We have deployed an open ondemand instance on a physical machine. At any given time, we usually have 50-100 interactive apps (Jupyter, RStudio server, matlab, etc.) running. How many users a single ondemand instance can support? Is the CPU, memory or network the potential bottleneck?
We have another cluster and the current open ondemand instance cannot submit jobs to this cluster. This cluster majorly supports traditional HPC users, and we are considering building an open ondemand instance for this cluster. It is likely that these users will run noVNC sessions through open ondemand. We are considering two options. 1) Upgrade the current open ondemand hosting machine so that we will use the current open ondemand instance to support both clusters. 2) Build a new open ondemand instance and use this new open ondemand to support these traditional HPC users.
This has more to do with the hardware that has all gone into you cluster and there is no easy or fast answer to this because it can depend on:
Server Capacity: The hardware of the server on which OOD is running will be the main factor. This includes the number of CPU cores, the amount of RAM, and the server’s network bandwidth.
User Activity: The type of activities users are performing in their interactive sessions can greatly influence the number of supported users. For instance, data-intensive jobs or jobs requiring high computation power will limit the number of concurrent users.
Cluster Configuration: OOD itself is a relatively lightweight web-based portal that interfaces with your High Performance Computing (HPC) resources. The number of users that can be effectively supported is primarily determined by the resources available on your HPC clusters that OOD submits jobs to.
This would also mean any bottleneck is largely going to depend again on the hardware and configuration chosen.