You are here

cluster.linuxgccrelease Killed error

2 posts / 0 new
Last post
cluster.linuxgccrelease Killed error

Dear Rosetta's users,

I'd run my score analysis employing the score options below [1]. But after some minutes the programs gave me this error message: killed. As I'm interested to cluster my 50 000 structures generated by Rosetta3.5, I wonder to know what I'm doing wrong with this cluster analysis.

Kind regards.

[1] cluster.linuxgccrelease -database /home/guest/rosetta/rosetta_database -in:file:fullatom -in:file:silent bamil2_50k.out -run:shuffle -cluster:radius 10 -cluster:sort_groups_by_energy -out:file:silent 2bamil_50k.cluster.10.out -silent_read_through_errors

Post Situation: 
Fri, 2015-11-20 05:56

The message "Killed" normally comes from the operating system, rather than Rosetta itself. Normally it means that the program exceeded some resource restriction of the machine and the operating system automatically killed the process in order to keep the system from going down.

The most common reason for a "killed" is running out of memory. Cluster is a notorious memory hog - it has to be, to some extent because it's operating on 50 000 structures and has to keep a lot of data in memory. There's various utilities which will allow you to look at the memory consumption of processes as they run. You may want to monitor the memory consumption as things run, and see if that's why things error out. 

Aside from simply running on a machine with more memory, one way around this is to reduce the number of structures you're clustering. A large number of those 50 000 structures you're clustering are probably poor scoring. You could filter the silent file for those structures which pass some energy cutoff, and then do clustering with many fewer structures, which should help memory usage.

The other alternative is to use another program to do clustering, one which might have a more memory efficient algorithm for clustering. Many people in the Rosetta community like Calibur. The drawback in using an external program is that you would need to output PDBs for each structure in the silent file, as Calibur and other non-Rosetta clustering tools can't read silent files.

Tue, 2015-11-24 16:07