PATTERN Cited by 1 source
Weighted-voting tier selection¶
Pattern¶
To select the optimal upper-tier data center for a cloud region, treat each IP subnet within that region as a voter. Each subnet's vote is weighted by the strength of its current upper-tier assignment (derived from continuous latency probing). The upper tier that accumulates the highest weighted vote total becomes the region's primary.
Mechanism (Cloudflare)¶
- Cloud provider IP-range files fetched every few hours (AWS
ip-ranges.json, GCP, Azure, Oracle equivalents). - Each file maps cloud regions → IP prefixes (subnets).
- The upper-tier database (built from latency probes refreshed every 15 minutes) assigns each known subnet to its current best upper tier.
- For a given cloud region, each matching subnet contributes a weighted vote toward its assigned upper tier.
- The upper tier with the strongest aggregate signal wins → becomes the region's primary.
Why Voting (Not Single Probe)¶
- A single subnet might have noisy probe data or be newly assigned.
- Aggregating across all subnets in a region provides statistical robustness.
- The weighting accounts for probe confidence / data freshness.
- Regions with many active subnets get strong consensus; sparse regions fall back to concepts/geographic-fallback.
Properties¶
- Self-updating: as cloud providers add/remove/reassign subnets, the vote pool changes automatically.
- Continuous: no one-time assignment — the election runs every probe cycle.
- Graceful degradation: insufficient votes → geographic fallback (not a wrong answer).
Seen in¶
- sources/2026-07-10-cloudflare-improving-smart-tiered-cache-for-public-cloud-regions — canonical description of the voting algorithm for Smart Tiered Cache