state may be replicated and that
網多書面報告
第三組
replicas reside at different locations
陳伯翰
Concurrency Hides the coordination of activities
between objects to achieve
consistency at a higher level
Distributed System:
Definition
Failure Hides failure and possible recovery of
A distributed system is a piece of software
Persistence Hides the fact that an object may be
that ensures
(partly) passivated by the system
that:
Note: Distribution transparency may be set
A collection of independent computers that
as a goal,
appears to its users as a single coherent
but achieving it is a different story.
system
Degree of Transparency
Two aspects: (1) independent computers
Observation: Aiming at full distribution
and
transparency
objects
(2) single system middleware.
may be too much:
Local OS Local OS Local OS
_
_
Users may be located in different continents;
Machine A Machine B Machine C
distribution
Network
is apparent and not something you want
Distributed applications
to hide
Middleware service
_
Completely hiding failures of networks and
Goals of Distributed
Systems
nodes
Connecting resources and users
– You cannot distinguish a slow computer
Distribution transparency
from
Openness
a failing one
Scalability
– You can never be sure that a server
Distribution Transparency
actually
Transparency Description
performed an operation before a crash
_
_
_
_
Access Hides differences in data
is (theoretically and practically) impossible
_
Full transparency will cost performance,
representation and invocation
exposing
mechanisms
distribution of the system
Location Hides where an object resides
– Keeping Web caches exactly up-to-date
Migration Hides from an object the ability of a
with
system to change that object’s location
the master copy
Relocation Hides from a client the ability of a
– Immediately flushing write operations to
system to change the location of an
disk
object to which the client is bound
for fault tolerance
Replication Hides the fact that an object or its
Openness of Distributed
Systems
Open distributed system: Be able to
interact with
services from other open systems,
irrespective of the
underlying environment:
_
Systems should conformto well-defined
_
Allow (dynamic) setting of caching policies,
preferably
per cachable item
_
Support different levels of trust for mobile
code
_
Provide adjustable QoS parameters per
data stream
_
Offer different encryption algorithms
Systems should support portability of
Scale in Distributed
Systems
applications
Observation: Many developers of modern
Systems should easily interoperate
distributed
Achieving openness: At least make the
system easily use the adjective “scalable”
distributed
without making
system independent from heterogeneity of
clear why their system actually scales.
the underlying
Scalability: At least three components:
interfaces
_
_
environment:
_
_
_
Hardware
Platforms
Languages
Policies versus Mechanisms
_
Number of users and/or processes
(size scalability)
_
Maximum distance between nodes
(geographical scalability)
_
Number of administrative domains
Implementing openness: Requires support
(administrative scalability)
for different
Most systems account only, to a certain
policies specified by applications and users:
extent, for
What level of consistency do we require for
size scalability. The (non)solution: powerful
clientcached
servers.
data?
Today, the challenge lies in geographical
Which operations do we allow downloaded
and administrative
code
scalability.
to perform?
Techniques for Scaling
Which QoS requirements do we adjust in the
Distribution: Partition data and
face
computations across
of varying bandwidth?
multiple machines:
_
_
_
_
What level of secrecy do we require for
communication?
Implementing openness: Ideally, a
_
_
_
Move computations to clients (Java applets)
Decentralized naming services (DNS)
Decentralized information systems (WWW)
distributed system
Replication: Make copies of data available
provides only mechanisms:
at different
machines:
_
Networks of Computers
tolerance)
Multiprocessors and
Multicomputers
Replicated databases
Distinguishing features:
_
_
_
_
Replicated file servers (mainly for fault
Mirrored Web sites
Large-scale distributed shared memory
_
_
Private versus shared memory
Bus versus switched interconnection
systems
Networks of Computers
Caching: Allow client processes to access
High degree of node heterogeneity:
local copies:
_
_
_
High-performance parallel systems
Web caches (browser/Web proxy)
(multiprocessors
File caching (at server and client)
as well as multicomputers)
Scaling – The Problem
Observation: Applying scaling techniques is
_
_
High-end PCs and workstations (servers)
Simple network computers (offer users only
easy,
network
except for one thing:
access)
Havingmultiple copies (cached or
replicated),
leads to inconsistencies: modifying one
copy
makes that copy different from the rest.
Always keeping copies consistent and in a
general way requires global
_
_
Mobile computers (palmtops, laptops)
Multimedia workstations
High degree of network heterogeneity:
_
_
_
_
Local-area gigabit networks
Wireless connections
Long-haul, high-latency POTS connections
Wide-area switched megabit connections
synchronization
Observation: Ideally, a distributed system
on each modification.
hides these
Global synchronization precludes
differences
large-scale
Distributed Systems:
Software Concepts
solutions.
Observation: If we can tolerate
inconsistencies, we
may reduce the need for global
_
_
_
Distributed operating system
Network operating system
Middleware
synchronization.
System Description Main goal
Observation: Tolerating inconsistencies is
DOS Tightly-coupled OS for
application
Distributed Operating
System
dependent.
Distributed Systems:
Hardware Concepts
Some characteristics:
Multiprocessors
other computers
_
_
Multicomputers
_
_
OS on each computer knows about the
OS on different computers generally the
same
system with
Services are generally (transparently)
networking facilities
_
distributed
_
Computers work independently (i.e., they
across computers
may even
Kernel Kernel Kernel
have different operating systems)
Machine A Machine B Machine C
_
Services are tied to individual nodes (ftp,
Network
telnet,
Distributed operating system services
WWW)
Distributed applications
_
Highly file oriented (basically, processors
Multicomputer Operating
System
share
Harder than traditional (multiprocessor)
OS: Because
Distributed System
(Middleware)
memory is not shared, emphasis shifts to
Some characteristics:
processor
only files)
_
OS on each computer need not know about
communication by message passing:
the
Often no simple global communication:
other computers
_
– Only bus-based multicomputers provide
_
OS on different computers need not
hardware
generally be
broadcasting
the same
– Efficient broadcasting may require network
_
Services are generally (transparently)
interface
distributed
programming techniques
across computers
No simple systemwide synchronization
Need for Middleware
mechanisms
Motivation: Too many networked
Virtual (distributed) shared memory requires
applications were
OS
hard or difficult to integrate:
_
_
to maintain global memory map in software
_
Inherent distributed resource management:
_
_
Departments are running different NOSs
Integration and interoperability only at level
no central
of primitive
point where allocation decisions can be
NOS services
made
_
Need for federated information systems:
Practice: Only very few truly multicomputer
– Combining different databases, but
operating
providing
systems exist (example: Amoeba)
a single view to applications
Network Operating System
– Setting up enterprise-wide Internet
Some characteristics:
services,
Each computer has its own operating
making use of existing information systems
_
– Allow transactions across different
databases
– Allow extensibility for future services (e.g.,
_
_
_
Authentication and authorization services
Simple encryption services
Auditing service
teleworking, collaborative applications)
Comparison of DOS, NOS,
and Middleware
Constraint: use the existing operating
1: Degree of transparency
systems,
2: Same operating system on each node?
and treat them as the underlying
3: Number of copies of the operating system
environment (they
4: Basis for communication
provided the basic functionality anyway)
5: How are resources managed?
Middleware Services (1/2)
6: Is the system easy to scale?
Communication services: Abandon
7: How open is the system?
primitive socketbased
Item Distributed OS Network Middlemultiproc.
message passing in favor of:
multicomp. OS ware DS
Procedure calls across networks
1 Very High High Low High
Remote-object method invocation
2 Yes Yes No No
Message-queuing systems
31NNN
Advanced communication streams
4 Shared
Event notification service
memory
Information system services: Services
Messages Files Model
that help manage
specific
data in a distributed system:
5 Global,
Large-scale, systemwide naming services
central
Advanced directory services (search
Global,
engines)
distributed
Location services for tracking mobile objects
Per node Per node
Persistent storage facilities
6 No Moderately Yes Varies
Data caching and replication
7 Closed Closed Open Open
Middleware Services (2/2)
Client–Server Model
mobility,
_
_
_
_
_
_
_
_
_
_
_
Control services: Services giving
applications control
over when, where, and how they access
_
_
_
Basic model
Application layering
Client–Server architectures
Distributed transaction processing
Basic Client–Server Model
(1/2)
Code migration
Characteristics:
data:
_
_
Security services: Services for secure
processing
and communication:
_
There are processes offering services
(servers)
_
There are processes that use services
(clients)
_
Clients and servers can be distributed
across different
machines
_
Clients follow request/reply model with
respect to
using services
Basic Client–Server Model
(2/2)
Servers: Generally provide services related
to a shared
resource:
_
Servers for file systems, databases,
implementation
repositories, etc.
_
Servers for shared, linked documents (Web,
Lotus
Notes)
_
_
Servers for shared applications
Servers for shared distributed objects
Clients: Allow remote service access:
_
Programming interface transforming client’s
local
service calls to request/reply messages
_
Devices with (relatively simple) digital
components
(barcode readers, teller machines,
hand-held phones)
_
Computers providing independent user
interfaces
for specific services
_
Computers providing an integrated user
interface
for related services (compound documents)