Re: [Sbcl-devel] back from the dead: lazy allocation for sbcl

[Brian M. <cha...@un...>]

On Jun 4, 2005, at 5:40 PM, Peter Van Eynde wrote:

> The main aim of this patch is to delay real allocation of memory until
> it is needed. For the static, read-only and linkage-table spaces  
> this is
> done by handling sigsegv's by quickly mapping the memory and retrying.
> For the dynamic space the memory allocator is modified to map in pages
> as required and as a backup (not needed it seems) the sigsegv is also
> checked.

Hi Peter,

I'm a bit confused here: as far as I'm aware, the dynamic space is  
only given real pages as it needs it.

> The net result is a sbcl with a memory footprint of 34MB. This to a
> system administrator looks a lot more reasonable then 800+MB and also
> allows the system to work with restrictive ulimits.

Only because most system administrators couldn't tell you the  
difference between virtual memory, real memory, and a toaster.

> Going further I think it could be possible to have 'holes' in the  
> memory
> spaces if the linker decides to place stuff in inconvenient places.

Wouldn't this require a huge page table of some kind?
--
Brian Mastenbrook
br...@ma...