Re: mempools conversions, more ..

On 18 Oct 2000, at 6:46, Adrian Chadd <adrian@creative.net.au> wrote:

> On Tue, Oct 17, 2000, Alex Rousskov wrote:
> >
> > -- Split Array into Array and PooledArray? MemPools will use Array that
> > does not use MemPools. PooledArray can use MemPools.

 This is the same as implement array within mempools itself as private.
 Would we need two public libs for arrays?

> I dare say that the real mempool implementation should really use
> mmap() or sbrk() and become a memory allocator in itself. However,
> that is probably something for 2.5 unless someone is very very
> resourceful and does the work now. :-P

 uh, I tend to respect evolution history of malloc, there has been lots
 of science and research put into it. I don't think we can beat dlmalloc,
 and I don't see much good in going through all the portability issues
 again.
 Of course its quite evident that squid mem usage patterns are too
 heavy for general purpose mallocs. So I think we should first try to
 not overload malloc, what mempools are exactly for.
 We should try to reduce different alloc sizes, perhaps even at a cost
 of some memory waste, just to allow best-fit algoritms to work and
 reduce fragmentation. For very small allocs I think its reasonable
 to do preallocation in larger chunks and manage these chunks within
 mempools.

------------------------------------
 Andres Kroonmaa <andre@online.ee>
 Delfi Online
 Tel: 6501 731, Fax: 6501 708
 Pärnu mnt. 158, Tallinn,
 11317 Estonia
Received on Wed Oct 18 2000 - 02:32:26 MDT