Re: [Boost-users] multi_index_container

From: Joaquín Mª López Muñoz (joaquin_at_[hidden])
Date: 2006-02-16 06:54:46

Hi Aashit,

Aashit Soni ha escrito:

> Hi all,
>
> I am using multi_index_container on MSVC6, WinXP. I ported my existing
> Hash and order list to multi_index_container. which is working quite
> expected? a pure vanila hash_index with ordered_index
>
> I am facing some problem in fine tuning the hash_index which is slower
> compared to my previous implementation.

Slower in which aspect? Insertion, deletion, lookup?

> I am looking for different
> ways how can i fine tune the hash_index in multi_index_container?
>
> specifically i am yet to discover,
> How can i set the bucket size for the hash container?

You cannot control bucket size explicitly. Hashed indices occupancy
level is controlled thru a parameter called load factor (lf) defined as:

  lf=size()/bucket_count()

In general, the lower lf, the faster lookup operations are. You can set
the maximum value of lf to be allowed with max_load_factor():

float mlf=c.max_load_factor(); // query max lf
c.max_load_factor(0.8f); // set max lf

When the load factor reaches the specified maximum, the number of
buckets is automatically increased.
By default max_load_factor()=1.0. Lower this for faster lookup,
at the expense of more memory usage (more buckets.)

> How can configure the container for maximum store it can reserve for
> me in one shot?

Use rehash(), documented at

http://boost.org/libs/multi_index/doc/reference/hash_indices.html#hash_policy

prior to mass insertion of elements. For instance:

// insert 1000000 elements. With lf=1.0, we will need 1000000 buckets
// so as to be under the specified max load factor
c.rehash(1000000);
c.insert(first,last);

Doing a prior rehash is not necessary, but will save you some
intermediate automatic rehashings, pretty much the same way as
it happens with std::vector and reserve.

> any other optimization parameter.
>

Other than this, your best option is to choose a good hashing function.
The default is expected to work well except for exceptional
situations. You might want to use the hashing function used in your
previous Hash container so as to check out whether it does a
better job than Boost.Hash, although I'd say this is unlikely.

> If any one has some idea. It will be of great help to me.
>
> regards,
> aashit

Hope this helps a little,

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo