Usage Dashboard
2 minute read
The usage dashboard at dashboard.daic.tudelft.nl shows how DAIC has been used over time, based on the SLURM accounting database. Log in with your TU Delft account.
What the numbers mean
The dashboard reports allocated resources: CPUs, GPUs, and memory that SLURM reserved for jobs, multiplied by the time the jobs ran. Allocation is not measured hardware load - a job that reserves 8 cores and uses one still counts as 8 allocated cores. The exception is the Efficiency section, which compares consumed resources with allocated ones where SLURM records them (consumed core-time and peak memory; GPU usage is not recorded by SLURM accounting).
All charts are computed for the date range selected in the sidebar. Filters for partition, account, QoS, and job state narrow every chart; the “Color By” selector splits the charts by that dimension.
Sections
| Section | Content |
|---|---|
| Users, Jobs | Active users and submitted jobs per period |
| CPU / GPU / Memory Usage | Allocated resource-hours per period and their distribution |
| Efficiency | Consumed over allocated resources per period. Memory efficiency is weighted by job runtime and based on peak usage, so it is an upper bound |
| Usage by Node | Resource-hours per node. With “Normalize %” enabled, the percentage of each node’s capacity, plus cluster-wide allocation gauges |
| Timing | Waiting times and job durations: trends, distributions per period, and histograms |
| Allocated Resources | CPUs, GPUs, memory, and nodes requested per job |
Reading the node charts
Node values come from jobs overlapping the selected range: hours outside the range are cut off, and a multi-node job is split equally over its nodes. With “Normalize %” enabled, each node is shown relative to its capacity as reported by SLURM; nodes whose capacity is unknown are listed separately. The gauges weight every node by its capacity, so idle nodes lower them.
Checking your own usage
For a single job, seff <jobid> and sacct -j <jobid> on the cluster give per-job numbers (see Checking Jobs). The dashboard is the aggregate view: pick a date range and filter by your account to see your group’s allocation, waiting times, and efficiency.
If the memory efficiency of your group is low, jobs request more memory than they use; requesting less lets more jobs run per node. See the reservation guidance in Policies & Usage Guidelines.