On Sat, Oct 28, 2017 at 11:59 PM John Souvestre <j...@souvestre.com> wrote:
> If it were uniformly fair then you couldn’t guarantee picking every > channel once in a while. Statistically it would be become more and more > probable, but never 100%. > > > You are touching on the subject of there being different ways to treat fairness in a system. Take TLA+ for instance: In a (strong) fairness setting, what you say are true in that the system can always pick another path and thus starve a channel. Even if that pick is random. However, there is also weak fairness in which it is adequate that a channel will eventually be chosen if it is enabled. That is, even if you "cycle" on other choices, you are golden as long as there is a path out. Randomness can provide weak fairness, but not strong fairness in most situations. In practice, many algorithms work on principles of weak fairness. Another example is the BPP complexity class. BPP problems are: * allowed to flip coins * must run in polynomial time * must give the correct answer 2/3 of the time If you just rerun the algorithm enough, then eventually you will know the correct answer with high probability. It's cousin class, BQP, is the quantum variant and it is important for quantum computation. As for round-robin, its weakness is in patterns. Since you can nil channels, you can end up in a pattern where things are never selected. Randomness is a guard against this because such patterns are eventually broken up by the dice roll. To be precise: a round-robin solution will be less robust and sometimes drop weak fairness in situations where random selection won't. -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
