Distributed Systems Architectures

• A distributed systems comprises of software components, processing nodes, and networks
• Software are under designer's control
• System hardware can be specified within the design or taken as-is

System Organization

• The software can become complex, especially in large systems
• Distinguishes logical organization of software and physical organization
• Software architecture deals with software components are organized and how they communicate
• Typical software architectures include layered, object-oriented, data-centred, service-oriented, and event-based architectures
• Once instantiated and placed on machines, it becomes a system architecture
• Architectures are distinguished by the roles of the communicating processes

Good Architecture

• Allows splitting functionality thus structuring the application
• There is no single best architecture, this depends application's requirements and the environment

Client-Server Architecture

• The most common and widely used model for communication between processes
• One process takes on the role of a server, while all other processes take on the roles of clients
• Server provides a service while clients consume it
• Client sends request to a server, the request is processed at the server and a reply is returned
• Typical application decomposes into three logical parts: the interface part, the application logic part, and the data part
• Implementations of the architecture vary in parts separated over the roles

Client Implementations

• A minimal user interface layer includes everything else to the server
• A fat client implementation includes the user interface and application logic in the client
• Only rely on the server to store and provide access to data
• Other implementations split the interface or application logic parts over the clients and server

Vertical Distribution (Multi-Tier)

• Extension of the client-server architecture that distributes the traditional server functionality over multiple servers
• Client request is sent to the first server
• During processing of the request this server will request the services of the next server, who will do the same, until the final server is reached
• The various servers become clients of each other
• Each server is responsible for a different step in the fulfillment of the original client request

Splitting Server Functionality

• Beneficial to a system's scalability and flexibility
• Improved scalability because the processing load on each individual server is reduced
• Allows the internal functionality of each server to be modified as long as the interfaces provided remain the same

Horizontal Distribution

• Splits up a server’s functionality over computers
• Horizontal involves replicating a server’s functionality over multiple computers
• Each server machine contains a complete copy of all hosted Web pages
• Client requests are passed on to the servers in a round robin fashion
• Architecture is used to improve scalability (reduce the load on server) and reliability (providing redundancy)
• It is possible to combine the vertical and horizontal distribution models

Peer to Peer

• Assumes different processes take on different roles in the communication architecture
• Takes the opposite approach and each plays the same role and are peers
• Each process acts as both client and server, both sending/receiving requests
• All processes provide the same logical services

Examples of P2P

• File-sharing applications
• Users start a program to search/download files from others and handles search/download requests from users
• Due to number nodes it becomes impossible to track all nodes & info they offer
• Reduce the problem nodes form an overlay network
• Nodes form a virtual network among themselves and have direct knowledge of other nodes
• Message to an arbitrary other node must first locate by propagating a request along overlay links

Types of overlay networks

• Distinction based on how they are built/maintained
• Node in the network maintains list of neighbours (partial view of network)
• Unstructured overlays the network often resembles a random graph
• Membership management is random, a node’s partial view consists from a random list of other nodes
• To keep network connected as nodes join/leave all nodes periodically their partial views with neighbours, creating a new list

Structured Overlays

• Choice of a node's neighbor is determined according to a specific structure
• Distributed hash table nodes work to implement
• Each node responsible for storing data associated with identifiers
• Joining a network, a node is assigned an identifier/locates node in the range and part of the space is taken

Hybrid

• Architectures can be designed by combining the previously described architectures
• Examples described are called hybrid architectures

Supepeer Networks

• Few superpeers form P2P network
• Regular peers are clients to a superpeer
• Architecture maintains a system with some of P2P system
• System simplifies by superpeers an indexes regular peers/brokers

Collaborative Distribution Systems

• Peers typically support to deliver content In a P2P and use client, server setup network
• For example BitTorren, nodes requesting to download a file from a server and contact the location of the tracker
• Tracker then tells the nodes the locations of the other nodes
• Nodes then must offer chunk to other nodes/ register to the tracker so other nodes can find them

Edge-Server Networks

• Servers at the "edge" such as ISP close to enterprise networks
• Client/home users then access the nearby edge servers
• Suited for large-scale distribution networks such as Akamai

Processes and Server Architecture

• Key property is distributed systems consist of Processes
• Communication is between processes on a single computer

Thread of Control

• Two Models
  • Process Model: Thread of Control with address space
  • Thread Model: Threads of control share a single address space
• Memory Access
  • Threads share all their memory, processes are prevented
  • Some systems provide only a process model and some a thread model
• Systems provide to be a Process while contain threads
• We can refer it as a Process

Server Process

Typically receives requests for work from various clients for responses

• A server should not refuse work for one client (E.G: Invoke Blocking) a Server as Single-Threaded
  • Alternatives:
    • Using many Threads
    • Design and Build a State Machine - Non Blocking Call
• Server Issue design Is whether they store the clients or NOT

The Stateful Model

• Stores a Persistent Information about a client
• Leads to good performance since clients need tell states
  • Flipside the need to keep track of client and ensure recovering after a crash
• The Stateless Model
  • Keep NO Information about Clients
  • Does not worries about state after a crash

Client - Server

• Separate virtual machines are hosted on single machine
• Consolidation on single Machine and Isolation
  • The Machines in Cluser assigned roles that included a Switch(To Receive and Route to Servers)
  • Typically The Server will store address and cycle

Code Mobility

• Mobility : In cases it can change the Location where Processes are being executed
  • This can be to Unloaded Servers within Clusters
  • Two types
    • Weak- Only Code is transferred and restarted at destination
    • Strong- Execution Code and Context are transferred at Execution from left off before moving

Communication Protocol

• Synchronization and sharing Data
  • Processes Synchronization:
    • Coordination
      • Finding if a Process lives
      • Determining Task a Process
      • Acquire Resources
      • Request tasks
  • Processes Shared Data
    • Cooperative Working
      • Sending data as tasks
      • Returning Cal

Ways to Communicate

• Shared Memory or by Message Passing
  • Processes must have Shared Memory
    • Threads: share memory file and resource
      • Communicate by sending each other Messages
• Message Passing has
  • (IPC) Inter Processes Communication available in Underlying Operating System
  • Examples : Pipes and Sockets

Communication In Distributed Systems

• Main Difference run computers are not access Direct Message Passing
  • Similar using message Passing Non-Distributed System
  • Message Passing Main operations involved SEND() and RECEIVE()
    • Connectionless
      • Very Simple
      • Many variations
    • Connection-Oriented
      • Sender/Receiver
      • Create before Send() and Receive()
      • Many consider dealing processes communication using Send(), Issues include Representation Communication Models
  • A number of Alternative can take place

First Distinction Model

• Data-Oriented Communication and Control Communication - Data Serves exa-Data - Although trigger and action no transfer
• Control-Orient Communication
  • Every data explicit
  • Type Communication:
    • Space and shared Memory
• Communication Operation
  • Syn or Asyn
  • SYN send messages blocked: Send/Recv Replies.
    • (Ex: Blocking)
  • ASYN: Execution after message without waiting, and possibly no answer

Alter Buffering Communication Model

- Message will be stored if message not able to pick

-UnBuffered: will be lost

Communication can be Tranisent or Persistent

 - Communication that's a Messaged will be delievered if the Reciever active
      - No intertested Message will be dropped

• Persistent Communication A message will be stored - Message Will Delievered

Degree of Reliability

• Reliable communication errors are Transparency
  • Communication, a message that's a sent at Destination
  • Destination exists Receive (With UNRELIABLE that's messages get Lost)
• Ordering messages
  • Ensure messages received
  • System guaranteed messages

Communication Abstractions

• Assumed messages can't easily write and send and Recv messages Abstraction to make it easier
• Way to make Programming Language easier
• Providing level programming assembly
• Level Do Higher provide to Do Communication
• attempt to completely the fact that communication is taking place

Used for Core Foundation in Middleware

• By Using Abstractions
• Communication not to know communicate Operating Primitives

Messaged -Orientes Communication

• Message Not Attempt fact to Hide place make Passing easier
• Message Oriented based Model on Process sending Messages
  • Communication Properties
    • Synchronization or Asyn
• Message Oriented provides By Message- Oriented Middleware
• MOM also provides abstractions as they are on Hardware Platforms, Abstract
• MOMS allows programmers use Message- Orientated
  • The Tasks RPC frameworks are PackingData
• XML-RPC- The Tasks RPC frameworks are PackingData
• Conversion different format of By The Recv

RMI Abstraction

• Must Specify to Server on the CALL on (Binding) The Server
• It's Server Keep Account The states belonging in different Clients and Invocation
• Server Metaphor for object metaphor
• Remote Object : once a client is in the network it longer to worry where object IS
• Also Can be object model Passed and Results IN RMI
• To relieve Problems associated
• Remote objects form a distributed paradigm lecture
• THE DANGER TRANSPARENCY
• The Allusion The Proc is bad FOR PCS can FAIL

Group Communication

• Departure the point of Style
• A Process Can Send a Group to other processes
• Broadcast (when a single is sent for every one) and mulicast (To tell Everyone the Printer is on FIRE)
• Implementation and using in Group , this includes reability
• Implementation Can Lead to Complication
• A single of reciepients Not Recv, Again OR the process not rec
• Different recipients
  • IP multi-cast Long Existed
  • Implementation routers

Increase Impairment Communicate is Gossip-Based Communication

• Ruined The Spread
  • (Data That receive some new will cont act Nodes in push Data ) , then
    • B Continues, and Push with Data Them

Event-Based Communication

• Decouple Sender/Recv to Comm SEND(Produce) and RECV (Listener) - A common Communication: Publish Subscribe sender recevie of - Subscription Based Topic or combo

Distributes Shared Memory

• The direct acc each is memory needs special machine the presence

Tuples Spaces

• Shared meme into generalized
• In the Model data Tuple other
• Underlying Middleware
• Tuples Ordering

Replication

• Involes create AND Maintian copys of a Service and Distributed Systems
• Unlik Comm WITHOUT Build the Fundamental Prinicrple
  • Bulid possible No Replication

##Replication Does

• Important WHEN key scale -- cases of Reliability Multiple server
• If a Server Become COrrupt replicas can use to restore state

Perfomance

• Replicating Helps Redue The Indivial

##Types of replication possbile

• Data replication

##control replication.

• In second cases

  • The Web With Caches Data Replication in Data Control: The web browsers are ex: replication data.

• improve & maintain performent of multiple servers: used over come or over computional

• During implement number adderssed Keep Replica

• the is keep the replicated data to addes

• The data consisten replica propogage how many replicas to create add and removed replicas etc

##Distributed-Datastore

• Generic Data data stored to data stores to include sharded,files web server objects etc.
• data is like a item
• clients connect and perform access
• From Clints view DATA Central Host Service Internally HOWEVER consisited Replica(servers)

Data Store Model Replicas Each

Data store performs for clients with client connecting