raxml-ng fails to finish with --search

[Jeff Bowman]

Hello,

I'm building trees using coarse-grained parallelization with --search, where I specify two physical cores for each tree to be built (36 threads total).  This works fine for most of my alignments but something strange is happening at the end of one taxon-rich alignment.  The log file only reflects output for 17 of the 18 trees requested, and all 36 cores are being used by raxml-ng.  The log file hasn't been updated after 48 hours, but the lastTree.TMP file is being rewritten over and over (it is the same each time).  Any thoughts on what's going on here would be appreciated.  I've attached my log file and am happy to provide the alignment or other files if this is helpful.

Thanks,

Jeff

[Alexey Kozlov]

Hi Jeff,

it seems that this last tree search simply takes much longer to converge than the rest. This is not
unusual on datasets with poor signal (many taxa + few sites). So please let it run a bit longer.

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>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/raxml/42d2ccd2-99a8-4637-b371-3fdc3482d776n%40googlegroups.com
> <https://groups.google.com/d/msgid/raxml/42d2ccd2-99a8-4637-b371-3fdc3482d776n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[bowm...@gmail.com]

Thanks Alexey,
This was my initial thought as well, but there are two things that seem inconsistent with this. 1) The log file hadn't updated even after 48 hours of additional run time, and 2) if raxml-ng were simply waiting for the final tree to converge, shouldn't it have only been running 1 worker and two threads (i.e. cpu = 200 % not 3600 %)? Let me know what I'm missing.

Cheers,
Jeff

You received this message because you are subscribed to a topic in the Google Groups "raxml" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/raxml/W8l4oIqgVRo/unsubscribe.
To unsubscribe from this group and all its topics, send an email to raxml+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/raxml/3d9e67a2-b404-82bd-d11b-c8b51d1ed405%40gmail.com.

[Alexey Kozlov]

Hi Jeff,

I know it could be confusing, but actually both things you observed are expected:

- to make log file more readable, detailed search progress is only reported for the worker #0

- raxml-ng uses busy waiting, that's why you see full CPU utilization even if only 1 worker is active

Best,
Alexey

[Jeff Bowman]

Excellent, thanks!  I was wondering about the logfile output...

Jeff

To view this discussion on the web visit https://groups.google.com/d/msgid/raxml/8eb0b8ea-3fc8-389b-0f6a-07717b8726e9%40gmail.com.