[ace-users] ACE vs/with Intel TBB ?

Ruslan Zasukhin sunshine at public.kherson.ua
Mon Jul 30 02:35:39 CDT 2007


On 30/7/07 10:02 AM, "Johnny Willemsen" <jwillemsen at remedy.nl> wrote:

Hi Johnny,

>> I wonder, how ACE framework is going collaborate with Intel TBB ?
> 
> We as Remedy IT recently started a project together with Intel to optimize
> ACE/TAO for Intel CPU's. We are using Intel C++ and are going to use Intel
> Vtune to tune ACE/TAO and get a better performance on Intel hardware. As a
> result of this project there are already a few new Intel C++ 10.0 builds on
> the scoreboard running on new Intel servers, more will follow. We are using
> IA32 and EM64T Intel CPU's. We have now direct access into the Intel
> organization to gain access to a lot of knowledge. We are looking for
> additional funding to add IA64 and do more extensive optimizations, if
> people are interested in this effort contact me.

Great.
 
> What is your idea about the Intel TBB? Do you have specific ideas that you
> want to see supported in ACE?

Just after reading of TBB description, they underline that this is
high-level framework, which

** hides low level OS-threads instead offering to developer high-level
concepts as Task. 

** They offer algorithms for parallel execution, e.g. Sort()
    Cool?  I have not see such things in ACE.
    I will be happy to get quick-sort which works on 8-32 CPUs :-)

** They offer containers with not single mutex but with lock per item.

---------------
I do not see problems why ACE cannot use the same ideas to become better
where it is possible. I think developers of ACE should be self interested at
least take a look on sources/design/docs of TBB.

As I see ACE is more cross-platform that Intel's TBB.
So unlikely we will switch from ACE to TBB in near time.

-- 
Best regards,

Ruslan Zasukhin
VP Engineering and New Technology
Paradigma Software, Inc

Valentina - Joining Worlds of Information
http://www.paradigmasoft.com

[I feel the need: the need for speed]




More information about the Ace-users mailing list