Distributed Systems: Design and Architecture

• A distributed system is a collection of systems that appear to the end user as one cohesive system.
• In a distributed system, nodes work together over a network to contribute processing power, storage, or memory to create a powerful, scalable, fault-tolerant system.

Design Principles of Distributed Systems

• Scalability involves designing systems to scale horizontally by adding more nodes to handle increased demand.
• Fault Tolerance means systems continue to function even if one or more nodes fail.
• Decentralization requires systems to to avoid single points of failure and bottlenecks
• Transparency requires systems to provide seamless user experiences by hiding complexities of the distributed architecture, so the system appears as a single, unified whole.
• Flexibility requires systems to accommodate different hardware, software, and network configurations.

Client-Server Architecture

• Client-server architecture is a computing concept where multiple client nodes request service from a central server system.
• The server processes requests and responds.

Client

• A client is the requesting entity, such as a device or application requesting services or data from a server, and initiates communication.
• Web browsers and mobile apps are examples of clients.

Server

• A server is the providing entity, such as a device or software application that provides services, resources, or data upon a client's request.
• Web servers, database servers, and file serversare examples of servers

Network

• The network is the communication medium connecting clients and servers which can be a LAN (Local Area Network) or WAN (Internet).

Two-Tier Architecture

• In two-tier architecture the client communicates directly with the server.
• Example: a desktop app accessing a database

Three-Tier Architecture

• Three Tier Architecture = Client + Application Server + Database Server
• The application server processes business logic before accessing the database using the server.
• Example: a web app where the browser communicates with an API (server) that fetches data from a database.

Multi-Tier Architecture (N-Tier)

• More than three layers, adding load balancing, caching, and security layers.
• This architecture is used in large scale cloud applications.

Advantages of Client-Server Architecture

• Data is stored and managed centrally.
• Servers enforce access control and authentication.
• Servers can handle multiple clients efficiently.
• Clients access data from one source, ensuring updated data.

Disadvantages of Client-Server Architecture

• The server may slow down if too many clients request data at once.
• Clients lose access to data if the server crashes.
• This setup can be costly

Examples of Client-Server Architecture

• Web browsing: browsers requesting a webpage from a server.
• Email services: email apps connecting to an email server (e.g., Outlook, Gmail).
• Cloud storage: Google Drive and Dropbox uses this model for file storage.
• Online banking: apps communicate with central servers for transaction processing.

Peer-to-Peer Architecture (P2P)

• P2P is a distributed system architecture where all nodes, or peers, are equal and act as both clients and servers.
• Peers share resources like files or computing power.

Key Characteristics of P2P Architecture

• All peers are equal
• There is no central authority so it is decentralized
• Each peer operates autonomously

Advantages of P2P Architecture

• Improved scalability because each peer can contribute to the network.
• Increased fault tolerance because other peers can take over responsibilities of failed peers
• Better resource allocation as the system efficiently utilizes resources.

Disadvantages of P2P Architecture

• This can be more complex to design and implement as compared to client-server.
• This has increased security risks
• Network congestion can occur

P2P Architecture Models

• Unstructured P2P does not have a centralized index.
• Gnutella an example of unstructured P2P architecture
• Structured P2P uses a distributed hash table (DHT) to index resources.
• Chord is an example of Structured P2P architecture
• Hybrid P2P combines elements of unstructured and structured P2P networks.
• BitTorrent is an example of Hybrid P2P architecture

Examples of P2P Architecture

• BitTorrent is a file sharing network
• Bitcoin allows users to share and verify transactions.

Middleware

• Middleware is software that acts as an intermediary between distributed applications to enable communication, data exchange, and coordination.
• Middleware provides a layer of abstraction to allow developers to focus on application logic.

Functions of Middleware

• Inter-process communication(IPC) facilitates communication between different processes across networks.
• Remote procedure calls(RPCs) and Remote Method Invocation(RMI) enables function/method execution across different machines.
• Message queuing ensures asynchronous communication between components.
• Security and authentication manages access control and encryption.
• Load balancing distributes workload across multiple servers.

Middleware Technologies

• Message-oriented middleware uses message queues or asynchronous communication.
• Apache Kafka, and RabbitMQ are examples of Message-oriented middleware
• Object-oriented middleware enables distributed objects to communicate.
• CORBA is an example of Object-oriented middleware
• Database middleware provides a bridge between databases and applications.
• ODBC and JDBC are examples of Database middleware
• Cloud middleware connects cloud applications and services.
• AWS Lambda is an example of Cloud middleware

Communication Protocols

• Communication protocols define the rules and formats for data exchange between nodes.

Key Communication Protocols:

• Transmission Control Protocol (TCP) is a Transport Layer Protocol
• User Datagram Protocol (UDP) is a Transport Layer Protocol
• HTTP/HTTPS: used in web-based distributed systems, is an Application Layer Protocols
• Message Queuing Telemetry Transport (MQTT) for IoT applicationsis an Application Layer Protocols
• RPC allows calling procedures on remote systems and is an Remote Communication Protocol
• RMI is an Remote Communication Protocol
• Websocket enables real-time bidirectional communication in web applications as an Event-Driven and Publish-Subscribe Protocol

Middleware and Protocol Selection Criteria

• Latency requirements
• Scalability
• Security needs
• Data consistency

Naming and Addressing

• Naming and addressing are critical components of distributed systems, enabling nodes to identify and communicate with each other.
• Naming involves assigning a unique identifier (e.g., name or ID) to each node or resource.
• Addressing consists of assigning a unique address, like an IP address, to each node or resource.

Common Naming and Addressing Schemes

• Domain Name System (DNS) is a hierarchical naming system that enables nodes to resolve domain names to IP addresses.
• Uniform Resource Locator (URL) is a naming scheme that enables nodes to identify resources on the web.

Synchronization and Concurrency Control

• Synchronization and concurrency control ensure that multiple nodes can access shared resources safely and efficiently.
• Synchronization involves coordinating the actions of multiple nodes to ensure consistent and predictable access to shared resources.
• Concurrency control manages multiple requests to shared resources, ensuring that only one node can access the resource at a time.

Locks

• Locks are mechanisms that enable nodes to acquire exclusive access to shared resources.
• A lock mechanism allows only one process or thread to access a shared resource at a time.

Lock Acquisition Process:

• Lock Acquisition: A process requests a lock on a shared resource.
• Lock Granting: If the lock is available, it is granted to the requesting process.
• Resource Access: The process acquires a lock and can now access the shared resource.
• Lock Release: When the process finishes accessing the resource, it releases the lock.

Types of Locks:

• Exclusive locks (X-locks): only one process can hold an X-lock on a resource at a time.
• Shared locks (S-locks): multiple processes can hold an S-lock on a resource at a time.

Challenges and Limitations of Locks

• Deadlocks: two or more processes can wait indefinitely for each other to release a lock.
• Starvation: a process may be unable to access a shared resource due to continuous lock acquisition by other processes.
• Performance overhead: Lock acquisition and release can generate significant overhead.
• Lock timeouts and lock escalation techniques are introduced to overcome these challenges

Semaphores

• A semaphore is a variable that controls access to a shared resource by multiple processes.

Semaphore Operations:

• P-operation: decrements the semaphore value; if the value is already zero, the process is blocked until the value becomes positive.
• V-operation: increments the semaphore value; if there are blocked processes, one of them is unblocked.

Semaphore States:

• Available (semaphore value > 0): the resource is available and can be accessed.
• Busy (semaphore value == 0): the resource is busy, and processes are blocked from accessing it.
• Blocked (semaphore value < 0): processes are blocked, waiting for semaphore value to become positive.

Distributed System Models

• Distributed systems support critical technologies such as cloud platforms, blockchain networks, and large-scale web services.
• Designing and managing these systems come with challenges related to concurrency, partial failures, unpredictable communication delays, and heterogeneous components.
• A distributed system model is an abstract framework that simplifies reasoning about system behavior, trade-offs, and solutions.

Client-Server Model

• Based on a request/reply protocol implemented with send and receive or Remote Procedure Calls(RPC) or Remote Method Invocation(RMI).
• A client sends a request message to the server.
• The server performs the work and returns the result or an error code
• A server may require services from other servers.

Key Features of the Client-Server Model

• Centralized control
• Asymmetric roles
• Scalability

Advantages of the Client-Server Model

• Simplicity in design and management.
• Strong access control and security .
• Easier maintenance of consistency

Disadvantages of the Client-Server Model

• Single point of failure
• Bottlenecks under high load
• Network latency for geographically distant clients from the server.

Use Cases f the Client-Server Model

• Traditional web applications
• Centralized databases
• Enterprise systems with access controls

Peer-to-Peer Models

• General and flexible framework
• All objects/processes have a similar role
• Objects or processes interact without any particular link between clients and servers.
• A large number of data files are transferred and a home computer has a tiny place in the database.
• Distributed share resources widely.

Key Features Peer-to-Peer Models

• Equal stakeholders
• Peer-to-peer are self-organized
• Flexible support for shared resources

Advantages of Peer-to-Peer Models

• No point of failure
• Improves scale as capacity increases
• Cost effective. Leverages remaining power

Disadvantages of Peer-to-Peer Models

• Compexity coordinating discovery
• Security risks such as Sybil attacks
• Data can be outdated and fragmented

Use Cases Peer-to-Peer Models

• Sharing networks
• Decentralized applications
• Collaborative applications

Publisher/Subscriber (Pub/Sub) Model

• A messaging pattern used in software architecture to facilitate asynchronous communication between different/ same systems.
• In this model, the messages sent are then transfered for each user.

Entities of Pub/Sub Model

• Publishers send and create messages
• Consumers utilize and acquire messages
• Topics use channels for certain types of messages
• Intermediate are Message Brokers, routing messages between subscribers and publishers

Characteristics of Pub/Sub Model

• Asynchronous communications
• Classifying messages by target
• The high throughput

Advantages of Pub/Sub Model

• Uncoupled by operating alone
• Live messaging distribution through loT
• The larger scale manages the number of streams by shared brokers

Disadvantages of Pub/Sub Model

• Prone to failure
• Message can be over-sent
• Subscriptions need back pressure

Use Cases of Pub/Sub Model

• Analytics in real-time
• Networks that are smart
• Support for driven message

Trade-offs and Design Choices

• Client and server environments are stable
• P2P provides safety from faulty blockchains
• The real time messages are pushed out via this structure