Understanding Distributed Systems

Introduction to Distributed Systems

• Before the mid-1980s, computers were very expensive and slow, with costs reaching into the millions of dollars.
• These computers were not interconnected.
• After the mid-1980s, major developments included the appearance of cheap and powerful microprocessor-based computers.
• Computer networks emerged, with LANs ranging from 10 to 1000 Mbps and WANs from 64 Kbps to gigabits/sec.

Definition of a Distributed System

• A distributed system consists of independent computers that appear to users as a single coherent system.
• This definition considers autonomous hardware and software providing a unified user experience.

Paradigmatic Examples

• The Web over the Internet is a key example.
• Mobile telephony over cellular networks is another example.
• Electronic funds transfer systems operate over special-purpose networks, facilitating transactions between bank accounts, credit card purchases, and cash machines.

Other Examples of Distributed Systems

• Email and instant messaging are common examples.
• Videoconferencing and multiuser gaming are also instances of distributed systems.
• Home entertainment and global positioning systems (GPS) further illustrate this concept.

Benefits of Scale

• Interconnecting systems enhances the ability to solve problems more efficiently than centralized systems in sequential mode.
• Distributed systems enable more users to accomplish more valuable work effectively compared to centralized systems.

Characteristics of a Distributed System

• It is independent and self-sufficient.
• It is often heterogeneous and autonomous.
• Its components are spatially separated.
• Components use a common interconnection to coordinate actions, presenting a single coherent system to users.
• Differences between computers and communication methods are hidden from users.
• Users and applications can interact consistently and uniformly, regardless of location.
• These systems are designed for continuous availability, even with partial failures, and are easily expandable and scalable.

Key Aspects of Distributed System Components

• The "independent, self-sufficient" aspect means components have their own processor, state (memory), and resource control/management, such as an operating system.
• "Autonomous" means each component can change or be changed on its own accord without prior agreement or notification.
• "Heterogeneous" implies components may have differing capabilities, such as performance.

Heterogeneity in Distributed Systems

• Heterogeneity arises from different hardware, software, and software interfaces.
• These differences can cause interacting components to drift apart over time.
• Failures force components to manage gaps in their knowledge of the system state.
• Greater spatial distance between components increases the system's representation as a distributed system.

Advantages of Distributed Systems

• Resource and Data Sharing: enables sharing of printers, databases, and multimedia servers.
• Availability, Reliability: allows masking of loss through redundancy.
• Scalability, Extensibility: facilitates system growth with demand.
• Performance: provides substantial computational resources.
• Inherent Distribution: supports organizational distribution via tools like e-mail and video.

Challenges of Distribution

• Concurrency, Security: requires preventing clients from interfering with each other.
• Privacy: needs to protect preference profiles and prevent unwanted communication like spam.
• Partial Failure: makes it difficult to identify error locations and impacts Location, Migration, and Replication.
• Heterogeneity: must address differences in hardware, platforms, languages, and management.

Organizing a Distributed System

• Distributed systems are often structured using middleware to support heterogeneous computers and networks.
• Middleware provides a unified system view across multiple machines.

Goals of Distributed Systems

• Easily connect users with resources, including printers, computers, storage, data, files, and Web pages, for economic reasons and to facilitate collaboration.
• Be transparent: Mask the distributed nature of resources and processes across multiple computers.
• Be open and scalable.

Transparency in Distributed Systems

• A system is considered transparent if users and applications perceive it as a single computer system.
• Access transparency hides data representation and resource access methods.
• Location transparency hides the physical location of resources (e.g., through naming).
• Migration transparency hides resource movement.
• Relocation transparency hides resource movement during use, such as with mobile wireless laptops.
• Replication transparency hides resource replication.
• Concurrency transparency hides resource sharing among competitive users.
• Failure transparency hides failures and recovery.
• Persistence transparency hides whether a resource resides in memory or on disk.

Openness in Distributed Systems

• Open distributed systems rely on well-defined interfaces.
• Interoperability allows components from different origins to communicate.
• Portability allows components to work on various platforms.
• Flexibility and extensibility are key, allowing easy configuration and the addition/replacement of components.
• Open Distributed Systems offer services based on standard rules, like network protocols.

Scalability in Distributed Systems

• Distributed systems should be scalable to accommodate more users/resources (size), geographically dispersed elements, and multiple administrative organizations.
• Centralized services, data, and algorithms pose scalability limitations.

Scaling Techniques

• Addressing scalability problems involves hiding communication latencies.
• Distribution and replication are key strategies to solve scalability problems.

Hiding Communication Latencies

• Avoid waiting for remote service requests.
• Let the requester perform other useful tasks while waiting.
• Employ asynchronous communication to allow applications to continue until a reply arrives.
• Batch processing and parallel applications benefit from this approach.
• For interactive applications, shift part of the processing to the client to reduce communication.

Strategies for Software Systems that Scale

• Distribution involves dividing namespaces into zones, exemplified by the Domain Name System (DNS).
• Replication involves replicating components across a system to increase availability and distribute load, which can be managed by the resource owner.
• Caching, a special form of replication, reduces communication latency and is managed by the user, but caching and replication can introduce consistency problems.

Types of Distributed Systems

• Distributed computing systems
• Distributed information systems
• Distributed pervasive/embedded systems

Distributed Computing Systems

• Used for high-performance computing tasks.
• Includes cluster computing and grid computing.

Cluster Computing

• Involves homogeneous workstations or PCs connected via a high-speed LAN.
• Each node operates under the same operating system.
• Used for computationally intensive programs run in parallel.
• A master node manages middleware for parallel programs and controls compute nodes.
• It allocates tasks and provides a user interface.

Grid Computing

• Grid computing focuses on resource sharing and coordinated problem-solving in dynamic, multi-institutional virtual organizations.
• It features a high degree of heterogeneity.

Distributed Information Systems

• These systems often face Interoperability issues with several networked applications.
• Wrapping requests into distributed transactions is one technique to make systems interoperable.
• Enterprise Application Integration seeks to allow applications to communicate directly with each other.

Example of a Distributed Transaction

• Assume a banking operation involving a withdrawal from one account and a deposit into another.
• Without proper handling, problems can occur if operations are only partially completed.
• Transactions group operations to ensure either all or none are carried out, with a rollback mechanism for incomplete transactions.

Properties of Reliable Transactions

• Atomic: Transactions happen indivisibly.
• Consistent: Do not violate system invariants.
• Isolated or Serializable: Concurrent transactions do not interfere.
• Durable: Changes are permanent once a transaction commits.

Strategies for Enterprise Application Integration

• Integrate applications independent from their databases.
• Design transaction systems to rely on request/reply patterns.
• Let applications communicate with each other.

Distributed Pervasive Systems

• These systems emphasize stability with fixed nodes and high-quality network connections.
• Wireless mobile and embedded computing devices are included.
• Embrace contextual changes.
• Encourage ad hoc composition.
• Recognize sharing as the default.

Hardware Concepts of Distributed Systems

• Multi-processors utilize shared memory.
• Multi-computers do not use shared memory.
• These systems can be homogeneous or heterogeneous.

Shared vs Private Memory Setups

• Shared Memory system shares a single backbone.
• Private memory system uses switches.

Multiprocessors and Shared Memory

• Shared memory must be coherent so that values written by one processor can be read by others.
• Bus-based organizations face performance bottlenecks as the number of processors increases.
• Cache memory aims to improve performance, which may cause incoherent memory.

Multi-processor Scaling Solutions

• Bus-based multi-processors struggle to scale
• Using a crossbar switch or omega network is required.
• Crossbar switches divide memory into modules connected via crosspoint switches, but are expensive.
• Omega networks reduce cost but introduce high latency between CPU and memory.

Software Concepts

• There are OSs in relation to distributed systems: Tightly-coupled systems and Loosely-coupled systems
• Tightly-coupled systems tries to maintain a single, global view of the resources.
• Loosely-coupled systems is a collection of computers each running its own OS.
• Middleware enhances services of network OSs to improve distribution transparency.

Summary of Main Issues

• DOSs are tightly-coupled operating systems for multi-processors.
• NOS are loosely-coupled operating system.
• Middleware is used to provide distribution transparency.

Distributed Operating Systems

• Multiprocessor OSs manage resources in a multiprocessor system.
• Multicomputer OSs exist for homogeneous multicomputer systems.
• Uniprocessor OSs separate applications and OS code via a microkernel.

Multiprocessor / Uniprocessor Operating Systems

• These systems extend uniprocessor OSs to support multi-processors with shared memory.
• A protection mechanism is used to guarantee consistency.

Multicomputer Operating Systems

• Processors cannot share memory, relying instead on message passing.
• Each node has its kernel for managing local resources.
• Each module handles interprocess communication.

Implementation of Distributed Shared Memory Systems

• Sharing can be emulated on distributed systems using a virtual shared memory that uses virtual memory capabilities.
• Addressing is handled at the individual node level.

How Distributed Memory Pages / Reads Work

• Read-only pages can be easily replicated.

Network Operating Systems

• These are possibly heterogeneous at a base level.
• These are constructed from a collection of uniprocessor systems with their own operating system.
• These systems are each independently connected to each other in a computer network.

Network OS Common Services

• Remote login rlogin
• Remote file copy rcp
• Shared file systems via file servers

Middleware

• These are not intended to handle a collection of independent computers as distributed OS do.
• The network operating system does not provide a view of a single coherent system that's scalable and open.
• Transparency and ease of use are combined.

Middleware Topologies

• Middleware services provide: access transparency, naming conventions, distributed transaction, security