You still need error checking (or some checksum) - if two packets are different, you still need to know which one was right. A very easy scheme is to send each packet twice (halving your bandwidth). Note that parity checks only work for 1-bit errors - in fact, a 2-bit error (in the same byte), the parity change will cancel out, and it will look ok.Īn alternative to requesting re-transmits, is to use forward error correction - send redundant data (reducing your bandwidth). If a small packet (very small!) is 20 bytes, and there is corruption 0.5% of those packets, this indicates a 0.5% / 20 bytes per packet / 8 bits per byte (assuming 7 data bits, 1 parity bit, for example). When used in communication (as it is here), parity errors usually mean you have to request a re-transmit (or, if acknowledgements are sent back, fail to send an acknowledgement and wait for an automatic re-transmit).Īlso, be aware that the relevant error rate is per bit, not per packet. Parity in memory is usually fatal - your code or data is corrupt, the solution is to halt the box, to stop this corruption (in memory) getting on to disk.
If you detect errors, you still have to handle them. Parity doesn't, and was never intended to, decrease corruption.Īll parity does is allow you to DETECT errors.
0 kommentar(er)
