Chapter 2: ARCHITECTURES

 Temesgen.H
Architectures
 Distributed systems are complex.
 In order to manage their intrinsic complexity,
distributed systems should be organized
properly.
 Organization is mostly expressed in terms of its
software components.
 Different ways to look at organization of
distributed systems –two ways
 Software architecture – logical organization (of
software components and interconnections)
 System architecture – physical realization (the
instantiation of software components on real
machines)
2
Architectural style
 A architectural style is formulated in terms of
 Components,
 The way that components are connected to each other,
 The data exchanged between components, and finally

A component is a modular unit with well-defined

interfaces that is replaceable within its environment.
 A connector is a mechanism that mediates
communication, coordination, or cooperation
among components.
 It allows for the flow of control between components
 E.g., facilities for remote procedure call, message passing, or
streaming data.

3
Types of Architectural Styles
 Common architectural styles of distributed
systems
• Layered architectures
• Object-based architectures
• Data-centered architectures
• Event-based architectures

 The common technique:

 Organize your system into logically different
components, and distribute those components over
the various machines.
 Goal :
 achieving (at a reasonable level) distribution
transparency.
4
Layered architectural style
 It is hierarchical organization
 Components are organized in a layered fashion
 Components of a layer makes a down-call to components of the layer
below
 Only in exception, an up-call is made to higher level component
 Each layer exposes an interface to be used by above layers
 “Multi-level client-server”
 Each layer acts as a
 Server: service provider to layers “above”
 Client :service consumer of layer(s) “below”
 Communication protocol-stacks are a typical examples
 OSI Reference model
 TCP/IP

 Figure 2-1. The (a) layered architectural style

5
Layered Architecture: Example

6
Object-Based Architectures
 Components are objects
 Objects are easy to be replaced so long as the interface is not touched
 It is less structured and hence a relatively loose organization
 The calling object might not run on the same machine as the
called object
 Connectors are RPC and RMI

Notes:
 Layered and object-based styles are the most important styles for distributed
systems today

7
Data-Centered Architectures
 Access and update of data store is the main purpose of the
system.
 Processes communicate/exchange info primarily by reading
and modifying data in some shared repository (e.g database,
distributed file system).
 Example, web based Distributed systems are largely data
centric

 Components:
 Data store,
 Components, that interact with the store
 Connectors:
 Queries
8
Event–based Architecture
 Event based architecture supports publish-subscribe
communication
 Publisher: components that announce data to be shared
 Subscriber: components register their interest for
published data.
 Decouples sender and receiver (asynchronous
communication)
 Both parties don’t need to be up at the time of communication
 Event can be considered as “a significant change in
state”
 Components:
 Can be an instance of a class or simply a module.

 Connectors:
 Event buses

9
System Architectures
 The software components, their interactions,
and their placement leads to an instance of a
software architecture, also called a system
architecture.

 System architecture are of three types:

 Centralized - most components located on a
single machine
 Decentralized - most machines have
approximately the same functionality
 Hybrid - some combination

10
Centralized Architecture
 In the basic client-server model, processes in a distributed system
are divided into two (possibly overlapping) groups.
 Server:- is a process implementing a specific service E.g File
server
 Client:-is a process that requests a service
 Clients and servers can be on different machines
 Clients follow request/reply model with respect to using services

 Figure General interaction between a client and a server.

11
Cont...
 Communication between a client and a server can be
implemented by :
 Connectionless protocol when the underlying network is fairly
reliable like local-area networks (UDP)
 Connection-oriented protocol in WANs, (TCP)
 Connectionless communication is efficient
 Simply packages a message for the server, identifying the service
it wants, along with the necessary input data

 But, it is hard for a sender to detect if the message is

successfully received
 Failure of any sort means no reply

 Possibilities:
 Request message was lost
 Reply message was lost
 Server failed either before, during or after performing the service
12
 Cont ...
 Typical tackling to lost request in connectionless
communication:
 Re-transmission (resending request )
 Good for idempotent operations, i.e., operations that could be
repeated more than once without harm. E.g., “Return
current value of X”
 Not good for non idempotent operations like “ increase
value of x by 100”
 Because, may result in performing the operation twice
 In this case reporting an error is appropriate, than resending

 For these reason many distributed systems use connection-

oriented protocols
 Not good enough in LAN as it is slow
 However, it fits the unreliable WAN environment
 Example, Virtually all internet applications are based on TCP/IP
connections
13
Application Layering
 The client-server model has been subject to many debates and
controversies
 One issue was how to draw a clear distinction between a client and a server
 Many client-server system can be divided into three levels
 The user-interface level
 The processing level
 The data level

 The simplified organization of an Internet search engine into three different layers.
14
Logical Architecture vs. Physical Architecture
 Layer and tier are roughly equivalent terms, but
 Layer typically implies software and
 Tier is more likely to refer to hardware.
 Logical organization is not physical organization.
 Physical architecture may or may not match the logical
architecture.
 Logically separate components might reside on single machine
or on different machines
 Clients and servers could be placed on the same node, or
be distributed according to several different topologies.
 Single-Tier Architecture: dumb terminal/mainframe
configuration
 Two-Tier Architecture: client/single server configuration
 Three-Tier Architecture: each layer on separate machine
 Two-tier and three-tier are the most common

15
Two-Tiered Architecture
 Where are the three application-layers placed?
 On the client machines, or on the server machines?
 A range of possible solutions:
 Thin-Client- A client machine only implements (part of) the
user-interface level
 A server machine implementing the rest, i.e, the processing
and data levels
 Pros: easier to manage, more reliable, client machines don’t
need to be so large and powerful
 Con: perceived performance loss at client

 Fat-Client - All user interface, application processing and

some data resides at the client
 Pros: reduces work load at server;
 More scalable

 Cons: harder to manage by system admin,

 Less secure

 Other solutions in between thin-client and fat-client

16
Two-tiered Architectures

Thin client --------

Fat-client

17
Three-tiered
 The server tier in two-tiered architecture becomes more and
more distributed
 A single server is no longer adequate for modern information
systems
 This leads to three-tiered architecture
 Server may acting as a client
 Three-tiered: each of the three layers corresponds to three
separate machines.

18
Decentralized Architectures
 Placing logically different components on
different machines is called vertical distribution
 User-interface, Processing components and a data
level are on different machine
 It is similar with the concept of vertical fragmentation
in distributed database where
 Tables are split into column wise and distributed on different
machines
 The advantage of VD is that each machine can be
tailored for specific type of function

19
Cont…
 An alternative to VD is horizontal distribution
 A client or server may be physically split up into logically
equivalent parts
 Each part operates on its own share of the complete data
set,
 This results in balanced work load
 Again this one is similar with that of horizontal
fragmentation in distributed database where
 Tables are split row wise, and subset of rows distributed
onto different machines
 Peer-to-peer systems are a class of modern
architectures that support horizontal
distribution.
 The functions that need to be carried out are represented
by every process that constitute the distributed system
20
Peer-to-peer systems
 P2P systems partitions tasks or work loads
between peers
 Often, the processes that constitute the system
are all equal
 Nodes act as both client and server;
 Much of the interaction is symmetric.

 Advantages of peer-to-peer system

 Have better load balancing
 More resistant to denial-of-service attacks,
 But, harder to manage than client-server systems.

21
Overlay network
 Nodes of the P2P distributed system are connected
using overlay network
 It is network that is built on top of another network
 Nodes are formed by the processes of the network.
 Overlay networks in the P2P system:
 Define the structure between nodes in the system.
 Allow nodes to route requests to locations that may not
be known at time of request.
 The main question for peer-to-peer system is
 How to organize the processes in an overlay network
 Their organization can be:
 Unstructured P2P:
 Structured P2P:
 Hybrid P2P:
22
Unstructured P2P architecture
 Largely relying on randomized algorithm to
construct the overlay network
 Each node has a list of neighbours, which is more
or less constructed in a random way
 One challenge is how to efficiently locate a
needed data item
 The two common approaches are
 Flooding
 Random walk

23
Cont…
 Flooding:
 Issuing node u passes request for data d to all
neighbors.
 Request is ignored when receiving node had
seen it before. Otherwise, v searches locally
for d (recursively).
 Return d if found, Otherwise forward the
request to the neighbors
 However, this approach causes high
signalling traffic over the network
 May be limited by a Time-To-Live: a maximum number
of hops.

24
Cont…
 Random walk:
 Issuing node u passes request for d to
randomly chosen neighbor, v.
 If v does not have d, it forwards request to one
of its randomly chosen neighbors, and so on.

25
Structured P2P
 Nodes are organized following a specific distributed
data structure.
 The most common one is distributed hash table
(DHT)
 In such systems, each data item is uniquely
associated with a key, in turn used as an index.
 Each node is responsible to store data that are
associated with subset of these keys
 P2P system now responsible for storing (key, value) pairs
 Looking up data d with key k means routing request
to node with identifier k.
 Example
 chord

26
Hybrid Architectures
 Many distributed systems require properties from both
client-server and peer-to-peer architectures.
 So, they put together features from both centralized and
decentralized architectures, resulting in hybrid architectures.
 Some nodes are appointed special functions in a well
organized fashion
 Examples
 Edge-server systems: placed at the edge of enterprise
network
 E.g., ISPs, which act as servers to their clients, but cooperate
with other edge servers to host shared content
 Collaborative distributed systems:
 E.g., BitTorrent, which supports parallel downloading and
uploading of chunks of a file.
 First, interact with client-server system to download the torrent
file, and then operate in decentralized manner.
27
End of Chapter 2