RAxML bootstrap takes too many time independently of the cores

[Lucía Moreyra]

Hi everyone,

I'm running a bootstrap analysis in Ubuntu using RAxML pthreads option. I've tried to run this analysis with 12 threads and 34 threads and the average time for each bootstrap is around 8 hours for both. I really don't understand why if I put more cores the time is not reduced. 

Can you help me figuring out what is happening? I'm just what to run the bootstrap since I already have the best tree.

The command line is the following:  

-raxmlHPC-PTHREADS-AVX -f d -p 12345 -b 12345 -N 100 -m GTRGAMMA -s concatenated70_75.phylip -n Cirsium202205bs100 -o Alfredia-acantholepis_L04P94H10S147 -q partitions.txt -T 34

The matrix has around 300.000 nucleotide and 332 species with a partition of 1050 genes.

The computer has a core i9, 18 cores/36 threads, 256gb RAM and Ubuntu 20.

I check the System Monitor and it shows all the cores working and using around 10gb of RAM memory. I don't think there is something wrong with this.

Also, do you think 8 hours per bootstrap makes sense for my matrix? 

I don't know if I can speed up the process changing some option.

Thank you in advance!

Lucía.

[Pfeiffer, Wayne]

Hi Lucia,

I see two problems.

1. RAxML does not take advantage of hyper-threading, so you should not specify more threads than you have physical cores.

2. The lack of speedup beyond 12 cores is because your data set has so many partitions. In fact, 12 cores might not even be much faster than 8 cores.

You should be able to speed up your analysis by making two simultaneous runs of 50 bootstraps using 8 threads each and starting from different seeds, say 12345 and 23456. This can also be accomplished using the hybrid parallel version of RAxML, but that requires MPI.

Good luck!

Wayne

--
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.
To view this discussion on the web visit https://groups.google.com/d/msgid/raxml/36546d94-e461-4895-9f03-558a4083a55dn%40googlegroups.com.

[Lucía Moreyra]

Thank you Wayne!

I understand now what was probably going on. I'm running the analysis as you said, hopefully, i'll work!

I'll tell you later how it's going :)

Lucia.

[Alexey Kozlov]

Hi Lucia,

even better solution: please switch to raxml-ng which has much more efficient parallelization for
partitioned datasets, and can automatically parallelize across bootstrap replicates whenever needed:

https://github.com/amkozlov/raxml-ng

Best,
Alexey

>> <https://urldefense.proofpoint.com/v2/url?u=https-3A__groups.google.com_d_msgid_raxml_36546d94-2De461-2D4895-2D9f03-2D558a4083a55dn-2540googlegroups.com-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dfooter&d=DwMFaQ&c=-35OiAkTchMrZOngvJPOeA&r=QDz-wiRlY9zE5rxxvqHCevtBQwPGExdv-9IJhNPslhk&m=KtEMIwGyj9p8Ede1dncpqDTc4Ufa8caFD4wW9svCKv76Io8cFeBQfk1EFyO16Qkw&s=HjwOFZkbMkETdN1944CCst67RWWYvY4-XiXyNzFu-xg&e=>.

>
> --
> 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>.

> To view this discussion on the web visit

> https://groups.google.com/d/msgid/raxml/804ead2b-a9bd-4ce1-bba7-22790bfc90dan%40googlegroups.com
> <https://groups.google.com/d/msgid/raxml/804ead2b-a9bd-4ce1-bba7-22790bfc90dan%40googlegroups.com?utm_medium=email&utm_source=footer>.