(SEM VII) THEORY EXAMINATION 2022-23 DISTRIBUTED SYSTEMS

SECTION A – Short Answers (2 Marks Each)

(a) Consistent global state and transitless global state

A consistent global state is a collection of local states of all processes such that if a message is received, it must have been sent. A transitless global state is a global state where no messages are in transit; all sent messages are already received.

(b) Why vector clocks are better than Lamport clocks

Lamport clocks only capture event ordering but cannot detect concurrency. Vector clocks precisely determine causal relationships and can detect concurrent events, making them more suitable for distributed systems.

(c) Resource vs communication deadlock

Resource deadlock occurs when processes wait indefinitely for resources held by each other. Communication deadlock occurs when processes wait indefinitely for messages that will never arrive.

(d) Performance metrics for distributed mutual exclusion

Key metrics include message complexity, synchronization delay, response time, throughput, and fairness.

(e) Applications of agreement protocols

Agreement protocols are used in leader election, clock synchronization, atomic commit, fault-tolerant systems, and replicated data consistency.

(f) Benefits of grouping files into volumes in Coda

It simplifies replication, improves availability, supports disconnected operation, and eases management and recovery.

(g) Consistent vs strongly consistent checkpoints

Consistent checkpoints avoid orphan messages. Strongly consistent checkpoints ensure no messages are in transit at checkpoint time.

(h) Forward and backward recovery

Backward recovery restores the system to a previous checkpoint. Forward recovery corrects the error state without rolling back.

(i) Timestamp ordering algorithms (in increasing abort order)

Basic → Multiversion → Conservative timestamp ordering.

(j) Flat vs nested transactions

Flat transactions have a single execution level. Nested transactions allow sub-transactions with independent commit/abort.

SECTION B – Long Answers (10 Marks Each)

(a) Major issues in designing a distributed system

Key issues include transparency, scalability, fault tolerance, security, heterogeneity, concurrency control, synchronization, and performance optimization. Designing a distributed system requires balancing efficiency with reliability while hiding complexity from users.

(b) Deadlock detection algorithms & Path-Pushing algorithm

Deadlock detection algorithms are classified as centralized, distributed, and hierarchical.
In the Path-Pushing algorithm, each site maintains a local wait-for graph and sends dependency paths to other sites. Deadlock is detected when a cycle is found across combined paths.

(c) Architecture of Distributed File System

A DFS consists of clients, servers, communication network, naming service, caching, replication, and consistency mechanisms. It provides transparency, scalability, and fault tolerance.

(d) Dynamic vs static voting protocol

Static voting assigns fixed votes to replicas, while dynamic voting adjusts votes based on replica availability. Dynamic voting improves availability and fault tolerance compared to static voting.

(e) Lock-based concurrency control algorithms

These include two-phase locking (2PL), strict 2PL, and distributed locking. They ensure serializability but may cause deadlocks and increased communication overhead.

SECTION C – Long Answers (10 Marks Each)

(3a) Limitations of distributed systems

Limitations include network latency, partial failures, lack of global clock, security risks, and complex debugging. These lead to problems like synchronization errors, inconsistent data, and fault handling difficulties.

(3b) Fundamental and architectural models

The fundamental model includes interaction, failure, and security models.
The architectural model defines system structure such as client-server, peer-to-peer, and layered architectures.

(4a) Deadlock handling strategies

Deadlock handling includes prevention, avoidance, detection, and recovery.
Centralized detection uses one coordinator, distributed detection shares responsibility, and hierarchical detection uses multiple coordinators.

(4b) Mutual exclusion in distributed systems

Mutual exclusion ensures only one process enters the critical section at a time. It is more complex in distributed systems due to lack of shared memory and global clock.