[erlang-questions] Trie fun

Gleb Peregud <>
Mon Jun 27 20:47:29 CEST 2011


Maybe creating a module in runtime, where a function returns a
prebuild Trie as a static term, will be a good solution? This way
processess will be able to share the same Trie in the same area of
memory. AFAIK static terms from modules are not copied from module's
"static term storage" to each process which uses it.

On Mon, Jun 27, 2011 at 20:38, Mike Oxford <> wrote:
> Erlang's MR against a trie seems ideal for parallelizing a word filter.
> However, the filter-terms becomes the bottleneck.
>
> In many languages we'd put it into memory and acquire a
> pointer/reference to it, spawn the worker threads and pass them the
> address of the filter list.
> This is efficient as it's static-data and only in memory in one place.
>
> In Erlang we build up the trie and then send it messages, leading to a
> bottleneck at the filter-list's message queue.
> Okay, so we can spawn multiple filters, and ship words to them, but
> every filter has to build up the filter-list from storage (compiled
> in, in this case, to minimize overhead.)
> Now we have 'm' words going to 'n' filters.  Except for the startup
> costs, that doesn't seem too bad.
>
> The Trie is static, prebuild it and store it as a binary "variable",
> so it's just blasted into memory instead of actually being built?  Any
> downsides? (Theory; I don't know if you
> can "image" memory this way in Erlang yet.)
>
> Is there a better way?  Yes, I could do it in C and link it in but I'm
> interested in the Erlang way.  :)
>
> Thanks!
>
> -mox
> _______________________________________________
> erlang-questions mailing list
> 
> http://erlang.org/mailman/listinfo/erlang-questions
>



More information about the erlang-questions mailing list