🚀 go-pugleaf

Gavin Estey <gavin@senator.demon.co.uk> wrote:
>In article <809779542snz@onex.demon.co.uk>
>           Martin@onex.demon.co.uk "Martin Randall" writes:
>
>> 
>> OK, here's a new one to throw into this thread - but not totally un-related.
>> Which is faster new or malloc ? My C++ book says that new should be faster.
>
>Why should you care which one is faster? You don't usually allocate memory
>in inner loops or speed critical areas. New is better, maybe not faster,
>but better.
>
>-- 
>Gavin Estey

Why should anyone care which is faster?!?  I can think of many applications
where the speed of dynamically allocating memory is key to performance.
For example, take any communications layer that is transferring packets
of data around.  Often receipt of a message would require a malloc and a
memcpy.  For packet sizes of less than 4Kbytes, the numbers on my platform
show that the cost of a malloc is more than that of a memcpy.  For very small
messages (on the order of tens of bytes) the cost is basically entirely
a malloc.  At this point the performance of malloc is very important,
and you have to decide whether to manage your own memory or not.

I would also think that anyone writing code that does heavy manipulation
of dynamic data structures (requiring lots of news and deletes as elements
are added and removed) would be concerned about the cost of memory
allocation.  Applications like this further highlight the difference
between new and malloc since new is invoking the constructor each time,
which may make the allocation of small objects relatively expensive.
Stroustrup addresses the small object allocation problem in his book
on pages 176-178.


--Paul