Cloud cost
Bin packing
How tightly a scheduler packs pods onto the fewest nodes, given their requests. Loose packing strands capacity as cluster idle across many half-empty nodes.
Last updated
Definition
Bin packing describes how densely a scheduler places workloads onto nodes. Suppose the scheduler, or a cluster-autoscaler or Karpenter configuration, spreads pods thin across many nodes. Each node then keeps unused capacity that nothing uses. That is cluster idle, multiplied by every node that runs below a useful packing density.
Packing density interacts with node shape. The same total requested capacity can run on many small, loosely packed nodes or on fewer, larger, tightly packed ones. The second option usually strands far less capacity for each dollar of node cost.
Where it shows up
Bin packing shows up as node count and per-node utilization in a cluster dashboard. The same total requested capacity can run on many lightly loaded nodes or on fewer, tightly packed ones. A cluster-autoscaler or Karpenter configuration file sets the packing behavior directly. It also sets the node shapes the autoscaler may choose.
What makes it expensive
An autoscaler set to spread pods thin, not consolidate them, strands schedulable capacity on every node it starts. The cluster looks busy while you pay for a lot of room that nothing asked for. Total requested capacity can look reasonable while average node utilization is low. The mismatch between those two numbers is the whole problem.
Related
More in Cloud cost
- Committed use discount (CUD)
- Reserved instance (RI)
- Savings plan (SP)
- On-demand vs. spot instances
- Data egress
- Amortized, blended, and unblended cost
- Billed vs. effective vs. list vs. contracted cost
- Rightsizing
- Idle and orphaned resources
- Storage class
- Lifecycle policy
- Incremental snapshot
- Allocatable capacity
- Requests vs. limits
- Cluster idle
- Workload idle
- Cost allocation tags
- Tag coverage
- Unallocated spend
- Showback vs. chargeback
- Unit economics
- RI/SP coverage and utilization
- Observability tax
- Metric cardinality
- High-water-mark billing
- Log ingest vs. index
- Snowflake credit
- Auto-suspend / resume minimum
- BigQuery slot
- Bytes scanned & partition pruning
- Time Travel & Fail-safe
The meter you don't watch still grows.
Waste rarely hides in the service you check every week. CostMon watches every meter on your stack, including the ones you forgot were running.