standalone jemalloc is slower than glibc's malloc (ptmalloc)

Ricardo Nabinger Sanchez rnsanchez at wait4.org
Thu Apr 12 06:04:02 PDT 2012


Hello Amol,

On Wed, 11 Apr 2012 17:01:40 +0530
amol pise <amolpise15 at gmail.com> wrote:

> I have check the performance of the jemalloc vs glibc's malloc
> (ptmalloc) using complex test in libxml2 package
> i.e. "runtest". This program allocate many memory of different size
> and supported multi-thread.
> 
> ...
> 
> Is it mean jemalloc() is slower than glibc's malloc ?
> Is there any way to improve the speed performance of jemalloc in this regard?

A few questions:

1- Is jemalloc hot?  Glibc should be.  First iteration for jemalloc
   will be cold, hitting startup latencies.  If that is the case, the
   measurement was not fair.

2- Do the results hold?  Ie, repeat 10 ~ 30 times each of them,
   check min/avg/max and stdev, while ensuring fair testing conditions.

3- Do those tests (as performed in libxml2) match whatever scenario you
   expect to hit when using jemalloc?  If they don't match or are too
   far apart, libxml2 will tell you nothing.

Cheers


-- 
Ricardo Nabinger Sanchez           http://rnsanchez.wait4.org/
  "Left to themselves, things tend to go from bad to worse."



More information about the jemalloc-discuss mailing list