ADL: Could I run more than one docking at the same time??

Collignon, Barbara C. collignonbc at ornl.gov
Mon Apr 13 14:59:12 PDT 2009


 

>This is not necessarily true. With the advent of new algorithms you
>can efficiently perform docking simulations in seconds.

I was talking about Autodock4.0.
One can estimate the computational time to screen 
*100 millions of compounds 
*flexible ligands
*rigid protein
*only one single CPU of 2.4Ghz

So, if one considers a mean of 30mn per ligand, then the total real time should be several thousands of years.
Even more if we take into account the flexibility of the side chains. 

> With the advent of new algorithms you
>can efficiently perform docking simulations in seconds. Things get
>even better If you have access to multi-core machines, where you can
>perform several individual docking simulations simultaneously

this is what we call "embarrassingly parallel" procedure. I mentioned it in my first sentence.

>To date, it seems meaningless to use thousands of processors for a single docking
>simulation, 

Of course...and that was not my point.

>as usually one can efficiently run a single docking job on
>a single core. 

Of course and this is actually what I mean with the expression: "embarrassingly parallel" procedure.

>This comment will definitively be appropriate for
>computationally intensive algorithms, like molecular dynamics
>simulations of systems with several thousands of atoms.

Fortunately, advances in computational technologies are much beyond of that...

We can simulate systems containing 3.2 millions of atoms using 12000 CPUs (or more of course).  

Also, The MPI version of DOCK6 may reach 92% of efficiency when running 27005 jobs on 2048 processors (1 job per processor).  However, for a number of processors greater than 8192, the MPI version did not scale well due to saturation of the master node and MPI communications. In this case, the High Throughput Computing (HTC) version using the dispatcher scheme is recommended to overcome the problem .  The HTC version consists of running simultaneously non-MPI version of DOCK6 on subsets of a large database using dispatcher programs on service nodes and scheduler/client scripts on computing nodes. 

see http://www.redbooks.ibm.com/redpapers/pdfs/redp4410.pdf

Barbara C. Collignon, Ph.D
Center For Molecular Biophysics
Post-doctoral Research Fellow
UT-Oak Ridge National Laboratory
Building 6011, Mail Stop 6309
Oak Ridge, TN  37831-6309




A couple of comments below.

> people use to perform embarrassingly parallel procedures to perform VHTS since
> 1) using only one single CPU to screen millions of compounds would take thousands of years of calculations.

This is not necessarily true. With the advent of new algorithms you
can efficiently perform docking simulations in seconds. Things get
even better If you have access to multi-core machines, where you can
perform several individual docking simulations simultaneously (this
comment is also meant as a reply to the original message by Mostafa).

> 2) MPI communications affect the scaling when using more than 8000 CPUs (average).

Again, this will depend on the program you are using. To date, it
seems meaningless to use thousands of processors for a single docking
simulation, as usually one can efficiently run a single docking job on
a single core. This comment will definitively be appropriate for
computationally intensive algorithms, like molecular dynamics
simulations of systems with several thousands of atoms.

>
> Barbara
>
> ge-----
> From: autodock-bounces at scripps.edu on behalf of Daniel Jana
> Sent: Mon 4/13/2009 3:17 PM
> To: autodock at scripps.edu
> Subject: Re: ADL: Could I run more than one docking at the same time??
>
> Hello Mostafa,
>
> Mostafa El-Miligy wrote:
>> Could I run more than one docking at the same time without any problem??
>>
>
> Computer programs don't mess up their paperwork as humans do. As long as
> you keep things separate (different files, possibly, different folders)
> things should work fine. Just remember that you don't get much
> performance out of running tons of simulations a single cpu...
>
> Best,
> Daniel
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>
>
> ________________________________________________
> --- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---
>

________________________________________________
--- ADL: AutoDock List  --- http://autodock.scripps.edu/mailing_list ---




More information about the autodock mailing list