to stand

Make it fast

Split resources

Static analysis

Dynamic translation

End-to-End

Log updates

Make actions atomic

Implementation Plan to throw one away

Keep secrets

Use a good idea again

Divide and conquer

Cache answers

Use hints

Use brute force

Compute in background

Batch processing

Make actions atomic

Use hints

cs-495 distributed systems fabián e. bustamante, winter 2004 introduction distributed systems goals...

Documents

distributed systems

systems os

distributed openness

systems network os

categories distributed

network slide

issue slide

hard slide