Principle: End-to-end reliability
Obtaining reliable communication:
- Rely on the end points
- Use lower level communication systems only to deliver bits with some probability
- End points detect errors and retry as needed
Example: big-bang network file copy
- File is sent and stored on disk
- Sender computes a checksum and sends it
- Received recomputes checksum and compares
- Retransmit if checksums are different
In such an end-to-end system, the total cost to send a message is 1 + r ? p, where
- 1 is the normalized cost to send a message
- r = cost of retry
- p = probability of retry
This is just like fast path