fragmentation-aware-packing

1# Fragmentation-Aware Packing
2 
3Use this skill when several feasible placements exist and the choice affects future capacity.
4 
5## Core Idea
6 
7A placement is not good just because it fits. Good placements preserve useful residual capacity. With fractional GPUs, this often means packing small compatible jobs together while preserving whole or scarce GPU slots. The same idea applies to any slots, bins, or resources with discrete capacities.
8 
9## Marginal Fragmentation
10 
11For each feasible placement, compute a local before/after estimate:
12 
131. Measure current free capacity by resource type and slot.
142. Copy the target machine or bin state.
153. Compute `fragmentation_before`.
164. Apply the candidate placement.
175. Compute `fragmentation_after`.
186. Set `marginal_fragmentation = fragmentation_after - fragmentation_before`.
19 
20```text
21best = None
22 
23for placement in feasible_placements:
24  target_before = copy(target_state)
25  fragmentation_before = estimate_fragmentation(target_before, workload_types)
26  target_after = apply(placement, target_before)
27  fragmentation_after = estimate_fragmentation(target_after, workload_types)
28  marginal_fragmentation = fragmentation_after - fragmentation_before
29  score = weighted_action_score(
30    marginal_fragmentation=marginal_fragmentation,
31    other_component_deltas=estimate_other_deltas(placement)
32  )
33  best = lower_score(best, placement, score)
34 
35choose best
36```
37…