malloc replacement for windows

[Tushar Udeshi]

Hello,

We have a single-threaded application running on Windows which does a
large number of news and deletes of small chunks of memory (<128 K). I
was looking to improve performance and compiled ptmalloc
(http://www.nedprod.com/programs/Win32/ptmalloc2/) into our
application. On doing so I got a 15% performance improvement. However,
I found the application consuming far more memory than it used to when
using the default malloc. I monitor the memory usage in Task Manager
and it looks as though memory consumption is continuously increasing.
On the other hand using the default malloc causes memory consumption
to increase and decrease through various phases of the program as
expected. I suspect there is too much fragmentation. Are there any
flags I can tweak to reduce this? I already tried enabling the
TRIM_FASTBINS flag.

Also, suggestions for any other malloc replacement under windows would
also be greatly appreciated

 

[Claudio Puviani]

...
Also, suggestions for any other malloc replacement
under windows would also be greatly appreciated

Look into SmartHeap and the HOARD allocator (let google be your
guide).

Claudio Puviani

 

[Dave Arnette]

Look into SmartHeap and the HOARD allocator (let google be your
guide).

Claudio Puviani

If speed is of any concern, also look into ESA from Cherrystone
software as well. We (comcast) hired a company to do performance and
scalability testing of various memory managers (because we were just
too darn busy to do it ourselves) and the final result proved to that
ESA from cherrystone was 2x faster than smartheap in the SMP tests.
ESA vs hoard and it was even greater, 10-20x faster than hoard.


Just my 0.02.



Dave Arnette.
(e-mail address removed)

 

[Claudio Puviani]

If speed is of any concern, also look into ESA from Cherrystone
software as well. We (comcast) hired a company to do performance and
scalability testing of various memory managers (because we were just
too darn busy to do it ourselves) and the final result proved to that
ESA from cherrystone was 2x faster than smartheap in the SMP tests.
ESA vs hoard and it was even greater, 10-20x faster than hoard.

It's very possible that those benchmarks may be valid, but without knowing
what methodology was used, it's equally plausible that the benchmarks could
easily be bogus or skewed or too narrowly scoped. More so if we don't know the
credentials of the company that performed them.

However, discounting hearsay, I agree that all options should be evaluated by
the OP.

Claudio Puviani