Le jeudi 17 février 2011 à 19:48 +0100, Juergen Harms a écrit : > >> Well, in fact, given the way tcp work, I do not see how it could > >> suffer from a transmission problem. > > Tcp corrects errors with a certain likelyhood. I do not know what > CRC-something is used by ftp - for CRC-16, depending on the type of > error and the package length, probablilities are smaller than 10 power > -6 (I did these figures many years back, rapid googling does not provide > any evident figures). Anyhow, with this order of magnitude, and with the > 100 M bytes(!) of a typical largish package, the probability of > receiving an apparently good package which after all is not so good is > non-negligeable. Doing another check at the application level therefore > may be of substantial merit. Would be good to get input from somebody > who is more competent.
IP and TCP both have error detections with 16 bits checksum. And the chance that a error slip in such a way that it is not detected by neither checksums, and at the same time that the error happen at the physical level ( ie, cable level ) and do not happen more than once are really really rare. Even if you would cause such a error, you would likely not be able to. -- Michael Scherer
