Efficient thread-local shared variables
Klaus D. Witzel
klaus.witzel at cobss.com
Tue Oct 24 17:17:29 UTC 2006
Hi Reinout,
on Tue, 24 Oct 2006 18:44:49 +0200, you wrote:
> Klaus wrote:
>>
>> I don't know why but the discussion with you "inspired" three new
>> operations for my implementation:
>>
>> - specialization, clones the anonymous class (shares the method
>> dictionary)
>> - proliferation of name space, locally copies the method dictionary
>> ("rapidly" doubles it, still shares methods, variables and state)
>
> I presume you want to do above at Process creation time, and
Yes, e.g. at "schema creation" time.
>> - proliferation of state, locally clones "un-shares" a variable (and so
>> its state)
>
> on global var assignment.
Yes, e.g. at "schema evolution" time.
>
> Another approach could be to (ab)use SearchOrder. Assuming you still
> have something like a 'global or 'default' scope you can put that at the
> bottom of your SearchOrder and have a per-thread Vocabulary (perhaps
> with scoping support) at the top of the SearchOrder.
>
> So at process creation time you only need to create an *empty*
> Vocabulary for each scope and chain it to the global SearchOrder,
> reading a global var will now fall through to the global definitions
> while assigning a new value to that var will put it in the topmost
> Vocabulary masking the global state of that variable.
:)
> Thinking this through even further: you could defer the creation of the
> thread-local Vocabularies in the SearchOrder until the first write is
> done, if the code only does reads the Process creation overhead can be
> pretty close to zero :-)
>
Thank you for sharing the ideas!
/Klaus
>
> R
> -
>
More information about the Squeak-dev
mailing list
|