On 16/07/10 22:17, Paul Eggert wrote: > In looking at the random part of coreutils some more, I see some > issues. > > 1. Apps that use random numbers typically must link to -lrt, > for a very small benefit (nanosecond resolution rather than > microsecond resolution time stamp for random seed). Often > this "benefit" is illusory as the time stamps really are not > nanosecond resolution. > > 2. If /dev/urandom is available, it should be used to seed > the ISAAC generator. This will cost more than invoking > gettimeofday() but it's far more random. > > 3. We could get about 2X CPU performance on 64-bit machines by using > ISAAC64 instead of ISAAC. > > The patch below implements (1); I haven't installed it. I'd like to > do (2) and (3) too, but thought I'd ask for feedback first.
All changes make sense. `shred` doesn't need the entropy anyway as it's just scribbling random data to make it more difficult to recover data, and to hide its tracks. Performance wise `shred` probably won't run faster as it should saturate any disk, but using less CPU is good: `sort -R` and `shuf` should benefit also. thanks, Pádraig.
