EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Heliotrope: Quorum- and SLO-Aware Adaptive Leader Placement for Raft-Based Geo-Distributed Databases

Victor Phillips, Serena Tran

International Journal of Information and Communication Technology Trends · 20.10.2026 · 6 (2) · 343-355

DOI: 10.66525/ijictt146 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Information and Communication Technology Trends, online erschienen am 20.10.2026. Verfasst von Victor Phillips und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In Raft-based geo-distributed databases, every write pays the round trip from the client to the leader of its Raft group plus the round trip from the leader to the slowest member of the fastest majority, so the region that hosts each leader largely determines user-visible latency. Deployed schedulers move leaders toward the region that issues most writes; this rule ignores quorum geometry, reacts to noisy short windows, and treats leader transfers, which stall a group and are rate-limited in practice, as free. We present Heliotrope, a fleet-level leader placement controller for Multi-Raft deployments. Heliotrope computes the exact commit-latency distribution of every leader candidate from measured round-trip-time distributions with a Poisson-binomial order-statistic recursion, scores candidates by mean latency plus an SLO-excess term whose fleet-wide aggregate is linear in demand, estimates near-term regional demand by shrinking the latest observation toward a seasonal prior, and assigns the leaders of all groups jointly under leader-load capacity while pricing every transfer with a dual variable that tracks a transfer budget. We evaluate Heliotrope with measured AWS and GCP inter-region RTTs and a geo-temporal workload derived from 838,178 human-authored commits to 440 public repositories over 13 weeks. At a budget of two leader transfers per group per day, Heliotrope lowers mean write latency by 2.2% and the share of writes slower than 200 ms by 18.4% relative to the best budget-matched baseline, wins on every one of the 63 test days, and has the fewest overloaded epochs and the lowest write latency of all evaluated policies when leader capacity is tight. A message-level replay of Raft confirms the analytic latency model to within 0.7 ms.

Zitieren

APA

Phillips, V., & Tran, S. (2026). Heliotrope: Quorum- and SLO-Aware Adaptive Leader Placement for Raft-Based Geo-Distributed Databases. International Journal of Information and Communication Technology Trends, 6(2), 343-355. https://doi.org/10.66525/ijictt146

Vancouver

Phillips V, Tran S. Heliotrope: Quorum- and SLO-Aware Adaptive Leader Placement for Raft-Based Geo-Distributed Databases. International Journal of Information and Communication Technology Trends. 2026;6(2):343-355. doi:10.66525/ijictt146

BibTeX
@article{Phillips2026,
  author = {Phillips, Victor and Tran, Serena},
  title = {Heliotrope: Quorum- and SLO-Aware Adaptive Leader Placement for Raft-Based Geo-Distributed Databases},
  journal = {International Journal of Information and Communication Technology Trends},
  year = {2026},
  volume = {6},
  number = {2},
  pages = {343--355},
  doi = {10.66525/ijictt146}
}
Kostenlos starten →