StarCluster

Back to the beginning: GotCloud

Back to GotCloud: Amazon

If you have access to your own cluster, your task will be much simpler. Install the GotCloud software (GotCloud: Source Releases) and run it as described on the same pages.

For those who are not so lucky to have access to a cluster, Amazon Web Services (AWS) provides an alternative. You may run the gotcloud software on a cluster created in AWS. One tool that makes the creation of a cluster of AMIs (Amazon Machine Instances) is StarCluster (see http://star.mit.edu/cluster/).

The following shows an example of how you might use StarCluster to create an AWS cluster and set it up to run GotCloud. There are many details setting up starcluster and this is not intended to explain all of the many variations you might choose, but should provide you a working example.

Getting Started With StarCluster

StarCluster provides lots of documentation (http://star.mit.edu/cluster/) which will provide more information on it than we have here.

To install and setup StarCluster for the first time, you can follow the QuickStart instructions: http://star.mit.edu/cluster/docs/latest/quickstart.html

• Includes installation instructions
  • See http://star.mit.edu/cluster/docs/latest/installation.html for more detailed StarCluster installation instructions if the QuickStart instructions are not enought (especially if running on Windows)
• Includes setting up a basic StarCluster configuration file
  • You will need your AWS Credentials to setup the configuration file
    • If you need help setting up your AWS credentials, see: AWS Credentials

You can skip actually starting the cluster in the QuickStart instructions if you want.

Troubleshooting: When I tried this, the starcluster start mycluster step failed similar to:

• http://star.mit.edu/cluster/mlarchives/2425.html
• So I followed the suggestions there and at https://github.com/jtriley/StarCluster/issues/455:

  1. $ sudo pip uninstall boto

  2. $ sudo easy_install boto==2.32.0

     • I was having trouble with pip install, but found that easy_install worked
• I had to force terminate mycluster after a failed start:

  •  starcluster terminate -f mycluster

• Then I was able to successfully start my cluster

Don't forget to terminate your cluster:

starcluster terminate mycluster

StarCluster and GotCloud

StarCluster Config Settings

By default, StarCluster expects a configuration file in ~/.starcluster/config.

• StarCluster will create a model file for you

Ensure your StarCluster configuration file is set for your usage.

• General AWS Settings:

[aws info]
aws_access_key_id = #your aws access key id here
aws_secret_access_key = #your secret aws access key here
aws_user_id = #your 12-digit aws user id here

• You should have set these in Getting Started With StarCluster above (quickstart guide and AWS Credentials) .

• GotCloud Cluster Definition
  • You may want to create a new cluster section for running GotCloud (or you can use smallcluster) in your configuration file: ~/.starcluster/config
  • You can call it anything you want, for example,

    gccluster

  • Example:

    [cluster gccluster]
    KEYNAME = mykey
    CLUSTER_SIZE = 4
    CLUSTER_USER = sgeadmin
    CLUSTER_SHELL = bash
    MASTER_IMAGE_ID = ami-6ae65e02
    NODE_IMAGE_ID = ami-3393a45a
    NODE_INSTANCE_TYPE = m3.large

    • Set KEYNAME to the key you want to use
    • Set CLUSTER_SIZE to the number of nodes you want to start up (this may be different than 4)
    • Set CLUSTER_USER to add additional users, like sgeadmin
    • Set CLUSTER_SHELL to define the shell you want to use, like bash
    • Set MASTER_IMAGE_ID to the latest GotCloud AMI, see: GotCloud: AMIs
      • Contains GotCloud, the reference, and the demo files in the /home/ubuntu/ directory that will be visible on all nodes in the cluster
      • Has a 30G volume, but only 6G available
    • Set NODE_IMAGE_ID to a StarCluster ubuntu x86_64 AMI
      • Since each node does not need its own 30G volume containing GotCloud, the reference, and the demo files, we use a separate image for the nodes.
    • The nodes can just access the master's copy of GotCloud, the reference, and the Amazon demo
    • Set NODE_INSTANCE_TYPE to the type of instances you want to start in your cluster
      • See http://aws.amazon.com/ec2/pricing/ for instance descriptions and prices
      • We do not recommend running GotCloud on machines with less than 4MB of memory
  • The CLUSTER_SIZE * CPUs in NODE_INSTANCE_TYPE = the number of jobs you can run concurrently in GotCloud

• Define Data Volumes
  • By default, the GotCloud AMI contains about 5G of extra space that you can use
    • /home/ubuntu/ directory is visible from all machines
      • Use /home/ubuntu/ for the output directory if it is <5G
      • This directory will be deleted when you terminate the AMI
  • Create your Own Volumes and attach them to the GotCloud cluster
    • Instructions TBD

Starting the Cluster

1. Start the cluster:

   • starcluster start -c gccluster mycluster

     • Alternatively, if you can the default template at the start of the configuration file in the [global] section to gccluster: DEFAULT_TEMPLATE=gccluster, you can run:

       • starcluster start mycluster

   • It will take a few minutes for the cluster to start

Copying Data to/from the Cluster

Copy data onto the cluster (command run from your local machine)

starcluster put /path/to/local/file/or/dir /remote/path/

Pull the data from the cluster onto your local machine (command run from your local machine)

starcluster get /path/to/remote/file/or/dir /local/path/

Reminder, if you write your output to /home/ubuntu/, it will be deleted when you terminate the cluster

Running GotCloud on StarCluster

• If you have not already, logon to the cluster as ubuntu:

  • starcluster sshmaster -u ubuntu mycluster

    • Type yes if the terminal asks if you want to continue connecting
• When running GotCloud:
  • Set the cluster/batch type in either configuration or on the command line:
    • In Configuration:

      BATCH_TYPE = sgei

    • On the command-line:

      --batchtype sgei

  • Set the number of jobs to run:

    --numjobs #

    • Replace number with the number of concurrent jobs you want to run (probably CLUSTER_SIZE * CPUs in NODE_INSTANCE_TYPE)
  • Otherwise, run GotCloud as you normally would.

To login to a specific non-master node, do:

starcluster sshnode -u ubuntu mycluster node001

Monitoring Cluster Usage

• Monitor jobs in the queue

  • qstat

  • This will show you how the currently running jobs and how they are spread across the nodes in your cluster

    

    • state descriptions:
      • qw : queued and waiting (not yet assigned to a node)
      • r : running
• View Sun Grid Engine Load

  • qhost

    

    • ARCH : architecture
    • NCPU : number of CPUs
    • LOAD : current load
    • MEMTOT : total memory
    • MEMUSE : memory in use
    • SWAPTO : swap space
    • SWAPUS : swap space in use
• View the average load per node using:

  • qstat -f

    

    • load_avg field contains the load average for each node

Terminate the Cluster

1. Reminder, check if you need to copy any data off of the cluster that will be deleted upon termination
   • Copying Data to/from the Cluster
2. Terminate the cluster

   • starcluster terminate mycluster

Run GotCloud Demo Using StarCluster

1. Create a new cluster section in your configuration file: ~/.starcluster/config
   • Add the following to the end of the configuration file:

     [cluster gccluster]
     KEYNAME = mykey
     CLUSTER_SIZE = 4
     CLUSTER_USER = sgeadmin
     CLUSTER_SHELL = bash
     MASTER_IMAGE_ID = ami-6ae65e02
     NODE_IMAGE_ID = ami-3393a45a
     NODE_INSTANCE_TYPE = m3.large

2. Start the cluster:

   • starcluster start -c gccluster mycluster

     • Alternatively, if you can the default template at the start of the configuration file in the [global] section to gccluster: DEFAULT_TEMPLATE=gccluster, you can run:

       • starcluster start mycluster

   • It will take a few minutes for the cluster to start
3. Logon to the cluster as ubuntu:

   • starcluster sshmaster -u ubuntu mycluster

     • Type yes if the terminal asks if you want to continue connecting

Examine the Setup

1. After logging into the Amazon node as the ubuntu user, you should by default be in the ubuntu home directory: /home/ubuntu
   1. You can check this by doing:

      pwd

      • This should output: /home/ubuntu
   2. Take a look at the contents of the ubuntu user home directory

      ls

      • This should output be 2 directories, example and gotcloud
        • The example directory contains the files for this demo
        • The gotcloud directory contains the gotcloud programs and pre-compiled source

   

2. Look at the example input files:

   ls example

   

   1. bam.list contains the list of BAM files per sample
   2. bams is a subdirectory containing the BAM files for this demo
   3. test.bed contains the region we want to process in this demo
      • To make the demo run faster, we only want to process a small region of chromosome 22. This file tells GotCloud the region. The region we are using is the APOL1 region

      

   4. test.conf contains the settings we want GotCloud to use for this run

      

      • For the demo, we want to tell GotCloud:
        1. The list of bams to use: BAM_LIST = example/bam.list
        2. The region to process rather than the whole genome: UNIFORM_TARGET_BED = example/test.bed
        3. The chromosomes to process. The default chromosomes are 1-22 & X, but we only want to process chromosome 22: CHRS = 22

Run GotCloud SnpCall Demo

1. Run GotCloud snpcall

   • gotcloud snpcall --conf example/test.conf --outdir output --numjobs 8 --batchtype sgei

     • The ubuntu user is setup to have the gotcloud program and tools in its path, so you can just type the program name and it will be found
     • There is enough space in /home/ubuntu to put the Demo output
       • /home/ubuntu is visible from all nodes in the cluster
     • This will take a few minutes to run.
     • GotCloud first generates a makefile, and then runs the makefile
     • After a while GotCloud snpcall will print some messages to the screen. This is expected and ok.
2. See Monitoring Cluster Usage if you are interested in monitoring the cluster usage as GotCloud runs
3. When complete, GotCloud snpcall will indicate success/failure
   • Look at the snpcall results, see: GotCloud: Amazon Demo -> Examining SnpCall Output

Run GotCloud Indel Demo

1. Run GotCloud indel

   • gotcloud indel --conf example/test.conf --outdir output --numjobs 8 --batchtype sgei

     • The ubuntu user is setup to have the gotcloud program and tools in its path, so you can just type the program name and it will be found
     • There is enough space in /home/ubuntu to put the Demo output
       • /home/ubuntu is visible from all nodes in the cluster
     • This will take a few minutes to run.
2. See Monitoring Cluster Usage if you are interested in monitoring the cluster usage as GotCloud runs
3. When complete, GotCloud indel will indicate success/failure
   • Look at the indel results, see: GotCloud: Amazon Demo -> Examining Indel Output

Terminate the Demo Cluster

1. Exit out of your master node

   • exit

2. Terminate the cluster
   • Since this is just a demo, we don't have to worry about the data getting deleted upon termination

   • starcluster terminate mycluster

     • Answer y to the questions Terminate EBS cluster mycluster (y/n)?

Old Instructions

StarCluster Configuration Example

StarCluster creates a model configuration file in ~/.starcluster/config and you are instructed to edit this and set the correct values for the variables. Here is a highly simplified example of a config file that should work. Please note there are many things you might want to choose, so craft the config file with care. You'll need to specify nodes with 4GB of memory (type m1.medium) and make sure each node has access to the input and output data for the step being run.

####################################
## StarCluster Configuration File ##
####################################
[global]
DEFAULT_TEMPLATE=myexample

#############################################
## AWS Credentials Settings
#############################################
[aws info]
AWS_ACCESS_KEY_ID = AKImyexample8FHJJF2Q
AWS_SECRET_ACCESS_KEY = fthis_was_my_example_secretMqkMIkJjFCIGf
AWS_USER_ID=199998888709 

AWS_REGION_NAME = us-east-1                 # Choose your own region
AWS_REGION_HOST = ec2.us-east-1.amazonaws.com
AWS_S3_HOST = s3-us-east-1.amazonaws.com

###########################
## EC2 Keypairs
###########################
[key east1_starcluster]
KEY_LOCATION = ~/.ssh/AWS/east1_starcluster_key.rsa   # Same region

###########################################
## Define Cluster
##   starcluster start -c east1_starcluster  nameichose4cluster
###########################################
[cluster myexample]          # Name of this cluster definition
KEYNAME = east1_starcluster                 # Name of keys I need
CLUSTER_SIZE = 4                            # Number of nodes
CLUSTER_SHELL = bash

#  Choose the base AMI using   starcluster listpublic
#   (http://star.mit.edu/cluster/docs/0.93.3/faq.html)
NODE_IMAGE_ID = ami-765b3e1f
AVAILABILITY_ZONE = us-east-1               # Region again!
NODE_INSTANCE_TYPE = m1.medium              # 4G memory is the minimum for GotCloud

VOLUMES = gotcloud, mydata
[volume mydata]
VOLUME_ID = vol-6e729657
MOUNT_PATH = /mydata

[volume gotcloud]
VOLUME_ID = vol-56071570
MOUNT_PATH = /gotcloud

Create Your Cluster

 starcluster start -c myexample myseq-example
 StarCluster - (http://web.mit.edu/starcluster) (v. 0.93.3)
 Software Tools for Academics and Researchers (STAR)
 Please submit bug reports to starcluster@mit.edu

 >>> Validating cluster template settings...
 >>> Cluster template settings are valid
 >>> Starting cluster...
     [lines deleted]
 >>> Mounting EBS volume vol-32273514 on /gotcloud...
 >>> Mounting EBS volume vol-36788522 on /mydata...
     [lines deleted]

When this completes, you are ready to run the GotCloud software on your data. Make sure you have defined and mounted volumes for your sequence data and the output steps of the aligner and umake. These volumes (as well as /gotcloud) should be available on each node.

 starcluster sshmaster myseq-example
 StarCluster - (http://web.mit.edu/starcluster) (v. 0.93.3)
 Software Tools for Academics and Researchers (STAR)
   [lines deleted]

 df -h
 ssh node001 df -h

If your data is visible on each node, you're ready to run the software as described in GotCloud.