Prof. Roberto Baldoni Ing. Alessia Milani Ing. Leonardo Querzoni Ing. Silvia Bonomi

Università di Roma “La Sapienza”Dipartimento di Informatica e Sistemistica

Distributed SystemsCorso di Laurea Specialistica in Ingegneria Informatica

AA 2006/2007

Introduction

Prof. Roberto Baldoni

Ing. Alessia Milani

Ing. Leonardo Querzoni

Ing. Silvia Bonomi

www.dis.uniroma1.it/~midlab

A definition

• A distributed system is a set of spatially separate entities, each of these with a certain computational power that are able to communicate and to coordinate among themselves for reaching a common goal

Primary Goal: sharing data/resources

Problems• Synchronization• Coordination

Coordination has to be implemented taking into account the following conditions that deviates from centralized systems:

1. Temporal and spatial concurrency2. No global Clock3. Failures4. Unpredictable latencies

These limitations restrict, for example, the set of coordination problems we can be solve in a distributed setting

Distributed Systems Examplesin

ile But also.....Service Oriented ArchitecturesOverlay NetworksGridP2PPervasive Systems&Ubiquitous Computing

Pervasive Systems

Internet everywhereOne-persons Many computersMobility……….. How can I keep consistent my mailbox???

Characteristics..and Challenges

• Heterogeneity• Openess• Security• Scalability• Fault Tolerance• Concurrency• Transparency

Heterogeneity

• Networks• Hardware• Operating Systems• Programming Languages• Implementations from different Developers

Soluzioni• Middleware (from RPC to Service oriented Architectures)• Mobile code and Virtual Machine

Openess

• Capability of a system to be extended and re-implemented• Necessary condition, set of documents with software interfaces

• Interface Definition Language (it describes the syntax and the semantic of a service/component, available functions/services, input parameters, exceptions, etc)

• Problem: semantic description of a service.

• A specification of a service/componet is well-formed if it is : – Complete. A specification is complete if everything related to the

implementation has been specified. If something has not been specified, the designer needs to add implementation dependent details.

– Neutral. A specification is neutral is it does not offer any detail on a possible implementation

Openess (ii)

• Interoperability. The capability of two systems to cooperate by using services/components specified by a common standard

• Portability. The capability if a service/component implemented on a distributed system A to work on a system B without doing any modification

• Flexibility. The capacity of a system to configure/ orchestrate components developed by various programmers

• Add-on Features. The capacity of a distributed system of adding components/services and be integrating in a running system

Openess (iii)

• Other recent capabilities:– Evolvability. The capacity of a system to evolve in time for example leaving active two different version of the same service. – Self-* (self organization, self management, self healing etc). The capacity of a system to reconfigure, to manage itself without uman

intervention

• The number of independent software developers make very complex the development of a distributed platform

• Examples:– RFC for internet– JBoss for J2EE platform

Security

• Confidenziality (protection against the interception of data from unauthorized users)

• Integrity (protection against data alteration)

• Availability (protection against the interference in the access to a resource)

Scalability

• A system is scalable if it remains running with adequate performance even if the number of “users” grow up of orders of magnitude

• Centralization is against scalability :– Service (single service for all users)– Data (single data structure for all processes)

Date Computers Web servers

1979, Dec. 188 0

1989, July 130,000 0

1999, July 56,218,000 5,560,866

Computers connected to the internet

Scalability (ii)• It becomes necessary using:

– Service Replication• Coordination Problems

– Data Replication• Consistency Problems

– Distributed Algorithms• No node has the current state of the whole system• Nodes base their decisions on data they own• A failure of a node should not compromize the goal of the algorithm

• Geographical Scalability

Internet-scale Applications

Enterprise Data Centers

Scalable Consistency-based Applications

First Open Workshop Budapest 21-3-2007

What is a large-scale distributed system?

Cooperative Information Systems

Scalable QoS based Applications

eGov eGov enveloenvelo

Failure Management

• Failure detection– Example: Checksum detects a corrupted packet

• Failure masking– Example: message retransmission

• Tolerating Failures– Example: intrusion tolerant systems

• Failure Recovery– Example: completing long running computation

• Redoundancy– Example: DNS

Concurrency

• Multiple access to shared resources– If clients invokes concurrently read and write methods on a shared

variable– Which value returns each read?

• Coordination • Synchronization

Transparency

• Access: allow to access remote and local resources with the same operations• Location: allows to access resources without knowing their physical location• Concurrency: allows a set of processes to run concurrently on shared resources

without interfering among themselves • Failures: allow to mask failures in order that users can complete remaining requested

operations • • Mobility: allows to move resources and users without influencing operation issued by

users• Performance: allow system reconfiguration changing the load

• Performance of a solution based on a distributed system not always improve with respect to a solution based on a centralized system.

Layering hw and sw

Interaction Models

client/server peer-to-peer

Interaction Models

• Interaction models impact on scalability, availability, cost, security performances

• Eg. client/server with replicated services:

availability, scalability

performance: replication imposes an extra work for maintaining consistency despite replica crashes

Web proxy server

Client

server

serverClient

Web applets

a) client request results in the downloading of applet code

Web server

ClientWeb serverApplet

Applet code

Client

b) client interacts with the applet

Thin clients and compute servers

ThinClient

ApplicationProcess

Network computer or PCCompute server

network

Spontaneous networking in a hotel

Internet

gateway

service

Music service

serviceDiscovery

Camera

Guestsdevices

LaptopTV/PC

Hotel wirelessnetwork

Real-time ordering of events

receive

Physical time

receive receive

receive receive receivet1 t2 t3

receive

Processes and channels

process p process q

Communication channel

Outgoing message buffer Incoming message buffer

receivem

Middleware: problems to face

• Heterogeneity: OS, clock speeds, data representation, memory, architecture HW

• Local Asynchrony: load on a noad, diffeent OW, Interrupts • Lack of global knowledge: knowledge propagates through messages whose

messages whose propagation times will be much slower than time taken by the execution of an internal event

• Network Asyncrony: propagation times of message can be unpredictable.

• Failures of nodes or network partitions.

• Lack of a global order of events

This limits the set of problems that can be solved through deterministic algorithms on some distributed systems

Prof. Roberto Baldoni Ing. Alessia Milani Ing. Leonardo Querzoni Ing. Silvia Bonomi

Documents

Mais Classificazione botanica Prof. Guido Baldoni · Radici...

LCBM: Statistics-based Parallel Collaborative …...LCBM:...

Baldoni Pub

1 Memorie Condivise Distribuite Prof.Roberto Baldoni, Ing......

C Ratti Ed Lojeski Manifesto futurista Federico Mochi vox...

Corso di Psicologia Dinamica Franco Baldoni Facoltà di...

Prof. Guido Baldoni Prof. Guido Baldoni Fasi dellinsilamento...

The clinical matching - iasa-dmm.org 2018 Slides(5... ·...

The clinical matching - iasa-dmm.org 2018 Slides.pdf ·...

Introduzione - Dipartimento di Informatica...

Prof. Guido Baldoni Prof. Guido Baldoni Miglioramento...

Final Notes Sestriere 23 Gennaio 2015 Roberto Baldoni.

Analysis of Deterministic Tracking of Multiple Objects...

Sistemi Distribuiti AA 2004/2005 Prof. Roberto Baldoni Ing.....

Prof. Guido Baldoni Prof. Guido BaldoniClassificazione...

Prof. Guido Baldoni Prof. Guido Baldoni Definizione e...