drastically different execution times between two runs (different machines; different seeds)

[pavlos]

Dear raxml community,

perhaps this is totally expected, but it looks strange at least to me... 

Yesterday, I started to run a pthreads job in a 24-core node with the followng report:

----------------------------------------------------------------------------------------------------------------------------

RAxML was called as follows:

raxmlHPC-PTHREADS-SSE3 -T 24 -f a -m GTRCAT -p 8989 -x 8493 -N 200 -s ./ji.filtered750.fa.MAFFT -n T2 

Time for BS model parameter optimization 3180.488848

----------------------------------------------------------------------------------------------------------------------------

Nothing else has been produced till yesterday, not even a single bootstrap replicate. 

Today, I just tried it in a different machine with 8 cores with other random number seeds. This is not a faster machine actually. 

-------------------------------------------------------------------------------------------------------------------------------------------------

RAxML was called as follows:

raxmlHPC-PTHREADS-SSE3 -T 8 -f a -m GTRCAT -p 889 -x 1123 -N 200 -s ./ji.filtered750.fa.MAFFT -n T2 

Time for BS model parameter optimization 9.029420

Bootstrap[0]: Time 142.424704 seconds, bootstrap likelihood -153386.327202, best rearrangement setting 9

Bootstrap[1]: Time 225.864484 seconds, bootstrap likelihood -140702.999807, best rearrangement setting 14

-------------------------------------------------------------------------------------------------------------------------------------------------

In less than 10 mins it has already produced two bootstrap results. Also the Time for BS model parameter optimization is drastically different 9 vs 3180. 

Is this something expected (due to random seeds for example)?

best

pavlos

[Alexandros Stamatakis]

looks like a parallel slowdown, how many site patterns does the
alignment have?

alexis

On 02.11.2014 09:00, pavlos wrote:
> Dear raxml community,
>
> perhaps this is totally expected, but it looks strange at least to me...
>
> Yesterday, I started to run a pthreads job in a 24-core node with the
> followng report:
> ----------------------------------------------------------------------------------------------------------------------------
> RAxML was called as follows:
>
> raxmlHPC-PTHREADS-SSE3 -T 24 -f a -m GTRCAT -p 8989 -x 8493 -N 200 -s
> ./ji.filtered750.fa.MAFFT -n T2
>
> Time for BS model parameter optimization 3180.488848
> ----------------------------------------------------------------------------------------------------------------------------
> Nothing else has been produced till yesterday, not even a single bootstrap
> replicate.
>
>
>
> Today, I just tried it in a different machine with 8 cores with other
> random number seeds. This is not a faster machine actually.
> -------------------------------------------------------------------------------------------------------------------------------------------------

> *RAxML was called as follows:*
>
> *raxmlHPC-PTHREADS-SSE3 -T 8 -f a -m GTRCAT -p 889 -x 1123 -N 200 -s
> ./ji.filtered750.fa.MAFFT -n T2 *
>
>
>
> *Time for BS model parameter optimization 9.029420*
> *Bootstrap[0]: Time 142.424704 seconds, bootstrap likelihood
> -153386.327202, best rearrangement setting 9*
> *Bootstrap[1]: Time 225.864484 seconds, bootstrap likelihood
> -140702.999807, best rearrangement setting 14*

> -------------------------------------------------------------------------------------------------------------------------------------------------
> In less than 10 mins it has already produced two bootstrap results. Also
> the Time for BS model parameter optimization is drastically different 9 vs
> 3180.
>
> Is this something expected (due to random seeds for example)?
>
> best
> pavlos
>
>
>

--
Alexandros (Alexis) Stamatakis

Research Group Leader, Heidelberg Institute for Theoretical Studies
Full Professor, Dept. of Informatics, Karlsruhe Institute of Technology
Adjunct Professor, Dept. of Ecology and Evolutionary Biology, University
of Arizona at Tucson

www.exelixis-lab.org

[pavlos]

Alexi kalimera,

here is the report from info:

Executing 200 rapid bootstrap inferences and thereafter a thorough ML search 

All free model parameters will be estimated by RAxML

ML estimate of 25 per site rate categories

Likelihood of final tree will be evaluated and optimized under GAMMA

GAMMA Model parameters will be estimated up to an accuracy of 0.1000000000 Log Likelihood units

Partition: 0

Alignment Patterns: 839

Name: No Name Provided

DataType: DNA

Substitution Matrix: GTR

Sequences are not very large (a bit less than 800bp) but there are many of them (about 2000 I think)

best

pavlos

[Alexandros Stamatakis]

kalimera pavlo,

that explains it the alignment is much too short to be run on 24 cores
with only 800 patterns, you should at most use two cores for running
this dataset, otherwise raxml will not scale,

alexis

[pavlos]

Thanks a lot!