[Boost-users] Re: multi_index : performance issue

From: Joaquin M Lopez Munoz (joaquin_at_[hidden])
Date: 2004-12-21 01:51:52

Joaquin M Lopez Munoz <joaquin <at> tid.es> writes:

> 
> Stanislas RENAN <stanislas.renan <at> nsktechnologies.com> writes:
> 
> > 
> > Hi,
> > 
> > I've got a performance problem during a test I've done with multi_index 
> > on MSVC 6.0 SP 6.
> > I'm using BOOST 1.32.0 .
> > The search performance is comparable with several maps, but
> > insertion is dramatically slow : 5mn 30s for 10 000 employees.
> > 
> > Here is my test code :
> > 
> [...] 
> > #if !defined(NDEBUG)
> > #define BOOST_MULTI_INDEX_ENABLE_INVARIANT_CHECKING
> > #define BOOST_MULTI_INDEX_ENABLE_SAFE_MODE
> > #endif
> 

I've taken a closer look at your program, and I think the
comparison is not exactly fair. After populating your
EmployeeMapper, their index sizes are:

m_byId size:10000
m_byName size:10000
m_byAge size:100

while all indices of a multi_index_container have the same
size (in this case, 10000). To balance this, I've changed

employee e(i,           // index numérique unique
           os.str(),    // nom, chaîne unique
           alea(100)); // valeur potentiellement dupliquée

to

employee e(i,           // index numérique unique
           os.str(),    // nom, chaîne unique
           i);          // unique

and got the following results:

Insertion
multi = 47
STL   = 78
Recherche
multi = 234
STL   = 266

If, instead of this change, I allow your EmployeeMapper
to contain duplicate values for the age index by
turning m_byAge into a multimap (and doing the necessary
changes, vg. multimap does not have an operator[]) the
numbers are:

Insertion
multi = 62
STL   = 94
Recherche
multi = 297
STL   = 312

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo