Processes

Chapter 3

Thread Usage in Nondistributed Systems

Context switching as the result of IPC– Process: a running program (mgmt. unit of OS)– Thread: execution unit within a process– Switching:

• Processes: expensive (CPU context, MMU, TLB, …)• Threads: cheap (almost only CPU context)

Thread Design Issues

– User-level threads:+: create, destroy, switching very cheap, because without OS.-: blocking system calls (e.g. communication) would block all threads. parallelism is restricted

– Kernel-level threads: +: kernel can do context switching when blocking system calls are used.

more parallelism -: create, destroy, switching are expensive (switches to kernel mode necessary).

– Hybrid solution:Seems to be the best one.Use of user-level threads (thread package).Use of kernel-level threads so-called light-weight processes (LWPs).

Thread Implementation

Combining kernel-level lightweight processes and user-level threads.

Multithreaded Clients (1)

– Example: WWW BrowserThread 1: get textThread 2: get imageThread 3: get banner…

Better responsiveness, but user may notice parallelism.

– Multiple connections to server: If server is replicated (i.e. horizontal distribution), different connections correspond to different server instances. more parallelism

Multithreaded Servers (1)

A multithreaded server organized in a dispatcher/worker model.

Multithreaded Servers (2)

Three ways to construct a server.Finite-state machine (rather intricate):– Only one thread (that simulates multiple ones)– Loop:

• Get next message.• Do some computation if needed.• Send asynchronous message(s) to device(s) (e.g. disk) – no blocking on I/O.• Go to first step (and receive next message, which may be from disk or from

client).

Model Characteristics

Threads (multi-threaded process) Parallelism, blocking system calls

Single-threaded process No parallelism, blocking system calls

Finite-state machine Parallelism, nonblocking system calls

Clients: User InterfaceThe X-Window System

The basic organization of the X Window System

Client-Side Software for Distribution Transparency– Access transparency: IDL-based stubs– Relocation/migration transparency: rebind within middleware (stub/proxy)– Replication transparency: see below– Failure transparency: time redundancy, use of caches as in browsers, …– Concurrency/persistence transparency: rather at server or intermediate server (transaction monitors manage concurrent transactions)

Servers: General Design Issues (1)

a) Client-to-server binding using a daemon as in DCE

b) Client-to-server binding using a superserver as in UNIX (Inetd server)

3.7

listens to all relevant ports

name to port mapping

Servers: General Design Issues (2)– Handling interrupts:

– Out-of-band data: highest priority (supported in TCP)– Server is interrupted (e.g. Unix signal)

– Stateful/stateless server:– Stateful: Server holds information about clients. For example, which client

has currently opened a file for writing.Recovery is needed after crashMay enhance performance

– Stateless: no information about clients is heldNo recovery after crashesLess performance

– Sometimes servers keep information about client behaviorCookies (web server information on user’s favorite pages kept at client

side)Problem: privacy

Object Servers: Object Invocation– Object server:

– Server that invokes local objects on behalf of client requests– It does not implement a service, objects implement services– Only a room where objects live

– Issues:– Transient objects:

In-memory ones that live as long as server lives (or shorter)Create object at first invocation and delete it when no clients are bound

to it : +: minimizes resources, -: some invocations may be slowCreate all transient objects at once

– Code/data sharing:All objects are separated (e.g. different segments)

+: promotes securityBut code sharing may be very useful for persistent objects (e.g. code of

class in memory and data of different objects on disk)– Threading:

On demand / poolPer invocation (sync needed) / per object (no sync needed)

Object Adapter (1)• Object adapter (also called object wrapper):

Software (part of middleware) that implements a specific

activation policy.

• Activation policy: e.g. per object / per invocation thread.

• Main idea: objects should be kept simple and should no know how they are invoked (+: reliability).

• Object adapters are generic (independent from object)

• They communicate rather with the skeleton/stub of the object and not with the object itself.

• Objects are rather passive until invocated by adapter.

• Multiple object adapters on same machine with different activation policies possible and desirable for reasons of flexibility.

Object Adapter (2)

Organization of an object server supporting

different activation policies.

Object 1.1

Object 2.1

Object 2.2

Policy 1

Policy 2

Object Adapter (3)

The header.h file used by the adapter and any program that calls an adapter.

/* Definitions needed by caller of adapter and adapter */#define TRUE#define MAX_DATA 65536

/* Definition of general message format: adapter/client*/struct message { long source /* senders identity */ long object_id; /* identifier for the requested object */ long method_id; /* identifier for the requested method */ unsigned size; /* total bytes in list of parameters */ char **data; /* parameters as sequence of bytes */};

/* General definition of operation to be called at skeleton of object */typedef void (*METHOD_CALL)(unsigned, char* unsigned*, char**);

long register_object (METHOD_CALL call); /* register an object: from stub */void unrigester_object (long object)id); /* unrigester an object: from stub */void invoke_adapter (message *request); /* call the adapter: from deMUX*/

Object Adapter (4)

The thread.h file used by the adapter for using threads.

typedef struct thread THREAD; /* hidden definition of a thread */

thread *CREATE_THREAD (void (*body)(long tid), long thread_id);/* Create a thread by giving a pointer to a function that defines the actual *//* behavior of the thread, along with a thread identifier */

void get_msg (unsigned *size, char **data);void put_msg(THREAD *receiver, unsigned size, char **data);/* Calling get_msg blocks the thread until of a message has been put into its *//* associated buffer. Putting a message in a thread's buffer is a nonblocking *//* operation. */

Object Adapter (5)

The main part of

an adapter that

Implements a

thread-per-object policy.

Demultiplexer thread

Adapter thread

Code Migration• Migration: shift entity (process, application, …) from A to B.

• Why?:

- gain more performance: utilize lightly-loaded nodes or exploit parallelism

(e.g. search in the Web).

- enhance reliability/availability: separate faulty and well-behaved applications (rather new).

- be more flexible (see next slide)

• Load: cumulative CPU requirements of ready-to-run processes, length of ready-to-run queue, utilization, …

• Goals (related to performance): a) minimize computation time, b) minimize advent of communication (important in distributed systems)

• Examples:

- client code is better executed at server, if too many data are needed to obtain some result.

- server code is better executed at client, if too many user interactions are needed prior to actual processing.

Dynamic Client Configuration(enhances flexibility)

The principle of dynamically configuring a client to communicate to a server. The client first fetches the necessary software, and then invokes the server.

Issue: security

e.g. protocol implementation (language that server speaks)

Models for Code Migration (1)• Mobility: refers to the fact that an entity can be migrated.

• Weak mobility: only code part of entity (with some initialization data) is migrated. Entity can only begin computation at target node.

Java applets

a) code executed in same process (less secure but efficient, e.g. applets are executed in browsers)

b) code executed in different process (more secure, but resources like files are not protected, and less efficient)

• Strong mobility: both code and whole state of entity (data, CPU registers, etc.) are migrated.

the entity (e.g. process) can continue computation on target node from where it has been stopped.

example: D’Agent

remote cloning: copy of entity on target machine, but original remains at sender machine (e.g. Unix fork and continue on different machine).

• Sender-initiated: initiative of migration is taken at sender side (e.g. upload code to a server)

• Receiver-initiated: initiative of migration is taken at receiver side (e.g. Java applets).

Models for Code Migration (2)

Alternatives for code migration.

Migration and Local Resources (1)• What to do with resources after a process has been migrated?

• Resources: monitors, files, ports, libraries, … Resource bindings may change after migration (unlike code and state)

• Process-to-resource bindings:

• Binding by identifier: Process needs exactly the resource used before migration.

new resource is equal old resource, e.g. local port

• Binding by value: Process needs the value of the resource used before migration.

new resource is equivalent to old resource, e.g. programming library

• Binding by type: Process needs a resource of a specific type.

new and old resources have same type, e.g. monitors, printer, …

• Resource-to-machine bindings:

• Unattached resources: Not bound to a specific machine, e.g. unshared file

• Fastened resources: In principle independent of any machines, but de facto (e.g. enormous moving overhead) bound to a machine, e.g. Web database

• Fixed resources: bound to a specific machine, e.g. devices

Migration and Local Resources (2)

MV: move resource

GR: use global referenceCP: copy resourceRB: rebind to (local) resource

Actions to be taken with respect to the references to local resources when migrating code to another machine.

Establishing global references may always solve the problem of resource bindings after migration, however it may introduce more overhead (e.g. communication, reconfiguring owners of global reference).

Unattached Fastened Fixed

By identifier

By value

By type

MV (or GR)

CP ( or MV, GR)

RB (or GR, CP)

GR (or MV)

GR (or CP)

RB (or GR, CP)

GR

GR

RB (or GR)

Resource-to machine binding

Process-to-resource

binding

Migration in Heterogeneous Systems

The principle of maintaining a migration stack to support migration of an execution segment in a heterogeneous environment

State-of-the-art: Use of a virtual machine (e.g. Java) relies on only one language

3-15

Overview of Code Migration in D'Agents (1)

A simple example of a Tcl agent in D'Agents submitting a script to a remote machine

agent_submit: sender-initiated weak migration (with separate process at receiver side)

proc factorial n { if ($n 1) { return 1; } # fac(1) = 1 expr $n * [ factorial [expr $n – 1] ] # fac(n) = n * fac(n – 1)

}

set number … # tells which factorial to compute

set machine … # identify the target machine

agent_submit $machine –procs factorial –vars number –script {factorial $number }

agent_receive … # receive the results (left unspecified for simplicity)

Main Program

Overview of Code Migration in D'Agents (2)

An example of a Tel agent in D'Agents migrating to different machines where it executes the UNIX who command

agent_jump: sender-initiated strong migrationagent_fork: sender-initiated remote cloning

proc all_users machines { set list "" # Create an initially empty list foreach m $machines { # Consider all hosts in the set of given machines agent_jump $m # Jump to each host set users [exec who] # Execute the who command append list $users # Append the results to the list } return $list # Return the complete list when done}

set machines … # Initialize the set of machines to jump toset this_machine # Set to the host that starts the agent

# Create a migrating agent by submitting the script to this machine, from where# it will jump to all the others in $machines.

agent_submit $this_machine –procs all_users-vars machines-script { all_users $machines }

agent_receive … #receive the results (left unspecified for simplicity)

Main Program

Implementation Issues (1)

The architecture of the D'Agents system.

Agent mgmt (e.g. Ids), authentication, communication

Agent start/end, migration, agent-level communication

Implementation Issues (2)

The parts comprising the state of an agent in D'Agents.

Status Description

Global interpreter variables Variables needed by the interpreter of an agent

Global system variables Return codes, error codes, error strings, etc.

Global program variables User-defined global variables in a program

Procedure definitions Definitions of scripts to be executed by an agent

Stack of commands Stack of commands currently being executed

Stack of call framesStack of activation records, one for each running command

Software Agents in Distributed Systems

Important properties by which different types of agents can be distinguished.

Other functional properties:Interface agents: personal agents that assist users in the use of applications (usuallylearning/intelligent agents)Information agents: process information from different sources (usually mobile agents that roam the network)

PropertyCommon to all agents?

Description

Autonomous Yes Can act on its own

Reactive Yes Responds timely to changes in its environment

Proactive Yes Initiates actions that affects its environment

Communicative YesCan exchange information with users and other agents

Continuous No Has a relatively long lifespan

Mobile No Can migrate from one site to another

Adaptive No Capable of learning

Agent Technology

The general model of an agent platform (FIPA model).FIPA: Foundation for Intelligent Physical Agents

Agent communication channel

Create/delete agent, end points mapping, …

What do other agents offer?

Some Agent Applications• Searching (e.g. in the Web):

Enhances parallelism. Mobility is asset.

• E-Commerce: Walk in the network to find best offers. Pay at seller side. Issue: security.

• Network management: Performance, fault, configuration mgmt, and other functions. Cooperation is asset: networks may span over different organizational domains. Examples:

a) Construct new complex services in cooperation with other agents (on-the-fly!).

b) Solve performance reductions or fault situations by cooperating with agents in other domains.

• Computer-assisted learning: Intelligence is asset: agent should be able to adapt to student’s knowledge. Relevant in distributed systems, if (distributed) agents cooperate in doing their work.

Agent Communication Languages (1)

Examples of different message types in the FIPA ACL giving the purpose of a message, along with the description of the actual message content.

Message purpose Description Message Content

INFORM Inform that a given proposition is true Proposition

QUERY-IF Query whether a given proposition is true Proposition

QUERY-REF Query for a given object Expression

CFP Ask for a proposal Proposal specifics

PROPOSE Provide a proposal Proposal

ACCEPT-PROPOSAL Tell that a given proposal is accepted Proposal ID

REJECT-PROPOSAL Tell that a given proposal is rejected Proposal ID

REQUEST Request that an action be performed Action specification

SUBSCRIBE Subscribe to an information sourceReference to source

Agent Communication Languages (2)

A simple example of a FIPA ACL message sent between two agents using Prolog to express genealogy information.

Field Value

Purpose INFORM

Sender max@http://fanclub-beatrix.royalty-spotters.nl:7239

Receiver elke@iiop://royalty-watcher.uk:5623

Language (syntax) Prolog

Ontology (semantics) genealogy

Content female(beatrix), parent(beatrix, juliana, bernhard)