Errors may be both detected and corrected

• Errors can occur at all layers of networking - physical interference, faulty RAM in intermediate routing devices, failure to conform to agreement/conventions, failure to conform to standards, coding errors in software implementations.
• Physical media errors occur in bursts - we are only interested in the presence of burst itself, not what it changed.
• How errors are addressed depends on:
  • the likelihood of errors,
  • the extent of errors,
  • the cost of detection,
  • the cost of ignoring the error, and
  • the cost of recovery/repair.
• Error correction is possible, but requires much redundant data, and is computationally expensive. Generally only employed for specific applications.
• Error detection, with CRCs or checksums, is much faster and widespread.
• Error detection is often employed at multiple layers, in hardware and software.
• Error detection algorithms should be chosen based on the extent and type of errors expected.
• Three strategies exist - just tolerate errors (retransmit), simply ignore data in error, or or pro-actively announce errors.