Body
Overview:
se-k8s-research is an ephemeral k8s pod service run by S&E IT for ARC researchers. There are no public environments on this cluster. Bespoke compute environments can be created upon request by faculty, and resources will be allocated as-needed. If you think this service may be appropriate for your or your student's research needs, please contact S&E IT for a consultation. There is no charge for compute services; storage needs may be charged if they exceed a reasonable threshold or require backups.
All pods are created programmatically from dockerfiles so that they are both portable and reproducible. All changes to environments are done in these dockerfiles and will only be pushed upon request or hardware migration. If your research needs involve static versions of software/libraries, please let us know during the consult so that we can bake those static requirements into your environment's configuration.
| DNS |
[varies] |
| PORT |
[varies] |
Authentication:
Authentication is generally AD-based (globalID and password), and access is controlled through automation. Non-domain-authed environments can be created if required.
Network Requirements:
- Any domain-bound, ethernet connected computer (with an ssh client installed)
OR
- A virtual machine in the Citrix virtual lab (with an ssh client installed)
Connecting via ssh:
A static DNS and Port will be assigned to your environment at creation time. For example, Andy Faculty (facul1a) might have an environment for his personal research. It might be assigned the DNS se-research-facul1a with a port assignment of 39999.
Assuming an openssh-compatible client is installed and appropriately configured, Andy might connect with the following:
ssh facul1a@se-research-facul1a.se.cmich.edu -p 39999
Most modern operating systems ship with an openssh-compatible client by default. In windows you can use the "ssh" command from a powershell prompt. scp can also be used on the same port to copy data to/from the pods.
Storage:
Home directories are backed up as a best-effort, and are an appropriate place to store very small datasets, scripts, and configuration files. Please limit your home usage to 10GB or less.
/home
Scratch directories are NEVER backed up, and will be allocated upon consultation/request. Scratch directories will generally be in one of the following locations, based on the physical storage locations. This information will be in the ssh motd.
/scratch
/storage/[server_name]/[storage_path]/[globalID]
Datasets with persistence and backups can be mounted to pods, and are generally required to be hosted on se-zfs. Some departments have already bought into this storage cluster, while others would need to buy-in in order to have backed up persistent storage. When mounted, they will generally follow the following convention and will also be in the ssh motd.
/storage/[server_name]/[storage_path]/[globalID]
/storage/[server_name]/[storage_path]/[project_name]
Compute:
The cluster currently has 11 nodes at the time of writing; at any given time, ~5 are typically idle and waiting for work. We can generally accommodate workloads of ~32 thread, 128GB of ram indefinitely, with up to 144 thread, 2TB of ram, allocated upon consultation request.