Hi all!
I installed the new version of gtdb v1.3.0 and database release 95! I tried to run the gtdb program with 4 genomes (but I have 130 genomes 
) and I received the following messagge:
WARNING: pplacer requires ~152 GB of RAM to fully load the bacterial tree into memory. However, 131.97GB was detected. This may affect pplacer performance, or fail if there is insufficient scratch space.
I would like to know the impact of this error in my analysis
Thank you!!
Hello,
Because your server has less than the required RAM (131/152GB) , GTDB-Tk will need to use scratch space. If program cant access the scratch space, it will crash and not return any results for the classify step.
Having less than the recommended amount of RAM doesn’t affect the accuracy of the tool but instead increases the risk it crashing.
To reduce the memory usage, you can use the --scratch_dir flag ( see https://ecogenomics.github.io/GTDBTk/commands/index.html ) .
Hope this helps.
Regards,
Pierre
Hi all,
If I have a server with 120GB of RAM whether GTDB will work on it, as pplacer requires ~152GB of RAM.
Please let me know how should I run GTDB in this case.
Thank you
Hi @p.chaumeil I have the same Problem using GTDB-Tk as part of the metabolic pipeline ( METABOLIC Usage · AnantharamanLab/METABOLIC Wiki (github.com). Would you please elaborate more the --scratch_dir flag? Thank you in advance
Hello @Kalonji_Abondance ,
The --scratch_dir flag is used to create a directory where pplacer will write a mmap-file ( by default gtdbtk.pplacer.scratch). This is equivalent to mmap-file in pplacer. From the test we have done so far on our servers, the difference of speed ,with or without scratch-dir, is minimal.
From the pplacer documentation:
In cases when there isn’t enough memory for pplacer to use for internal nodes, or it’s otherwise disadvantageous to use physical memory, it’s possible to instead tell pplacer to mmap a file instead using the --mmap-file flag. This will, very roughly, perform disk IO instead of using physical memory.