GROMACS version: any
GROMACS modification: No
Here post your question
Hi Team!
Not sure this is a right place for my message please correct me if I’m wrong.
I would like to ask Gromacs society how they consider cloud (on-line) service to execute Gromacs calculations?
Main features would be like this:
Ability to execute any Gromacs task, any Gromacs version
Execution on nodes with several GPUs (1080Ti)
Task management system - GUI for user tasks with task statuses, completion percentage, etc.
Ability to estimate execution time in advance for free
Ability to specify any Gromacs command parameter
Submit archive with user data and run your own bash script for data transformation (if needed)
We already did some tests of mdrun on 6-12 GPUs and it shows very good performance (version 2019.5 was used).
I am a Gromacs user. I think using Gromacs as a cloud service would be useful, but there are some caveats. Mainly the price. Many research groups have access to their own cluster or a high-performance supercomputer at academic prices, which often means for free. The cost of owned clusters is also not real. I had to buy the hardware from my research project, but I don’t pay the electricity it uses. Access to supercomputers is not trivial, there are often competitive calls, but again if your proposal is accepted you get lots of computing hours.
Having said that, there are situations when I would pay to get more computing power for a given time lapse. Some people have explored using Amazon cloud computing (AWS) so you should also consider how you deal with that (see this rather old post http://fowlerlab.org/2016/01/17/gromacs-on-aws-performance-and-cost/) . There are even some companies offering these services with a different MD engine (see https://www.acellera.com/cloud-molecular-dynamics-simulation-acecloud/). These are old posts. Maybe this has been discontinued for reasons I ignore.
In a nutshell, the cloud services are interesting, but it all boils down to the conditions of use.