Skip to content

Latest commit

 

History

History
180 lines (137 loc) · 7.12 KB

workshop.md

File metadata and controls

180 lines (137 loc) · 7.12 KB
Raw

AWS HPC Workshop

First, log in to your Console and see your Instances

Under "Services", click "EC2". You should see this:

Initial EC2 Instances

Take note of the instances that you have at the beginning of the exercise.

Next, get to your console through Cloud9

  1. Click "Services" and search for Cloud9

Cloud9 menu

  1. Open the IDE

Cloud9 IDE

Note: You Must enable third party cookies in your browser for this to work properly

Your screen will look like this:

Cloud9 IDE

Set in to your Python Virtual Environment

source ~/pcluster-env/bin/activate

Your command prompt should now look like this:

(pcluster-env) ec2-user:~ $

Now Create Your Cluster

pcluster create mycluster

This operation will take anywhere from 10-13 minutes.

ssh into your cluster master to submit jobs

(pcluster-env) ec2-user:~ $ pcluster ssh mycluster

Last login: Fri Oct  4 18:44:57 2019

       __|  __|_  )
       _|  (     /   Amazon Linux AMI
      ___|\___|___|

https://aws.amazon.com/amazon-linux-ami/2018.03-release-notes/
12 package(s) needed for security, out of 20 available
Run "sudo yum update" to apply all updates.
[ec2-user@ip-10-0-1-152 ~]$

Go to the /shared/AWS-HPC-Workshop directory

[ec2-user@ip-10-0-1-152 ~]$ cd /shared/AWS-HPC-Workshop

Run the submit.sh command to submit 10 job.sh jobs

[ec2-user@ip-10-0-1-152 ~]$ ./submit.sh job.sh 10

You should see this:

Your job 1 ("job.sh") has been submitted
Your job 2 ("job.sh") has been submitted
Your job 3 ("job.sh") has been submitted
Your job 4 ("job.sh”) has been submitted
Your job 5 ("job.sh") has been submitted
Your job 6 ("job.sh") has been submitted
Your job 7 ("job.sh") has been submitted
Your job 9 ("job.sh") has been submitted
Your job 10 ("job.sh") has been submitted

See your jobs in queue

job-ID  prior   name       user         state submit/start at     queue                          slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
      1 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      2 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      3 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      4 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      5 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      6 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      7 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      8 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
      9 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1        
     10 0.00000 job.sh     ec2-user     qw    10/01/2019 15:56:05                                    1

Go back to the AWS Console and click EC2 and then Instances to see your intances spinning up

Navigate from "Services" -> "EC2" -> "Instances"

Instances coming up

When your jobs are done, see the output files in your home directory:

[ec2-user@ip-10-0-1-66 ~]$ cd
[ec2-user@ip-10-0-1-66 ~]$ ls -l
total 40
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e1
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e10
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e2
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e3
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e4
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e5
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e6
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e7
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e8
-rw-r--r-- 1 ec2-user ec2-user  0 Oct  1 16:00 job.sh.e9
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o1
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o10
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o2
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o3
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o4
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o5
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o6
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o7
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o8
-rw-r--r-- 1 ec2-user ec2-user 35 Oct  1 16:00 job.sh.o9

You can see the output and that jobs ran on different hosts

[ec2-user@ip-10-0-1-66 ~]$ cat job.sh.o1
This job executed on ip-10-0-1-125


[ec2-user@ip-10-0-1-66 ~]$ cat job.sh.o2
This job executed on ip-10-0-1-231

Now, let's submit an MPI run

[ec2-user@ip-10-0-1-66 AWS-HPC-Workshop]$ cd /shared/AWS-HPC-Workshop/
[ec2-user@ip-10-0-1-66 AWS-HPC-Workshop]$ qsub mpirun.sh
Your job 11 ("mpirun.sh") has been submitted


[ec2-user@ip-10-0-1-66 AWS-HPC-Workshop]$ qstat
job-ID  prior   name       user         state submit/start at     queue                          slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
     11 0.00000 mpirun.sh  ec2-user     qw    10/01/2019 16:14:29                                    1

Now, watch a large-ish file get written:

[ec2-user@ip-10-0-1-224 AWS-HPC-Workshop]$ cd /shared/AWS-HPC-Workshop/

[ec2-user@ip-10-0-1-224 AWS-HPC-Workshop]$ ./dd.sh /fsx/dd.out 4096
4275044352 bytes (4.3 GB) copied, 22.068620 s, 194 MB/s
4096+0 records in
4096+0 records out
4294967296 bytes (4.3 GB) copied, 22.1729 s, 194 MB/s

Notice the speed of the writes to the FSx for Lustre partition. FSx for Lustre is a managed Lustre filesystem that scales with the size of the partition you create. The filesystem really excels at high speed scratch space. And, just like scratch space on-premise, it is not as durable.

Finally, see how FSx for Lustre can write back data to S3 for durable storage

Start by going to the console, selecting S3 and looking at your S3 bucket

S3 Bucket

Use the provided script to write back the contents of your /fsx directory to S3

[ec2-user@ip-10-0-1-224 AWS-HPC-Workshop]$ ls -al /fsx
total 4133638
drwxrwxrwt  3 root     root          25088 Oct  1 19:23 .
dr-xr-xr-x 28 root     root           4096 Oct  1 19:07 ..
-rw-rw-r--  1 ec2-user ec2-user 4294967296 Oct  1 19:23 dd.out
-rwxr-xr-x  1 root     root        2263040 Sep 24 19:15 ior-3.2.1.tar.gz
[ec2-user@ip-10-0-1-224 AWS-HPC-Workshop]$ ./archive.sh
[ec2-user@ip-10-0-1-224 AWS-HPC-Workshop]$ nohup: ignoring input and redirecting stderr to stdout

Now go back to your S3 bucket and look for the "workshop" folder

S3 Bucket

That's it! You can now log out of your lab account