13
1/13 Artjom Lind ([email protected]) 11.09.2013 https://courses.cs.ut.ee/2013/hs Hajussüsteemid MTAT.08.024 Distributed Systems Features of Distributed Systems

Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

1/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Hajussüsteemid

MTAT.08.024 

DistributedSystemsFeatures of Distributed Systems

Page 2: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

2/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Features

● Sharing resources● Openness● Parallelism● Scalability● Fail-safity● Transparency

Page 3: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

3/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Sharing Resources

● What is resources ?● Hardware

– Not every single resource is for sharing● Data

– Databases– Proprietary software– Software production– Collaboration

Page 4: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

4/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Sharing Resources

● Different resources are handled in different ways, there are however some generic requirements:

– Namespace for identification– Name translation to network address– Synchronization of multiple access

Page 5: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

5/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Client-Server

● Server is sharing resources● Clients demand resources● Service – abstract entity, which is powered by multiple

processes● Resource names are usually server specific● Client is demanding services and server is indicates

services● Example: WWW● Disadvantages of Client-Server model

– Not all resources can be shared with this model– Non-effective for internal sharing

Page 6: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

6/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Distributed objects

● Objects (resources)– Unique identifier– Objects can migrate over network

● Object users● Messaging between objects

– Responses are optional● Disadvantages of object model

Page 7: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

7/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Openness

● Hardware openness– Adding new devices, compatibility– Hardware API specification

● Software openness– Adding new features, new services, communication with

other software– Open API and protocol

● Open system can contain hardware of software of different vendors

● Each component can be exchange according to interface● Validation – check if component is complaint according

to the standards

Page 8: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

8/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Parallelism

● Centralized system, with n processors makes computing n times faster (in ideal case)

● Distributed system, with m single-process machines gives also m times faster computations but all processes run in different machines

● Users running the same program simultaneously● Many server processes work in parallel for serving

different client requests● Drawbacks of parallelism (synchronization)

Page 9: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

9/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Scalability

● Distributed systems design should allow to extend the system in future

● Extending system not affecting the software● System has to remain stable even with big

number of attached components● Scalability is measured by performance of one

single processor (overall performance / number of processors)

● Design doesn't allow services to be unavailable

Page 10: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

10/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Scalability affecting factors

● Centralized services – one central server● Centralized data – one central database● Centralized algorithms – decision making

based on all data

Page 11: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

11/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Scalability of algorithms

● Requirements:– Each single machine doesn't rely on all data

in distributed system– Decision is made based on local information

only– Failure of one machine does not block all

the others– Absence of global synchronization clock

Page 12: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

12/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs

Scalability methods

● Asynchronous communication in order to maintain geographic scalability

● Replication, Caching● Vertical scalability – different nodes for

different layers in the system● Horizontal scalability – many nodes in on

particular layer of the service

Page 13: Distributed Systems - Kursused€¦ · Centralized system, with n processors makes computing n times faster (in ideal case) Distributed system, with m single-process machines gives

13/13

Artjom Lind ([email protected])11.09.2013https://courses.cs.ut.ee/2013/hs