DC PPT
DC PPT
Algorithms in
Distributed Systems
By- Nishtha Desai E217
Rasanath Kulkarni E218
PROBLEM STATEMENT
• In distributed systems, achieving consensus among nodes is critical
for ensuring consistency and fault tolerance. Traditional consensus
algorithms like Paxos and Raft face challenges such as high
communication overhead, slow convergence in large-scale systems,
and vulnerability to Byzantine faults. This project aims to analyze
existing consensus protocols and propose an optimized hybrid
consensus algorithm that improves efficiency, fault tolerance, and
scalability.
PAXOS
• Paxos is a consensus algorithm that
ensures agreement among distributed
nodes, even in the presence of failures.
How It Works:
• Uses a three-phase process: Prepare,
Accept, and Learn to achieve consensus.
• Designed for fault tolerance but is
complex and difficult to implement.
Use Case: Distributed databases, cloud
storage, Google’s Chubby lock service.
RAFT
• Raft is a consensus algorithm that
simplifies leader election and log
replication for distributed systems.
How It Works:
• Elects a leader node to manage log
replication among followers.
• Follows a simpler and more
understandable approach compared
to Paxos.
Use Case: Kubernetes etcd, Apache
InfluxDB, and replicated storage
systems.
PROPOSED SOLUTION
• We propose a Hybrid
Adaptive Consensus
Algorithm (HACA) that
dynamically switches
between different
consensus mechanisms
based on network
conditions and failure
rates.
ALGORITHM STEPS
1. Use Raft for normal operations, ensuring leader-based fast
decision-making with minimal communication.
2. Switch to Byzantine Fault Tolerant (BFT) consensus in case of
detected malicious activity or untrusted nodes.
3. Optimize message passing by reducing redundant communication
overhead through adaptive quorum selection.
4. Leverage machine learning to predict network failures and
preemptively adjust leader election strategies.
ARCHITECTURE
ANALYSIS
Advantages:
• Improved Performance: Reduces network latency by using Raft in
normal conditions.
• Enhanced Security: Detects and mitigates Byzantine faults
dynamically.
• Scalability: Adaptive quorum selection optimizes communication
overhead.
• Resilience: Predicts failures and proactively adjusts consensus
mechanisms.
ANALYSIS
Challenges:
• Increased complexity due to hybrid model integration.
• Higher computational requirements for machine learning-based
failure prediction.