Small typo: 99.95% for regional. It's similar to AWS with a 99.9% for EKS clusters. Also not sure what GKE's new SLA reimbursement structure will be going forwards.
I really enjoyed that write up. The focus on a good UI/UX in platform teams is definitely something that resonates with me. I also really liked the article placing an emphasis on Platform teams providing value by having a better "glue" I thought was interesting. I hadn't really thought that glue could be a component where value could be added. I really like the pattern of tools like "Devhose" and in developing a tool specifically to be the glue that integrates well with other tools.
What I've done before is use branch-buildID (develop-009) for tags, and then add detailed information like the git commit sha, git version tags, as labels. All of these labels were passed in as build args to Docker from the CI/CD system. One other important thing to consider is if you have docker tags that can be overwritten, which is the default in most repos, in the docker file it is useful to specify FROM sha256 checksum instead of labels for reproducible builds.
Yes, MSMBuilder is software package that uses them explicitly in protein dynamics simulations. The rationale behind it is it allows you to take trajectories from molecular dynamics simulations, and construct a markov state model out of the conformational state space of the protein. This is extremely useful in MM simulations, because it allows for the conformational space to be sampled by many smaller trajectories, instead of one longer trajectory where it is highly unlikely to sample the full state space anyways(gets stuck in local minima). This allows for molecular dynamics simulations to be done as High Thruoughput computing instead of HPC. Its the rationale behind Folding@home.
Input data could come from the protein database, which has catalogs of thousands of protein structures. The issue being is when is a protein folded? Do we have some known ending structure and we just need to get within an acceptable Root Mean Square Deviation? The other issue would be there need to be some sort of centralized server used to collect the resulting data, and create the Markov State Model.