DS properties and interaction model

Properties of a distributed system

• Concurrency: components execute in parallel and share resources.
• No global clock: coordination is only by message passing; each computer has its own clock, with drift.
• Independent (partial) failures: a component or link may fail while others continue.
• Heterogeneity: variation in hardware, OS, network and languages.
• Openness, scalability and transparency.
• Resource sharing and message-passing communication with delays.

Interaction model

Processes in a distributed system interact by passing messages, and the performance of communication limits what can be done. The interaction model addresses the following issues.

1. Performance of communication channels: characterised by latency (delay between sending and receipt), bandwidth (amount of data per unit time) and jitter (variation in delivery time of messages, vital for multimedia).
2. Computer clocks and timing events: each process has its own clock, and clocks drift at different rates (clock drift rate). Two processes cannot give exactly matching timestamps.
3. Two variants of the model
   • Synchronous: known upper bound on process execution time, message delay and clock drift. Timeouts can be used to detect failure, but real systems seldom guarantee this.
   • Asynchronous: no bounds; a slow process cannot be told from a failed one. It reflects the Internet and is the safer general assumption.
4. Event ordering: because there is no global time, the order of events is found using logical clocks (Lamport, "happened-before") instead of physical time.

So the interaction model states the assumptions about delay, clock and ordering, and thereby decides which algorithms (such as failure detection or mutual exclusion) are possible.