ics362 distributed systems

51
ICS362 Distributed Systems Dr. Ken Cosh Week 3

Upload: dexter

Post on 17-Jan-2016

41 views

Category:

Documents


0 download

DESCRIPTION

ICS362 Distributed Systems. Dr. Ken Cosh Week 3. Review. Architectural Styles Layered Architectures Object-Based Data-Centred Event-Based System Architectures Centralised / Decentralised / Hybrid. This Week. Processes Threads Virtualisation Clients Servers Code Migration. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: ICS362 Distributed Systems

ICS362 Distributed Systems

Dr. Ken Cosh

Week 3

Page 2: ICS362 Distributed Systems

Review

Architectural Styles– Layered Architectures– Object-Based– Data-Centred– Event-Based

System Architectures– Centralised / Decentralised / Hybrid

Page 3: ICS362 Distributed Systems

This Week

Processes– Threads– Virtualisation– Clients– Servers– Code Migration

Page 4: ICS362 Distributed Systems

What is a Process?

OS– A program in execution

Key Issue– Managing and Scheduling Processes– Both in terms of OS and in terms of Distributed

Systems

Page 5: ICS362 Distributed Systems

Processes

The OS creates virtual processors– Each for running different programs

OS has a process table– To keep track of the virtual processors– CPU register values, memory maps, open files, privileges,

etc.

OS manage these independent processes– One process can not affect another– i.e. concurrency transparency

Page 6: ICS362 Distributed Systems

Achieving Concurrency Transparency

Computationally Expensive– Allocation

Create complete independent address space Initialising memory segments

– Potentially moving existing data

– Switching Saving CPU context Modify Memory Management Unit (MMU) Potentially swapping processes between main memory and

hard disk– If more processes than main memory

Page 7: ICS362 Distributed Systems

Threads

Threads are like processes– Executing code independently from other threads– But with no attempt at concurrency transparency

Threads can share the same address space

– Computationally less expensive– But programming requires more intellectual effort

Page 8: ICS362 Distributed Systems

Multithreading

Consider a spreadsheet program– Editing one cell can cause updates to formulae

throughout multiple sheets.– How do we allow user input concurrently with

calculations? And perhaps autosaving

Page 9: ICS362 Distributed Systems

Thread Implementation

Normally in the form of a thread package– Containing operations to create and destroy

threads & synchronisation variables such as mutexes.

Implementation can be done in different ways– Executed entirely in user mode– Kernel responsible for scheduling– Hybrid

Page 10: ICS362 Distributed Systems

User Level thread library

All thread administration is kept in user’s address space;

– Cheap to create & destroy threads

Switching between threads is also cheap

However, a blocking system call will block the process – and hence any other threads in the process.

– E.g. when waiting for input

Page 11: ICS362 Distributed Systems

Kernel level threads

That problem is avoided by implementing threads in the OS kernel.

Then, every thread operation is carried out by the kernel, through system calls– Creation, deletion, synchronisation etc.

Switching threads becomes as expensive as switching processes

Page 12: ICS362 Distributed Systems

Hybrid Implementation

Light Weight Processes (LWP)– LWPs run in user space in the context of a single

process, but several LWPs can run per process with a shared address space.

– Creation, deletion operations etc. run in user space without intervention by the kernel.

Page 13: ICS362 Distributed Systems

Threads in DS

A convenient way of allowing blocking system calls without blocking an entire process.

In DS, threads can be separated to different processors (clients or servers)

Page 14: ICS362 Distributed Systems

Multithreaded Clients

Example: A Web Browser– Setting up connection– Read incoming data– Pass to display component

– Potentially connecting to multiple load balanced replicated servers to transfer data in parallel streams

Page 15: ICS362 Distributed Systems

Multithreaded Servers

Suppose a request comes into a single threaded server.

– The request needs to be completed before any more requests can come

With multithreads, a dispatcher thread can assign a worker thread to read data from a file system

– Normally a blocking operation

Meanwhile the dispatcher thread can deal with more reads, or responses from other worker threads.

Page 16: ICS362 Distributed Systems

Multithreaded Servers

Operating System

Request from network

DispatcherThread

WorkerThread

Page 17: ICS362 Distributed Systems

Virtualisation

Of course, we are still referring to a single processor, but giving the illusion of simultaneous execution.

– ‘Resource Virtualisation’

Renewed interest in virtualisation as distributed systems become popular

– Extending (or replacing) an existing interface to mimic the behaviour of another system

– E.g. to allow software to run on a different platform.

Page 18: ICS362 Distributed Systems

Applications of virtualisation

Hardware & Low level systems software evolves quicker than application software.– Virtualisation allows older, more stable, software

to run on newer platforms

Networking is pervasive across heterogeneous computers– Virtualisation helps by letting each application run

on its own virtual machine.

Page 19: ICS362 Distributed Systems

Virtual Machine Implementation

Virtualisation mimics the behaviour of various interfaces:

– Interface between Hardware and Software, which can be invoked by any program.

– Interface between Hardware and Software, which can be invoked by privileged programs such as the OS.

– Interface of system calls offered by an OS– Interface of library calls offered by an API (essentially hiding

the OS system calls).

Page 20: ICS362 Distributed Systems

Computer Systems Interfaces

Hardware

Operating System

Library

Application

PrivilegedInstructions

System Calls

Library Functions

GeneralInstructions

Page 21: ICS362 Distributed Systems

Process Virtual Machine

Hardware

Operating System

Runtime System

Application

Virtualisation is for single processes

Page 22: ICS362 Distributed Systems

Virtual Machine Monitor

Hardware

Operating System

Virtual Machine Monitor

Application

Multiple different Operating Systems running concurrently on the same platform

Page 23: ICS362 Distributed Systems

Clients

Provide means for users to interact with remote servers– 1) Fat Client

Application specific protocol E.g. Agenda application running on a PDA,

synchronising with a remote (possibly shared) agenda.

– 2) Thin Client Application neutral protocol Client is a terminal with no need for local storage Example – X Window

Page 24: ICS362 Distributed Systems

X Window

Oldest, still widely used networked user interface.

The heart is the ‘X-kernel’ OS running on the client– Containing terminal specific device drivers for

monitor, keyboard & mouse.– Captures keyboard / mouse events and made

available through Xlib library, running elsewhere and communicating through X protocol.

Page 25: ICS362 Distributed Systems

X Window System

X Kernel

Device Drivers

Local OSLocal OS

Window Manager

Application

Xlib Xlib

Page 26: ICS362 Distributed Systems

X Window Manager

Application which dictates the “look and feel” of the display– How to display windows, what buttons etc.– Other applications adhere to these rules.

So.. Confusingly the X Kernel acts as a server where applications play the role of clients!

Page 27: ICS362 Distributed Systems

Communication Efficiency

This takes a lot of data communication!– Video stream 30 frames per second– 320x240 screen (PDA)– 24bits per pixel– = 53 Mbps!

Compression techniques are necessary

Page 28: ICS362 Distributed Systems

Increasing UI Complexity

Drag & Drop Functionality Consider a document which contains text &

images– Document contains multiple applications– Suppose we want to rotate an image

It needn’t reside on the client, but the client may require more processing power.– Or the Client needs to be fatter

Page 29: ICS362 Distributed Systems

Servers

A process implementing a specific service on behalf of a collection of clients– It waits for an incoming request– Then ensures the request is taken care of– Returns to wait for the next request

Page 30: ICS362 Distributed Systems

Iterative vs Concurrent

Iterative Server – Handles the request itself and returns response if

necessary to the user.

Concurrent Server– Passes the request to a separate thread or

another process.– The other process is responsible for responding

to the requesting client.

Page 31: ICS362 Distributed Systems

Ports

Clients send requests to an endpoint (port) Servers listen to a particular port

– E.g. FTP on TCP port 21, HTTP on TCP port 80

Or a Daemon can listen to a port and contact the specific server when requests arrive.

Page 32: ICS362 Distributed Systems

Stateless Servers

Does not keep information on the state of its clients. Can change its state without informing clients

Once an HTTP server delivers the files it forgets about the client.

– Technically the web server might log all client requests as useful information, but if the information is lost there is no damage to the service.

Page 33: ICS362 Distributed Systems

Soft State Servers

Server which promises to maintain state on behalf of the client – for a limited time– E.g. server promises to inform client about

updates, but only for a limited time.

Page 34: ICS362 Distributed Systems

Stateful Server

Server maintains persistent information on its clients.– E.g. server allows a client to keep a local copy of

a file (possibly for updating). Server then maintains a list of client/file entries.

This can improve performance, however if the server crashes it has to restore the list.

Page 35: ICS362 Distributed Systems

Session State vs Permanent State

Bear in mind that Permanent State is generally stored in a database, such as customer information, purchases etc.

Maintaining Session State is part of Stateful Server design.

Page 36: ICS362 Distributed Systems

Stateful vs Stateless

Should not affect services provided by the server.– If files have to be opened before they can be read

from or written to, then either design should mimic this behaviour.

Page 37: ICS362 Distributed Systems

Server Clusters

A collection of machines connected through a network, where each machine runs one or more servers.

Logically organised into three tiers (layers)– Switch – through which requests are routed– Application – servers where programs run– File/Database – where the data is stored

Page 38: ICS362 Distributed Systems

Server Clusters

Client Requests

LogicalSwitch

Application ServersDatabase / File Servers

First Tier Third TierSecond Tier

Page 39: ICS362 Distributed Systems

Server Clusters

Some systems combine the data and application into a single server resulting in a two tiered system architecture– E.g. Streaming Media

Page 40: ICS362 Distributed Systems

Server Clusters

Often Offer Multiple Services, with different machines running different services

– Switch needs to be able to distinguish requests and services.

– Often caused by limitations on available software & hardware, and different administrators not wanting to interfere with each other.

Leads to some overloaded servers, while others are idle.

– Possibility for code migration?

Page 41: ICS362 Distributed Systems

Code Migration

So far we have been concerned with passing data. Sometimes it is useful to pass programs

– Even while they are running

Sometimes overall system performance can be improved by moving processes from overloaded machines to lightly loaded machines.

– Load in terms of CPU queue length or CPU utilisation

Page 42: ICS362 Distributed Systems

Examples

1) A client application needs to perform many database operations– Why not move the operations closer to the data

and then only return the results?

2) A form needs to be filled in, processed and then translated into a series of database operations– Why not do the form processing on the client?

Page 43: ICS362 Distributed Systems

Code Migration

Benefits– Flexibility– Dynamically configured distributed systems– Software need not be pre-installed

Concerns– Security

Page 44: ICS362 Distributed Systems

Strong vs Weak Mobility

A process consists of 3 segments– Code Segment – set of instructions– Resource Segment – references to necessary

external resources (files, devices, printers etc.)– Execution Segment – current execution state of

the process

Page 45: ICS362 Distributed Systems

Weak Mobility

The code segment is sent– In this case the code must always begin from a

predefined point (perhaps the start)– E.g. Java Applets– Only requires that the target machine is capable

of running the code

Page 46: ICS362 Distributed Systems

Strong Mobility

The execution segment is transferred as well as the code segment.– Code can be stopped, moved to a new machine,

and restarted from where it was.– More general than weak mobility, but harder to

implement.

Page 47: ICS362 Distributed Systems

The Resource Segment

Consider; you’ve designed your website on localhost, and then upload it to a webserver– Absolute vs Relative URLs?– Changing IP addresses

Page 48: ICS362 Distributed Systems

The Resource Segment

Binding by Identifier– A precisely referenced resource, and nothing else!

Binding by Value– It doesn’t matter which resource is used, so long as the

correct value is returned.– Does it matter if I use my C++ libraries or yours?

Binding by Type– What type of device are we referring to – a printer? A

monitor?

Page 49: ICS362 Distributed Systems

Kinds of Resource

Unattached Resources– Easily moved – e.g. data files associated with the program

Fastened Resources– Possible, but relatively high cost – e.g. moving a database –

not dependent on current machine, but infeasible to move to a new environment

Fixed Resources– Cannot be moved from environment or machine – e.g. local

resources, or a local comunication end point.

Page 50: ICS362 Distributed Systems

Solutions?

With 3 kinds of process-to-resource bindings and 3 kinds of resource-to-machine bindings, there are 9 combinations to consider.

And 4 common solutions– Establish a Global System wide reference (GR)– Move the Resource (MV)– Copy the Value of the Resource (CP)– Rebind process to a locally available resource

(RB)

Page 51: ICS362 Distributed Systems

Solutions

Unattached Fastened Fixed

By Identifier MV (or GR) GR (or MV) GR

By Value CP (or MV,GR) GR (or CP) GR

By Type RB (or MV, CP) RB (or GR, CP) RB (or GR)