Re: [boost] [lockfree::fifo] Review

Subject: Re: [boost] [lockfree::fifo] Review
From: Chris M. Thomasson (cristom_at_[hidden])
Date: 2010-01-23 19:48:05

"Chris M. Thomasson" <cristom_at_[hidden]> wrote in message 
news:hj84tr$7r8$1_at_ger.gmane.org...
> "Helge Bahmann" <hcb_at_[hidden]> wrote in message 
> news:alpine.DEB.1.10.1001191553030.25332_at_m65s28.vlinux.de...
[...]
> Now, I can think of a possible optimization/enhancement to the proxy 
> algorithm itself. Take note that an odd reference count denotes that a 
> collection cycle is in progress. Therefore, if a reader thread can simply 
> test if the reference count of it's acquired collector object is even, 
> then it simple does not actually "need" to release a reference to it. If 
> it's odd, then a release is necessary in order to move along the current 
> collection process. This would allow the reader threads acquire/release 
> pattern to be amortized across the actual garbage deferment threshold. I 
> think I am going to alter the algorithm in order to support the following 
> usage pattern:
> __________________________________________________________________
> #define DEFER_THRESHOLD 256
>
>
> typedef proxy<DEFER_THRESHOLD> proxy_type;
>
>
> static proxy_type g_proxy;
>
>
> void reader_thread()
> {
>    proxy_type::collector* c = &g_proxy.acquire();
>
>    for (;;)
>    {
>        query_type* q = try_to_get_a_query_request();
>
>        if (! q)
>        {
>            g_proxy.release(*c);
>
>            q = wait_for_a_query_request();
>
>            c = &g_proxy.acquire();
>        }
>
>        execute_query(q);
>
>        if (c->is_collecting())
>        {
>            g_proxy.release(*c);
>
>            c = &g_proxy.acquire();
>        }
>    }
>
>    g_proxy.release(*c);
> }
> __________________________________________________________________
>
>
>
>
> This would allow reader threads to curn along during periods of load with 
> fairly low overhead. I say this because the only time the readers would 
> have to release their collector object and re-acquire it is when they must 
> explicitly wait for query requests, or when the garbage threshold was 
> reached and a collection cycle was initiated by a writer thread. There are 
> some fairly promising enhancements I think I can make to the collector, 
> and that is one of them.
>
>
> Humm...

Refer to the following message over in `comp.programming.threads' for a 
model of the pattern above in Relacy 2.2:

http://groups.google.com/group/comp.programming.threads/msg/f03f8435a1792061

Enjoy!

;^)