RAxML 8 running slow?

[Olaf R.P. Bininda-Emonds]

Hi Alexis,

We've had some discussion about this before, but I do think that RAxML 8 is running slow for me in general and definitely slower than it ought to be.

As a test, I gave RAxML v7.2.8 and v.8.2.9 the exact same matrix to chew on: 728 taxa and 16491 bp divided among 33 partitions and with no gaps other than those needed for the alignment (i.e., all 33 genes are present for all 728 taxa). After about eight hours, SSE-PTHREADS with T = 12 has completed 33 of 500 fast bootstrap replicates for v7 and one-third LESS (only 24!) for v8, both compiled on the same machine in the same way. Each node on the cluster does have 12 CPUs and each set of 12 is dedicated to the single task.

I know that PTHREADS is tricky sometimes depending on the number of partitions (18 of which here are really small: < 100 bp) and distinct patterns, but should it really be this slow and should v8 really be that much slower than v7?!

Best,

Olaf

[Alexey Kozlov]

Hi Olaf,

just my 2 cents: I don't know the reason for the performance discrepancy you are observing, but I'm pretty sure that you
can get much better runtimes if you use MPI version of RAxML. RAxML-MPI parallelizes across BS replicates (and not
alignment patterns), which will be more efficient on a dataset like this.

Best,
Alexey

> --
> You received this message because you are subscribed to the Google Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

[Olaf R.P. Bininda-Emonds]

Thanks Alexey,

Have never used the MPI version, but I might give it a shot after / if these analyses finish.

PTHREADS is definitely needed for this one though because of the number of alignment patterns: 14000+. Interestingly, however, the two versions also disagree as to the number of alignment patterns: 14342 for v7 and 14331 for v8, but I assume that this might have been a bug that was squashed in the meantime?

Cheers,

Olaf

---

[Alexey Kozlov]

Hi Olaf,

> PTHREADS is definitely needed for this one though because of the number of alignment patterns: 14000+. Interestingly,
> however, the two versions also disagree as to the number of alignment patterns: 14342 for v7 and 14331 for v8, but I
> assume that this might have been a bug that was squashed in the meantime?

most probably, and since version 7 is *really* old, I wouldn't recommend using it even if it's faster :)

Best,
Alexey

> <javascript:>
> > <mailto:raxml+un...@googlegroups.com <javascript:>>.
> > For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.

[Alexandros Stamatakis]

Dear Olaf,

There have been so many and partially very important bug fixes since
v7.2.8 that you should just not use 7.2.8 any more.

As a consequence of those fixes RAxML might indeed have become slower
for certain parameter combinations and dataset types.

Alexis

On 29.11.2016 20:35, Olaf R.P. Bininda-Emonds wrote:

> --
> You received this message because you are subscribed to the Google
> Groups "raxml" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to raxml+un...@googlegroups.com
> <mailto:raxml+un...@googlegroups.com>.
> For more options, visit https://groups.google.com/d/optout.

--
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

[Olaf R.P. Bininda-Emonds]

Hi Alexis,

It was mostly just curiosity on my part because I had the feeling that RAxML 8 was running slower than I remembered 7 to be. I'm just surprised that it is in fact slower because you tend to tweak the speed of RAxML with each release and am especially surprised by just how much slower it is!

Any chance of re-introducing a few bugs in the next version to speed it up again? ;)

Best,

Olaf

---

[Alexandros Stamatakis]

hi olaf,

> It was mostly just curiosity on my part because I had the feeling that
> RAxML 8 was running slower than I remembered 7 to be. I'm just surprised
> that it is in fact slower because you tend to tweak the speed of RAxML
> with each release and am especially surprised by just how much slower it is!

well, just wait for the release of the complete raxml re-design that
will be faster again ...

> Any chance of re-introducing a few bugs in the next version to speed it
> up again? ;)

easy, we can just have it return random trees, that will be fast :-) and
be in line with the garbage in garbage out strategy

alexis

> > an email to raxml+un...@googlegroups.com <javascript:>
> > <mailto:raxml+un...@googlegroups.com <javascript:>>.

> > For more options, visit https://groups.google.com/d/optout

> <https://groups.google.com/d/optout>.

>
> --
> 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 <http://www.exelixis-lab.org>

[George Pacheco]

Hej Alexis, 

Sorry for coming with so many question recently, but I was just wondering if you would have any estimation of when RaxML will be running fast again (due to whatever reason it got slow). I have also noticed a huge difference regarding the running time, which I fear will make my analyse unpractical considering my limited timeframe.

Thanks very much in advance, George.

[Alexandros Stamatakis]

Dear George,

The completely re-designed RAxML will be released for sure in January.

All the best,

Alexis

[George Pacheco]

Thanks, Alexis!

Your best, George.