soellman/copernicus repository overview

⁠Copernicus and Gromacs on CoreOS

This is a work in progress. The eventual goal is to have a simple AWS cloudformation with a single small server, and an auto-scaling group of larger workers.

The next steps are to craft user-data to bring up the single node scenario, then a cloudformation to script the single node-scenario, and then a multi-node setup with EFS as a backing store and scaling policies to reduce the worker pool to 0 on idle.

The cpcc command could be extended to add a new cloud subcommand which can drive the following operations:

1. new-cluster - create a new cluster
2. status - show cluster status (number/size of server/workers)
3. scale - scale workers to desired number
4. download - download all data from cluster
5. shutdown - scale workers to 0 and shut master down
6. revive - start up master
7. terminate - terminate entire cluster, destroying all resources

⁠Status

Milestone one is almost complete.

⁠Proposed Milestones

⁠One

• Cloudformation with single-node and cooperating cpc-server and cpc-worker.
• Drive with shell/python script

⁠Two

• Cloudformation with small master and ASG with workers.
• Allow specification of gromacs/cpc versions
• implement new-cluster, set-cluster, status, terminate using boto

⁠Three

• Add persistent storage and scale-down on idle (pre-requisite: EFS in GA)
• implement scale, shutdown, revive

⁠Unscheduled

• download??
• GPU support
• gromacs release docker builds using jenkins
• detect cpu optimizations to choose best gromacs image

⁠Use it

⁠Create a cluster in AWS

The create-cluster.py script (which has dummy variables) will create a cluster in AWS (us-east-1 for now) consisting of a single node which runs the server and a worker. The script will output its IP address which you can use to login via cpcc.

You will need to have your AWS credentials in the file used by boto (python AWS library): http://boto.readthedocs.org/en/latest/boto_config_tut.html⁠

This essentially fulfills the first milestone.

⁠Run the server/worker

Bring up a CoreOS host, and add the cpc-server.service and cpc-worker.service to /etc/systemd/system and run sudo systemctl daemon-reload. Then start them with sudo systemctl start cpc-server cpc-worker.

⁠Run the client (on the same host)

• docker run --rm -ti --net=host --volumes-from soellman/gromacs copernicus bash
• cpcc add-server -n test localhost
• cpcc use-server test
• cpcc login cpc-admin

⁠Strategies

⁠cluster upgrades

• Don't support them. Really. One cluster has one version of Copernicus and one version of gromacs.
• Maybe a simple cpc/gromacs version update on an EMPTY cluster.

⁠gromacs versions

• one tag set per version
• e.g. 5.1-intel-avx-gpu, 5.1-intel-sse-gpu, 5.1-amd-avx-gpu
• we need some util to detect right version so copernicus can create the right gromacs volume container