Many interoperability problems ◦ Various system has a various environment ▪ Cannot resolve complex library dependencies ▪ Cannot run program on other operating system ◦ On HPC: each site has a different type of the batch job system • This can be solved by ◦ Multi-sites federated Kubernetes clusters using container technology
providers such as AWS, GCP, and Azure though they has a great experience on deployment of the cloud • AWS Tokyo regional catastrophe on Friday afternoon, August 23, 2019 ◦ duration: ~10 hours • GCP Google Cloud Storage Incident on 2019-03-13 ◦ duration: ~4 hours  https://publickey1.jp/blog/19/aws23.html  https://www.publickey1.jp/blog/19/sregmail313google.html
environment ▪ Popular container runtime: Docker, rkt ◦ Users can create containerized program including libraries it depends ◦ Users can run it on any environment Cloud On-premise HPC Open Container Runtime (Docker, rkt) Ubuntu CentOS Go Python Fortran C
Open source & initially developed by Google ◦ Can run any containerized application automatically • Provided features ◦ Service discovery ◦ Storage orchestration ◦ Secret & config management ◦ Batch execution Source: Least Privilege Container Orchestration - Docker Blog https://www.docker.com/blog/least-privilege-container-orchestration/
about federation between cloud and grid computing • However, it can be enhanced with Kubernetes HPC On-prem ise HPC Cloud Cloud  Buyya, Rajkumar, and Rajiv Ranjan. "Federated resource management in grid and cloud computing systems." Future Generation Computer Systems 26.8 (2010): 1189-1191.
on various environments: ◦ Cloud: for flexible on-demand computing and storage ◦ On-premise: for cost-efficient/special hardwares ◦ HPC: for the large-scale compute-intensive programs HPC On-prem ise HPC Cloud Cloud