[DGD] task rollback and random() bias

[Felix A. Croes]

Raymond Jennings <shentino at gmail.com> wrote:

> In Hydra, if you have a task rescheduled AFTER it has accessed random(),
> could that bias the RNG if certain values of random() at that spot made the
> task less or more prone to being rescheduled?

The seed of the pseudoramdom sequence will be the same when the task is
rescheduled.  It will only be updated when the task commits.

Each LPC-running thread has its own independent pseudorandom generator
seed.  You can expect slightly more random than usual sequences, because
variance in network timing will create scheduler entropy, affecting which
thread will be chosen to run which task.

Of course, this is still very far from cryptographically random.  For that,
/dev/random is needed.

Regards,
Felix Croes