What is GARLI, and how can I use it on Quarry?

On this page:

• About GARLI
• Choosing Garlijob or ParallelGarlijob
• Preparing to use GARLI
• Using Garlijob for serial jobs
• Using ParallelGarlijob for parallel jobs

About GARLI

The GARLI package conducts phylogenetic searches on aligned sequences using maximum likelihood criteria. It supports codon- and amino-acid-based models in addition to standard nucleotide models. The current version of GARLI on Quarry is 0.96b8. Garlijob is the customized script for using the serial version of Garli. ParallelGarlijob is the customized script for using the parallel version of Garli.

For information about GARLI packages, see the GARLI home page, GARLI FAQ, and GARLI manual (also available in PDF format). At Indiana University, an example data set is available on Quarry at /N/soft/linux-rhel4-x86_64/Garli/example.

Choosing Garlijob or ParallelGarlijob

Both Garlijob and ParallelGarlijob can set up the working environment, write a script for TORQUE (also called PBS), and submit the script to run as a batch job that requires more than 20 minutes to finish. Here are few things to consider in deciding which script to use:

• Garlijob submits serial jobs; ParallelGarlijob submits parallel jobs.
  
  
• ParallelGarlijob divides a number of search or bootstrap replicates to multiple CPUS. Each individual search or bootstrap replicate won't be faster or better, but the time to complete all the searches will be shorter.
  
  
• Garlijob is suitable for submitting a few different serial GARLI jobs simultaneously. ParallelGarlijob won't let the same user handle multiple jobs at the same time. ParallelGarlijob requires using garli.conf as the default configuration name; this name cannot be changed.
  
  
• Although all the output results will be in the same working directory where the configuration and input files are, the output of the parallel version of GARLI will appear in a number of result files (just as the result of using Garlijob several times would). To understand how to deal with the results, see the GARLI advanced topics page.

Preparing to use GARLI

1. If you don't already have a Quarry account, request one through the Account Management System (AMS); see At IU, if I already have some computing accounts, how do I get others? and Getting started on Quarry.
   
   
2. Set up SoftEnv by entering softadd +garli and then resoft at the command prompt.
   
   
3. Create a working directory that contains all the data files (in .nex format ) and a parameter file named garli.conf. All the parameters that GARLI takes are stored by default in garli.conf. To learn how to write your garli.conf file, see the Garli manual (PDF) or the GARLI Configuration Settings page.

If your job will run for fewer than 20 minutes, call GARLI in the directory where your input files are. If your job will take longer than 20 minutes, use the Garlijob or ParallelGarlijob script (see the next section).

The output GARLI generates will be in the same directory as your input files. Note that if you did not pass a configure file on the command line, the program will look in the current directory for a file named garli.conf.

Using Garlijob for serial jobs

To learn how to use Garlijob, enter garlijob at the prompt. You will see the following information:

The first option, configure_file_of_Garli, is necessary.

The output GARLI generates will be in the same directory as your input files. Note that if you did not pass a configure file on the command line, the program will look in the current directory for a file named garli.conf.

Using ParallelGarlijob for parallel jobs

To learn how to use ParallelGarlijob, enter parallelgarlijob at the prompt. You will see the following information:

ParallelGarlijob requires garli.conf as the default name of the configuration file. It also requires that you specify how many times to repeatedly use the garli.conf file. Do this by supplying an integer X to the -NumRepeatGarliConf option and the number of CPUS (Y) to use. Therefore, the number of search replicates or bootstrap replicates is X*Y.

Following are a few examples:

The first command is a serial job running on a single CPU for an hour. If the bootstrap is 20 times, the second command will run on 10 CPUs for 2 hours (2x10=20), and the third command will run on 4 CPUs for 5 hours (4x5=20).

For more information, see Configuring and submitting a run on the GARLI MPI version page.

Comments/Questions/Corrections