[phenixbb] Autobuild requires more then 24 hrs to run on 32 cpu?

Thomas C. Terwilliger terwilliger at lanl.gov
Mon Jun 29 17:33:10 PDT 2009


Hi Tanya,

Autobuild can take quite a while for a large structure such yours, yes.

The command you set up makes 5 models and merges them to make one good
composite model.  Each of the  5 models is made by running 5 cycles of
rebuild-in-place, which rebuilds the entire model in overlapping 
6-residue segments.

This particular process is parallelized only up to the number of models,
so your 32 processors unfortunately don't help you; you are using only 5
of them and getting only 5-fold improvement over one processor.

This run seems like it was going fine and was 95% of the way done:

AutoBuild_run_1_1_/TEMP0/AutoBuild_run_1_N.log ends with
>>
>> *Copying  refined_pdb_in.pdb  to  composite_model.pdb  in
>> AutoBuild_run_1_/TEMP0
>> Beginning merge with current best model:  refined_pdb_in.pdb
>> Merging models: composite_model.pdb refine_1.pdb
>> Refining model:  Build_composite_3.pdb
>> Model: AutoBuild_run_1_/TEMP0/Build_composite_refined_3.pdb
>> R/Rfree=0.23/0.25

This run doesn't look good:

>> or with error *
>> TRIED resolve_extra_huge ...but not OK
>>
>> ********************************************************************************
>> Failed to carry out AutoBuild_build_cycle:
>>
>>

but my guess is that if you use a recent "nightly build" of phenix
(available on the web site) the error may go away or at least it *should*
give you a more informative error message if not.

You can speed up your job by saying "n_cycle_rebuild_max=1" or =2 (instead
of =5); that will be 5x or 2.5x faster.  I am guessing your job could take
about a day on 5 processors (just more than you had), depending on
resolution.

All the best,
Tom T

>> Hello!
>>
>> I am trying to use Autobuild wizard and it seems that it takes longer time
>> that I expected.
>> That's why I have next questions:
>>
>> 1. What is "usual" time range estimate for Autobuild run with default
>> parameters if my model has 7 chains with 284 aa in each?
>> (if I'm running Autobuild on a single machine it does not finish in 3 days
>> or on cluster with usage of 32 processors - in 24 hours)
>>
>> 2. Do I use phenix comman correctly for using several machines at once?
>>
>> I used /phenix-1.4-89, command line
>> *phenix.autobuild  data=e239a.hkl seq_file=seq.dat model=model.pdb
>> nproc=32*
>>
>>
>> In 24 hrs, which is my limit on cluster time, I get next things -
>> AutoBuild_run_1_1_/AutoBuild_run_1_1.log ends with
>>
>> *Setting up to build 5 models to be combined into final model #1
>> NOTE: input model will be rebuilt in place: nothing will be added or
>> deleted.
>> Building  5  models ...
>> Running up to  32  jobs in parallel...
>> Splitting work into 5 jobs and running 32 at a time with csh in
>> /home2/tas354/phenix_runs/20090610/1/AutoBuild_run_1_/TEMP0
>>
>> Starting job 1...
>> Starting job 2...
>> Starting job 3...
>> Starting job 4...
>> Starting job 5...*
>>
>> AutoBuild_run_1_1_/TEMP0/AutoBuild_run_1_N.log ends with
>>
>> *Copying  refined_pdb_in.pdb  to  composite_model.pdb  in
>> AutoBuild_run_1_/TEMP0
>> Beginning merge with current best model:  refined_pdb_in.pdb
>> Merging models: composite_model.pdb refine_1.pdb
>> Refining model:  Build_composite_3.pdb
>> Model: AutoBuild_run_1_/TEMP0/Build_composite_refined_3.pdb
>> R/Rfree=0.23/0.25
>>
>> Rebuilding side chains from  Build_composite_refined_3.pdb*
>>
>> or with error *
>> TRIED resolve_extra_huge ...but not OK
>>
>> ********************************************************************************
>> Failed to carry out AutoBuild_build_cycle:
>>
>>
>> ********************************************************************************
>> *
>>
>> Should I change parameters of the Autobuild which will dramatically reduce
>> time for computations or should I ask for extension of my time for letting
>> the programs to run as it is?
>>
>> I would give more details if it is needed I just could not think of some
>> other right now. And I would appreciate any advices and suggestions!
>>
>>
>> Thank you
>> Tanya
>> _______________________________________________
>> phenixbb mailing list
>> phenixbb at phenix-online.org
>> http://www.phenix-online.org/mailman/listinfo/phenixbb
>>




More information about the phenixbb mailing list