Aligner memory limit vs Clustering memory limit ?

[Sherman-1]

Hi,

I'm currently running Diamond 2.0 blastp for 500.000 queries against the last NR I could find ( 600Millions index ). I encountered a behavior that I don't understand.

When running my blastp command with --block-size 10, I had a warning issue in my log file of this shape :
"Warning: resident size (" << (res_size + last_size) << ") exceeds memory limit."

As I understand it, this warning is triggered when the following condition is met :
if (res_size + last_size > mem_limit)

However, I find intriguing that the mem_limit constant is defined in the align_queries object as following :
const int64_t mem_limit = Util::String::interpret_number(config.memory_limit.get("16G"));

Why is this variable is fixed at 16GB for the user ? I also saw in the config.cpp file that there is a memory_limit variable that can be modified but only for clustering commands. Are these two limits related in any way ?

Last remark, when I run the exact same blastp run but with --block-size 8, the warning is not triggered anymore.
Both runs have 200GB RAM available while my PBS executor estimate a peak consumption of 130GB :

Job Id: 564147.pbsserver
Job_Name = diamd_smple
Job_Owner = simon.herman@node04.example.org
resources_used.cpupercent = 4225
resources_used.cput = 12:25:39
resources_used.mem = 138869908kb
resources_used.ncpus = 75
resources_used.vmem = 45310404kb

[bbuchfink]

You can use the -M option to set a memory limit, but only for the clustering workflow. It does not apply to the alignment workflow and you can ignore that warning.

[Sherman-1]

Thank you very much !