Hardware & Network Management | TULPAR — HPC Cluster Management Panel

What's Inside

The Physical Layer, in the Panel

From power to network, GPU to file system — the daily hardware chores run from a browser.

Node power states in one list; on, off, cycle and reset actions execute from the panel.

Serial-over-LAN consoles open from the panel; where a vendor console path isn't verified, the panel won't point you the wrong way.

Define subnets, allocate IPs automatically and release them; provisioning feeds from the same source, so collisions don't happen.

Per-node GPU discovery and DCGM metrics: utilization, memory, temperature, power and ECC counters.

HCA and port status with GUID/rate discovery — fabric data is discovered, not typed in by hand.

NFS, Lustre, GPFS and BeeGFS mount profiles are defined per node category; a node's effective mount set is resolved in the panel.

To manage hardware you first have to know it correctly — TULPAR fills the inventory by discovery instead of condemning you to manual entry.

Node hardware summaries, GPUs and InfiniBand attributes are filled in by discovery; the inventory doesn't have to be maintained by hand.

Node, network and fabric relationships collect into one view; the cluster's layout is read from the panel.

BMC and DCGM targets are constrained by an allowed subnet list — the panel cannot be made to call arbitrary addresses.

Storage profiles and automation rules apply to node categories; 10 nodes and 500 nodes are managed by the same rules.