On Sat, Sep 5, 2009 at 5:30 PM, Krukow<karl.kru...@gmail.com> wrote:
>
>
>
> On Sep 5, 8:18 pm, Rich Hickey <richhic...@gmail.com> wrote:
>> Given sufficient history, readers will not be retried due to the
>> activity of writers. It is true that while history is being
>> dynamically acquired there may be retries. Unless you have some
>> pathological transaction relationships, that history acquisition will
>> occur in a small amount of time, and thus is being ignored as would a
>> constant factor. Perhaps 'never' is too strong a term, but history
>> acquisition is somewhat of an implementation detail. If that build-up
>> phase is an actual problem, you now have :min-history to mitigate it.
>
> Ok - thanks for the clarification, and :min-history seems a very
> practical tool.
>
>> There is potential for a 'separate worlds' design, but the devil is in
>> the details of allocation of refs to worlds. While the single shared
>> CAS is a theoretical bottleneck, I'd have to see it become an actual
>> bottleneck before introducing that complexity.
>
> Yes, I understand. Perhaps it may become another "knob" of the STM
> system if there is a convincing use-case, e.g.
> (ref {} :min-history 42 :world :x)
>As I said, devil's in the details. Any name-based scheme that would have transactions looking up STM instances by name would just introduce another bottleneck (the name-to-STM-instance map). Rich --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Clojure" group. To post to this group, send email to clojure@googlegroups.com Note that posts from new members are moderated - please be patient with your first post. To unsubscribe from this group, send email to clojure+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/clojure?hl=en -~----------~----~----~----~------~----~------~--~---
