BOINCAMI

How to Create an AMI to Run BOINC

Why Run BOINC In the Cloud

Consumer PCs are not designed for power-hungry, memory-intensive scientific computing applications. Server hardware is. Even with the performance penalty for virtualization, it’s much better to have numbers being crunched in a data center than on your laptop; you won’t drive up your electric bill and incur needless wear and tear on your everyday electronics.

Launch An Instance Using the Starcluster AMI

Start an instance with one of the public Starcluster AMIs, since they already have many drivers and applications pre-installed for HPC (BLAS,CUDA,etc). Choose an HVM AMI, since these will include GPU support (should you ever want to use a GPU-based instance type). I chose Ubuntu 12.04 (starcluster-base-ubuntu-12.04-x86_64-hvm; ami-52a0c53b). Avoid Ubuntu 13.04, as support for that from Canonical is spotty.

Install A Lightweight GUI and Remote Desktop Server

There are some operations for BOINC that you can accomplish from the command line (boinccmd). Other operations (such as changing resources allocated to BOINC) are not possible or are difficult. A lot of BOINC’s design seems to be centered upon the user interacting with the client via the graphical BOINC manager.

To get around this, install a lightweight GUI like lxde and a remote desktop server like x2go.

Create an Attachable/Detachable Volume

Make a new volume in the AWS console. Attach to the instance. Partition, create a file system and mount at a suitable location (like ‘/boinc’).

Install BOINC

Use apt-get or yum to install BOINC client and BOINC manager. Move BOINC client’s data from /var/lib/boinc-client to the new volume you created (mounted at ‘/boinc’ or wherever you decided to mount it). Edit /etc/default/boinc-client and change BOINC_DIR from “/var/lib/boinc-client” to the new mount point (‘/boinc’).

Separating the data from the instance’s root volume allows the tasks that were currently running to be preserved in the case of a poweroff (i.e., if you are using spot instances and the instance is abruptly terminated).

Save the Instance as an AMI

Remove the bash history, ssh directory in your home folder. Save as an EBS-backed AMI in the AWS console.

When Running

When you start a new instance with the AMI, don’t forget to a) mount the attached volume and b) start the BOINC client service with sudo /etc/init.d/boinc-client start.

With Spot Instances

Make a snapshot of the BOINC data volume beforehand, since you don’t know which AZ the instance will pop up in. Make new volume from snapshot in the instance’s AZ accordingly.


Scientific-Computing