Anthony DeRobertis, le Sat 08 Dec 2001 10:43:36 -0500, a écrit : > Seriously, why the second test? Assuming MD5 isn't horribly broken, > there is a higher chance of a random bit error reading the data than of > MD5 giving a false answer.
Just a few gross figures for a 4Ko file. A typical disk MTBF is of the order of a million hours. Considering 40Mo/s speed, reading 4Ko takes roughly 1/10,000th second. As a result, you have a rough read error probability of 1/36,000,000,000,000. On the other hand, MD5 (128bits) collision probability is roughly 1/2^128, i.e. something like 1/256,000,000,000,000,000,000,000,000,000,000,000,000. I indeed doubt the disk industry will ever be able to get a MTBF better than that. Samuel
