Distributed Operating Systems (Unit 4)

Scheduling

Normal time-sharing scheduling algorithms could lead to latency issues. When a processor tries to call another processor which will only get executed in the next time slice, it only increases the execution time.

• Co-Scheduling - Uses a conceptual matrix in which each particular processor which is being called is in a particular row. This makes a round-robin scheduling algorithm with all processors running first the processor for a fixed time then, the next, and so on.
• Real-time
• Static
• Dynamic

Fault Classifications

• Component Faults
• Transient - Can occur once and then disappear.
• Intermittent - Can occur, disappear, and reoccur later.
• Permanent - They stay till the component is repaired.
• System Failures - Processor Faults
• Fail Silent - Just stops upon fault and does not respond.
• Byzantine - Continues to run and gives malicious or wrong output.
• Synchronous - Responds to a message within known finite bound if working.
• Asynchronous
• Redundancy
• Information
• Time - More time for intermittent and transient faults.
• Physical - Extra equipment for better fault detection.
• Active Replication - Uses physical redundancy (same as Triple Modular)
• Primary Backup - Primary server works, when down, the backup takes a load.
  
  
  
• Triple Modular
  

Threads

System Models

• Workstation model
• Diskless
• Diskful
• Hybrid
• Idle
  
  
  

Design Issues of Processor Allocation Algorithms

• Deterministic vs Heuristic
• Centralised vs Decentralised
• Optimal vs sub-optimal
• Local vs Global
• Sender Initialted vs Receiver Initiated.

Processor Allocation Algorithms

Dynamic Load Distribution Components

Dynamic Load Distribution Algorithms

• Centralised
• Hierarchical
  
  
  
• Sender-Initiated
  
  
  
• Receiver-Initiated
  
  
  
• Bidding
  
  
  
• Direct
• Nearest Neighbour
  
  
  

RTS