Setting up a Cluster-on-the-Cloud

How to use the cloud?

1.  Get an account on the supported cloud

2.  Start a master instance via the cloud web console or CloudLaunch

3.  Use CloudMan’s web interface on the master instance to manage the platform

Agenda details

•  Launch an instance •  Demonstrate the following CloudMan

features and prepare for the data analysis part: •  Manual & Auto-scaling •  Using an S3 bucket as a data source •  Accessing an instance over ssh •  Customizing an instance •  Submitting an SGE job •  Sharing-an-instance

Inte

rac

tio

n fl

ow

YOUR TURN

Launch an instance 1.  Load biocloudcentral.org

2.  Enter the access key and secret key provided at http://bit.ly/bioworkshop

3.  Type any name as the cluster name

4.  Set any password (and remember it)

5.  Keep Large instance type

6.  Start your instance Wait for the instance to start (~2-3 minutes)

For more details, see wiki. galaxyproject.org/CloudMan

Scaling computation

YES

YES

NO

Workshop plan

•  Launch an instance ✓ •  Demonstrate the following CloudMan

features and prepare for the data analysis part: •  Manual & Auto-scaling •  Using an S3 bucket as a data source •  Accessing an instance over ssh •  Customizing an instance •  Submitting an SGE job •  Sharing-an-instance

Inte

rac

tio

n fl

ow

Manual scaling •  Explicitly add 1 worker node to your cluster

•  Node type corresponds to node processing capacity

•  Research use of Spot instances

Auto-scaling

Auto-scaling in action

Computation time: 9 hrs

Fixed cluster size

5 nodes Computation cost: $20

20 nodes Computation cost: $50

Computation time: 6 hrs

1 to 16 nodes

Computation time: 6 hrs

Dynamic cluster size

Computation cost: $20

Using an S3 bucket as a data source

COMMAND LINE ACCESS

Accessing an instance over ssh

Use the terminal (or install Secure Shell for Chrome)

SSH using user ubuntu and the password you chose when launching an instance:

[local machine]$ ssh ubuntu@<instance IP address>

Once logged in

•  You have full system access to your instance, including sudo; use it as any other system

•  A suite of bioinformatics tools readily installed

•  Can submit jobs to a job manager via the standard qsub command

Distributed Job Manager (DRM)

•  Controls job execution on a (set of) resource(s)

•  A job: an invocation of a program

•  Manages job load

•  Provides ability to monitor system and job status

•  Popular DRMs: Sun Grid Engine (SGE), Portable Batch Scheduler (PBS), TORQUE, Condor, Load Sharing Facility (LSF)

Customize your instance - install a new tool

$ cd /mnt/galaxy/export $ wget http://heanet.dl.sourceforge.net/project/dnaclust/parallel_release_3/dnaclust_linux_release3.zip $ unzip dnaclust_linux_release3.zip $ cd dnaclust_linux_release3 $ chmod +x *

$ cp /mnt/workshop-data/mtDNA.fasta .

Get a copy of a sample dataset Don’t forget

Use the new tool in the cluster mode 1.  Create a new sample shell file to run the tool; call it job_script.sh

with the following content: #$ -cwd ./dnaclust -l -s 0.9 /mnt/workshop-data/mtDNA.fasta

2.  Submit single job to SGE queue qsub job_script.sh

3.  Check the queue: qstat -f 4.  Job output will be in the local directory in file job_script.sh.o# 5.  Start a number of instances of the script:

qsub job_script.sh (*10) watch qstat ‒f

1.  See all jobs lined up 6.  See auto-scaling in (using /cloud) [1.5-2 mins] 7.  Go back to command prmopt, see jobs being distributed

Sharing-an-Instance •  Share the entire CloudMan platform

•  Includes all of user data and even the customizations

•  Publish a self-contained analysis

•  Make a note of the share-string and send it to your neighbor

Management Console

Application(s)(eg, Galaxy)

1°2°

3°

6°, 8°

9°

Persistent data

repositoryStart CloudMan

Setup services

5°

4°

7°

10°

CM-wCM-wCM-w

...

FS 2 FS ...

Instance block storage

Contextualize image

CloudMan MICloudMan MI

...

CloudMan MI

CloudMan machine image

11°

S3/Swift

CloudMan instance

Snap2 Snap..

/mnt/galaxy[Indices]

/mnt/cm/paster.log cm-<hash>

/usr/bin/ec2autorun.log /tmp/cm/cm_boot.log

Troubleshooting

1°

2°

3°

Time for lunch?

Should now have a functional CloudMan instance, configured as Galaxy cluster

•  If not, start one over lunch