UberCloud HPC Experiment
UberCloud is best known for the HPC Experiment, the free, community driven effort that explores the end-to-end processes of accessing and using remote computing resources in HPC centers and HPC clouds, on demand. Participants of the experiment get hands on experience on using remote computing resources in HPC centers and in the Cloud for their own projects; in the process learning how to cope with the challenges and resolve many of the common roadblocks. The project teams working on the end-users applications discuss their experience, the hurdles they ran into, their lessons learned, and the best practices, on the UberCloud community web site.
The UberCloud HPC Experiment was started in July 2012, and within a year has grown to about a thousand organizations and individuals from over 60 countries. More than 120 teams have been formed since then and many of these teams are covered on the UberCloud community web site. UberCloud HPC Experiment offers free trials for up to 1,000 CPU core hours. We have the experience of having handled over 120 such projects. You can review the schedule of upcoming HPC Experiment “rounds” here.
Top 10 Recommendations to Conduct a Successful Experiment
After 4 rounds with 124 team experiments we have collected a few must-read recommendations on how to perform a successful experiment, and how to avoid some severe pitfalls. You as a participant want to follow this etiquette, to appreciate the precious time of your fellow team members.
1. Consider that this is just an experiment, and not a production project. Nobody wants free experiments to compete with commercial service providers and thus jeopardize their business.
2. As the end-user in your team, select an application example which you know is manageable within a 4-6 weeks long experiment project.
3. Select a 6-8 week time window for your experiment where you know you will be able to spend at least 2-4 hours per week on average on this experiment.
4. Prepare for the experiment by defining the main goals and describing the needed hardware, software and expertise.
5. Once your experiment project is launched, you must complete the required team documents on the UberCloud Basecamp collaboration platform (Project Definition, Key Contacts, Computing Resources, and Software Resources).
6. All preparatory work such as testing of geometry meshes, boundary and initial conditions, physical parameters, etc., should be done on your in-house system. Only when the test problem runs smoothly on your system it should be moved to a remote system for larger runs with more resources (up to a max of 1000 cpu core hours).
7. While thinking about the set-up of the experiment, also start thinking about your next steps. What comes after the experiment if it concludes successfully? Would you be ready for including cloud computing as an option for additional computing needs? If pricing is reasonable, would you consider running production jobs in the cloud? Will you be ready for Requesting a Quote on the UberCloud Marketplace from selected service providers?
8. As a service provider, provide your services in time, for example access to resources, software, and expertise shouldn’t take longer than a few days. We don’t ask for Service Level Agreements for the experiments, but consider the opportunity that the end-user might become your customer in the near future.
9. Follow up with your team at least once a week. If you can’t be available, announce this to your team well ahead of time.
10. And finally, whenever your team faces a roadblock, contact immediately your Team Mentor who will help to resolve your roadblock.