Memory chunks: are they much speed-up?

Memory chunks: are they much speed-up?

[Victor Porton]

Is allocating memory in chunks by several (small) objects at once a 
significant speed-up (compared to standard allocators) in typical Ada 
impl.?

Re: Memory chunks: are they much speed-up?

[Mark Johnson]

Victor Porton wrote:
> 
> Is allocating memory in chunks by several (small) objects at once a
> significant speed-up (compared to standard allocators) in typical Ada
> impl.?

It is hard to say because it depends upon a number of factors:
 - your allocation, deallocation pattern
 - the algorithm used in a specific allocator
 - the relative timing of allocations
and so on.

In general, a special purpose allocator will always be "faster" than a
general purpose one. It may also be "more efficient" (less unused
space). However, in general, I find it better to avoid allocation /
deallocation in my applications - especially in real time and "not have
that problem". There are several ways to do that; ask if you need some
examples.
  --Mark

Re: Memory chunks: are they much speed-up?

[Stephen Leake]

porton@ex-code.com (Victor Porton) writes:

> Is allocating memory in chunks by several (small) objects at once a 
> significant speed-up (compared to standard allocators) in typical Ada 
> impl.?

It might be. Depends on whether the standard allocator already does
that. So you'll have to test and see; the answer will vary with
compiler & os & target chip.

-- 
-- Stephe

Re: Memory chunks: are they much speed-up?

[David Marceau]

Victor Porton wrote:
> 
> Is allocating memory in chunks by several (small) objects at once a
> significant speed-up (compared to standard allocators) in typical Ada
> impl.?
If you follow ravenscar or spark I don't think you're even allowed to
allocated dynamically small or big chunks.
Everything is static.
Memory leaks are not a problem in that way.

Going back to your question though if it wasn't critical or
high-integrity, then yes using memory pools each dedicated to a certain
type of object and then reserving the memory pools in advance is
certainly much more efficient.  If ever one of the memory pools is full,
no problem just dynamically allocate another memory pool dedicated to
the object type in question.

I have seen one library product dedicated to c/C++ programmers in this
area in the past.  SmartHeap by Microquill.  You might want to check it
out.

Do memory pools impact/improve software performance?  Yes.

I hope this helps.

Cheers,
David Marceau