Distributed Systems
Delhi Technological University
Introduction
Instructor: Divyashikha Sethia
[email protected]Sites.google.com/site/divyashikhasethia
Divyashikha Sethia (DTU) 1
Course Objective
Understanding of the issues involved in distributed computer systems and to
investigate the fundamental characteristics of distributed computing systems,
including their models, architectures and designs that exploit rapidly evolving
technology.
Divyashikha Sethia (DTU) 2
Course Introduction
Prerequisites:
- Operating System
- Computer Networks
Divyashikha Sethia (DTU) 3
Course Outline
•Unit1:
- Introduction: Definitions and motivation, characteristics, issues in design
of DS
- Architecture: Distributed System Models
Unit2: Inter-process communication: Networking concepts, Sockets and
•
streams, remote procedure calls, remote method invocation
Unit3: Time Synchronization: Logical clocks, vector clocks, direct
•
dependency clocks, matrix clocks
Unit 4: Resource Allocation Synchronization: Distributed Shared Memory,
•
Process Scheduling, Load Balancing & Load Sharing, Mutual Exclusion,
Election algorithms.
Divyashikha Sethia (DTU) 4
Course Outline
Unit 5: Replication and Consistency
Need for replication, consistency models and protocols
Unit 6: Fault Tolerance
–Basic concepts, reliability and availability, recovery
Unit7: File Systems
–Distributed file services, example file systems
Divyashikha Sethia (DTU) 5
Text
Text Books:
Distributed Systems: Principles and Paradigms, 2nd Ed., Andrew S. Taenbaum and
Maarten Van Steen, Prentice Hall, 2007.
References:
• Distributed Operating Systems: Concepts and Design, P.K.Sihna, PHI, 2007.
•Distributed Operating Systems and Algorithms. R. Chow, T. Johnson. Addison-Wesley
Publishing Company, 1997. ISBN 0-201-49838-3.
•Distributed Systems: Concepts and Design, 4th Ed by Coulouris, G, Dollimore, J., and
Kindberg, T., Addison-Wesley, 2006.
Divyashikha Sethia (DTU) 6
Evaluation
Scheme in Syllabus
7
Divyashikha Sethia (DTU)
Definition of a Distributed System (1)
• A distributed system is:
A collection of independent computers that appears to its users as a
single coherent system.
• Utilizes the power of :
- Advancement in computing power of microprocessor
- High speed network communication between computers
8
Divyashikha Sethia (DTU)
Some uses..
Pool of processors dynamically assigned to users to execute a job in the best
•
manner possible
•Large bank with branch offices over world, each office with:
- Master computer to store local accounts and handle local transactions
- Ability to communicate with other branch offices and central computer
- Transactions can be made without regard to where customer account is
located.
9
Divyashikha Sethia (DTU)
Definition of a Distributed System (2)
•Middleware to support heterogeneous computers and networks and offering a single
system overview.
•Middleware layer extends over multiple machines.
•Applications distributed over computers.
Divyashikha Sethia (DTU) 10
Goals
Resource Availability
•
Distribution Transparency
•
Openness
•
Scalability
•
Divyashikha Sethia (DTU) 11
Resource Availability
• Access to Remote resources by all users eg: printers, computers, storage
• Helps collaborate and exchange information
- exchanging files, mail. documents, audio, and video
- electronic commerce
• Threat to Security
Divyashikha Sethia (DTU) 12
Transparency in a Distributed System
Different forms of transparency in a distributed system.
Divyashikha Sethia (DTU) 13
Openness
• Offers services according to standard rules in the form of interfaces
described by Interface Definition Languages (IDL)
• Allows an arbitrary process to talk to another process
• Extensible: add new components or replace old ones
Divyashikha Sethia (DTU) 14
Scalabity
• Add more users and resources
• Geographical scalability wrt to distribution of users resources across
different places
• Administrative scalability
Divyashikha Sethia (DTU) 15
Scalability Problems when users increase
Concept Example
Centralized services A single server for all users
Centralized data A single on-line telephone book
Centralized algorithms Doing routing based on complete information
Examples of scalability limitations.
Divyashikha Sethia (DTU) 16
Centralized algorithm
• Collect information about load on machines and lines and compute
optimal routes for communication.
• Collecting and Transporting information to centralized server and then
distributing information after analysis overloads network.
•Distributed algorithms characteristics:
-No machine with complete information about system state
-Machine decision based on local information only
-Algorithm not ruined by failure of a single machine
-Global clock
Divyashikha Sethia (DTU) 17
Geographic scalability
• Client server architecture which requires synchronous communication
works well on a LAN.
- client, blocks until a reply is sent back
• Communication delays in WAN for response can be longer
• Unreliability of communication over WAN
- local-area networks generally provide highly reliable communication
facilities based on broadcasting
Divyashikha Sethia (DTU) 18
Administrative scalability
Scaling across multiple independent administrative domains resolving
•
conflicting policies for resource usage, management, security
Security issues across domains
•
Divyashikha Sethia (DTU) 19
Scaling Techniques
1. Hiding Communication latencies
2. Distribution
3. Replication
Divyashikha Sethia (DTU) 20
Scaling Techniques – Hiding
Communication latencies
Avoid waiting for responses to remote service requests (asynchronous
•
communication)
Reply invokes interrupt to complete the previous request
•
Alternative to start a new thread of control to perform request – only single
•
thread blocked for reply while others are in process
•Not suitable for interactive applications requiring quick response time.
eg: accessing databases using forms
•Solution: interactive form filling to be done at client side rather than server side
Divyashikha Sethia (DTU) 21
Hiding Communication latencies(2)
1.4
The difference between letting:
a) a server or b) a client check forms as they are being filled
Divyashikha Sethia (DTU) 22
Scaling Techniques – Hiding
Communication latencies (3)
Server may check for syntactic errors before accepting an entry
•
Better solution: code for filling in form, and possibly checking entries, to the
•
client, and have the client return a completed form
Eg: Web in the form of Java applets and Javascript
Divyashikha Sethia (DTU) 23
Scaling Techniques –Distribution (1)
• Splitting components into smaller parts and spreading them
across system
• Eg: Domain Name System name space is hierarchically
organized as tree of domains, which are divided into
nonoverlapping zones,
Divyashikha Sethia (DTU) 24
Scaling Techniques –Distribution (2)
1.5
An example of dividing the DNS name space into zones and
distributing the naming service
Resolving “nl. vu.cs.flits”: Traverse Z3,Z2,Z1
25
Divyashikha Sethia (DTU)
Scaling Techniques –Distribution (3)
World Wide Web:
Web is physically distributed across a large number of servers, each handling a
•
number of Web documents.
The name of the server handling a document is encoded into that document's
•
URL.
It is only because of this distribution of documents that the Web has been
•
capable of scaling to its current size
26
Divyashikha Sethia (DTU)
Scaling Techniques - Replication
Replicate components across distributed system – increasing availability
•
and balancing load between components for better performance
eg: Geographically close copy can hide communication latency
Caching is a form of replication – copy of resource in the proximity of the
•
client accessing the resource
- Difference from replication: on-demand rather than planned
Leads to consistency problems
•
- update must be immediately be propagated to all other copies
- Two concurrent updates must be made in the same order.
Requires global synchronization
•
Divyashikha Sethia (DTU) 27
Scalability Conclusion
• Size scalability is the least problematic
• Geographical scalability is a much tougher problem
•Administrative scalability seems to be the most difficult one, because need to
solve nontechnical problems
- peer-to-peer technology demonstrates what can be achieved if end users
simply take over control
Divyashikha Sethia (DTU) 28
Advantages over Centralized systems
Resource utilization provides better performance for the price.
•
Provide more computing power compared to centralized mainframe
•
Computer supported cooperative work involving spatially separated
•
machines
Higher reliability
•
Incremental growth in computing power
•
Divyashikha Sethia (DTU) 29
Advantages over PC
Divyashikha Sethia (DTU) 30
Disadvantages of DS
Divyashikha Sethia (DTU) 31
Classification
SISD: Single Instruction and Single stream (traditional computer)
•
SIMD: Single Instruction and Multiple Stream (Array of processors with one
•
instruction to be processed on multiple data handled by multiple data units in
parallel) eg: supercomputers
MIMD: Multiple instruction and multiple streams – group of independent
•
computers each with its program counter, program and data. (Distributed
Systems)
Divyashikha Sethia (DTU) 32
MIMD Classification
Less delay high High delay low
data rate - Parallel data rate -
Distributed
Wires from
Single medium
machine to
connecting all
machine
machines
33
Hardware Concepts
Different basic organizations and memories in distributed computer systems
Divyashikha Sethia (DTU) 34
Multiprocessors (1)
A bus-based multiprocessor.
•Bus has address lines, data lines and control lines in parallel.
•Single coherent memory ( memory written is readable to others coherently
after a minute delay)
•Cache memory to increase performance
•Write through cache for uniformity of caches (i.e word written to cache is
written to memory as well so that new CPU accessing it gets updated value)
•Limited can have at most 64 CPUs
35
Divyashikha Sethia (DTU)
Multiprocessors (3)
• Write through cache for uniformity of caches (i.e word written to cache is
written to memory as well so that new CPU accessing it gets updated value)
• Snoopy cache – cache constantly monitors the bus for any write occurring
to a memory address that is in the cache and updates it.
36
Divyashikha Sethia (DTU)
Multiprocessors (2)
1.8
a)A crossbar switch
b)An omega switching network
37
Divyashikha Sethia (DTU)
Multiprocessor
•Crossbar Switch:
- Every intersection between CPU and memory is a physical switch that can be
opened or closed
-Many CPUs can access memory at the same time provided memory locations
are different
-Disadvantage – n CPU, n memory require switches
•Omega network:
-Contains 2x2 switch
-n CPU n memory location require
-Delay due to switching stages
Conclusion: tightly coupled, shared memory multiprocessor is difficult and
-
expensive
38
Divyashikha Sethia (DTU)
Bus Based Multicomputer
•Each computer has own memory and communicate across bus
•Since traffic is less compared to multiprocessor bus can be a lower speed LAN
39
Divyashikha Sethia (DTU)
Homogeneous Multicomputer Systems
1-9
a) Grid b) Hypercube
40
Divyashikha Sethia (DTU)
Types of Distributed Systems
• Distributed Computing System
-used for high-performance computing tasks
Distributed Information System –
•
- interoperatabilty & communication of network applications in organization
Distributed Pervasive Systems
•
41
Divyashikha Sethia (DTU)
Distributed Computing System(1)
i) Cluster computing
- underlying hardware consists of similar workstations closely connected by
means of a highspeed LAN
- each node runs same OS
ii) Grid computing
-consists of distributed systems often constructed as federation
of computer systems, where each system may fall under different
Administrative domain, and may have different hardware, software, and
deployed network technology.
42
Divyashikha Sethia (DTU)
Cluster Computing
Build supercomputer by simply hooking up collection of simple computers in
•
high-speed network.
Cluster computing is used for parallel programming in which single (compute
•
intensive) program is run in parallel on multiple machines
43
Divyashikha Sethia (DTU)
Linux Cluster
Each cluster consists of collection of compute nodes controlled and accessed
•
by master node that:
- allocates nodes to particular parallel program
- maintains a batch queue of submitted jobs
- provides interface for users of system.
Master runs middleware for execution of programs and management of cluster,
•
while compute nodes only need a standard OS.
Middleware has libraries for executing parallel programs and effectively
•
provides advanced message-based communication facilities, handling faulty
processes, security, etc.
44
Divyashikha Sethia (DTU)
Grid Computing
Computing systems have a high degree of heterogeneity
•
Resources from different organizations are brought together to allow the
•
collaboration
Software provides access to resources from different administrative domains,
•
users and applications that belong to a specific virtual organization
45
Divyashikha Sethia (DTU)
Layered architecture for
grid computing
Fabric layer
•
- Provides interfaces to local resources at a specific site.
- Allows sharing of resources within a virtual organization
Connectivity Layer
•
- Communication protocol for grid transactions for usage of multiple resources
- Security protocols to authenticate users and resources
46
Divyashikha Sethia (DTU)
Layered architecture for
grid computing
Middleware
Layer
•Resource Layer
- Responsible for managing a single resource
- Functions for obtaining configuration information on a specific resource, perform
specific operations such as creating a process or reading data
•Collective layer
- Access multiple resources
-Services for resource discovery, allocation & scheduling tasks onto multiple resources,
data replication
•Application layer
- Applications that operate within virtual organization &which use grid computing
47
Divyashikha Sethia (DTU)
Difference between Cluster and
Grid computing
Characteristics of Grid Computing
•Loosely coupled (Decentralization)
Diversity and Dynamism
Distributed Job Management & scheduling
Characteristics of Cluster computing
•Tightly coupled systems
Single system image
Centralized Job management & scheduling system
48
Divyashikha Sethia (DTU)
Distributed vs Cloud Computing
Distributed computing/distributed system involve breaking up a
•
problem which can be solved by a group of computers working at the
same time.
Cloud computing usually refers to providing a service via the internet.
•
That service can be pretty much anything, from business software that
is accessed via the web to off-site storage or computing resources.
Divyashikha Sethia (DTU) 49
Cloud vs Grid Computing
Grid computing:
•
-Used in environments where users make few but large allocation requests.
Eg: lab may have 1000 node cluster and users make allocations for all 1000,
or 500, or 200, etc.
- only a few of these allocations can be serviced at a time and others need to
be scheduled for when resources are released
-results in sophisticated batch job scheduling algorithms of parallel
computations.
•Cloud computing:
- lots of small allocation requests. The Amazon EC2 accounts are limited to
20 servers each by default and lots and lots of users allocate up to 20 servers
out of the pool of many thousands of servers at Amazon.
-Allocations are real-time and there is no provision for queuing allocations
until someone else releases resources
50
Divyashikha Sethia (DTU)
Distributed Information System
Networked applications can be integrated to form enterprise-wide
information system
•Types:
i) Transaction Processing Systems (Database – self study)
-server running application (often including a database) and making it
available to remote programs, called clients.
- clients wrap a number of requests, possibly for different servers,
into a single larger request and have it executed as a distributed transaction
ii) Enterprise Application Integration
-integration should also take place by letting applications communicate
directly with each other
51
Divyashikha Sethia (DTU)
Enterprise Application Integration
Integrate applications independent from their databases &communicate directly
through Middleware in the form of :
i) Remote procedure calls (RPC):
- Application component can effectively send request to another application
component by doing local procedure call, which results in request sent as
message to callee.
-Result sent back and returned to application as result of procedure call 52
Divyashikha Sethia (DTU)
Enterprise Application Integration
ii) Remote Method Invocation ( RMI)
-allow calls to remote objects instead of application
•Disadvantage: caller and callee need to be up and running at time of
communication
53
Divyashikha Sethia (DTU)
Message Oriented middleware (MOM)
Disadvantage of RPC & RPI:
•
- caller and callee need to be up and running at time of communication
- need to know exactly how to refer to each other
•Message Oriented Middleware (MOM)
-applications send messages to logical contact points, described by means of
subject and specifying type of message.
-Middleware takes care those messages are delivered to applications.
- Publish/subscribe systems form an important and expanding
class of distributed systems
54
Divyashikha Sethia (DTU)
Distributed Pervasive System
Small, battery-powered, mobile, devices with only a wireless connection hence
not very stable
Configured by owners, but should automatically discover environment
Requirements:
1.Embrace contextual changes.
• Device aware of environmental changes
• Network disconnectivity :automatically connect to another network/
appropriate actions
2. Encourage ad hoc composition.
• Easy to configuration of applications
3. Recognize sharing as the default.
• Easily read, store, manage, and share information
• Efficient discovery of services
55
Divyashikha Sethia (DTU)
Example of pervasive systems
Home Systems:
• PCs, gaming devices, smart phones, surveillance cameras, clocks, kitchen
appliances hooked to distributed system
• Self configuring and self managing
• Universal Plug and Play (UPnP) standards by which devices
automatically obtain IP addresses, can discover each other
Electronic Health Care System
• Devices to monitor well-being of individuals and to automatically
contact physicians when needed and avoid hospitalization
• Various sensors organized in a (preferably wireless) body-area network
(BAN) minimally hindering a person, operating while person is on the
move
56
Divyashikha Sethia (DTU)
Pervasive Electronic Health Care
Monitoring a person in a pervasive electronic health care system:
(a) Local hub collects data as needed and offloads it to external storage
device time to time.
(b) Continuous wireless connection hooked up to external storage
57
Divyashikha Sethia (DTU)
Sensor Networks
Used for processing information and form the basis for many medium-scale
distributed systems
Consists of tens to hundreds or thousands of relatively small nodes, each
equipped with a sensing device, battery operated
Distributed Database
58
Divyashikha Sethia (DTU)
Sensor Network database
Storing and processing data:
(a) only at the operator's site or (b) only at the sensors. 59
Divyashikha Sethia (DTU)
Sensor Network database
a) Sensors do not cooperate but simply send their data to centralized database
located at operator's site
• may waste network resources and energy
b) Forward queries to relevant sensors and let each compute an answer,
requiring operator to sensibly aggregate returned answers
• wasteful as it discards aggregation capabilities of sensors which would
allow much less data to be returned to operator
60
Divyashikha Sethia (DTU)
In-network processing
Forward query to all sensor nodes along a tree encompassing all nodes and
subsequently aggregate results as they are propagated back to root, where
initiator is located
Aggregation will take place where two or more branches of the tree come to
together
TinyDB, implements a declarative(database) interface to wireless sensor
networks
• Use any tree-based routing algorithm.
• Intermediate node collect and aggregate tree results from its children, along
with its own findings, and send that toward root
61
Divyashikha Sethia (DTU)
Summary
Distributed systems consist of autonomous computers that work together to
give the appearance of a single coherent system.
Advantage:
• Makes easier to integrate different applications running on different
computers into a single system.
Issues : Resource Availability, Distribution Transparency, Openness, Scalability
Types of Systems:
• Distributed Computing System, Distributed Information System,
Distributed Pervasive Systems
62
Divyashikha Sethia (DTU)
Resources
•Distributed Systems - Tanenbaum
•http://www.jatit.org/research/introduction_grid_computing.htm
•http://www.thepicky.com/tech/difference-cloud-computing-vs-grid-
computing/
63
Divyashikha Sethia (DTU)
