Re: GC tuning idea

[Mark Hamburg <mark@...>]

The differences relative to my proposal is:

1. I don't want to stop the GC entirely. I just want to stop the  
atomic step from running. Incremental work would be viewed as safe  
(though of course a really big table could cause issues there as well).

2. I would allow the callback to decide that memory load was high  
enough that we should let the atomic step proceed even with the risk  
that it will cause a noticeable pause since we're probably at the  
point where we're going to be taking paging pauses.

Mark

On Aug 8, 2008, at 11:54 AM, Bilyk, Alex wrote:

> I was thinking this is a great idea. However, one can already  
> control GC
> manually, as far as I understand, by doing
>
> collectgarbage("stop ")
> ...
>
> if(app_thinks_it_is_a_good_time_and_place_to_GC())
> 	collectgarbage("step", application_GC_step)
> ...
> if(app_thinks_it_is_a_good_time_and_place_to_GC ())
> 	collectgarbage("step", application_GC_step)
>
> Plugging a custom application allocator allows gathering Lua  
> allocation
> statistics on the application side as well. So, looking back at  
> this, it
> feels like application can already do whatever it needs with respect  
> to
> GC as it is.
>
> Alex
>
> -----Original Message-----
> From: lua-bounces@bazar2.conectiva.com.br
> [mailto:lua-bounces@bazar2.conectiva.com.br] On Behalf Of Mark Hamburg
> Sent: Thursday, August 07, 2008 11:32 PM
> To: Lua list
> Subject: GC tuning idea
>
> Just a quick thought...
>
> It might be interesting to provide a hook in the GC process so that
> clients could postpone the atomic phase for the GC. One would probably
> need to pass in some statistics regarding memory usage so that the
> client could decide whether the memory usage outweighed the risk of
> taking a GC pause.
>
> Mark