Phase-Aware Dual-Signal Allocation forPredictive Cold-Start Mitigation in Serverless Computing
International Journal of Information and Communication Technology Trends · 19.10.2026 · 6 (2) · 356-367
DOI: 10.66525/ijictt145 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Information and Communication Technology Trends, online erschienen am 19.10.2026. Verfasst von James Lee und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Warm pools mitigate Function-as-a-Service cold starts, but equal treatment of warm hits ignores heterogeneity in startup cost. We present Phase-Aware Dual-Signal Allocation (PADA), a lightweight, causal scheduler that fuses exponential recent-activity smoothing with previous-day seasonal activity and ranks candidate warm instances by predicted miss cost per reserved megabyte. A three-phase startup decomposition makes the scheduling objective sensitive to infrastructure, runtime, and application initialization. We conduct a reproducible minute-granularity evaluation on six independently generated synthetic workload fleets comprising 144 functions each, using three memory budgets, seven comparators or ablations plus a reactive reference, and a separate readiness-lag and retention-stability stress test. At an 8,192-MB cap, PADA achieves 43.22 modeled cold-start milliseconds per request, versus 89.31 for a fixed 15-minute retention rule and 68.12 for an EMA-only allocator. The comparison also shows that minimizing the number of cold starts can differ from minimizing their latency penalty, and that a raw histogram-hazard feature can be detrimental under drift. In a separate readiness-lag test, a one-minute lag yields 44.00 versus 90.02 ms/request for PADA and Fixed15. A retention-biased extension reduces planned slot activation events by 55.1% with minimal change in modeled latency. All reported arrivals are synthetic and startup costs are assumed, rather than measured cloud-service latency.
Zitieren
APA
Lee, J., & Martin, J. (2026). Phase-Aware Dual-Signal Allocation forPredictive Cold-Start Mitigation in Serverless Computing. International Journal of Information and Communication Technology Trends, 6(2), 356-367. https://doi.org/10.66525/ijictt145
Vancouver
Lee J, Martin J. Phase-Aware Dual-Signal Allocation forPredictive Cold-Start Mitigation in Serverless Computing. International Journal of Information and Communication Technology Trends. 2026;6(2):356-367. doi:10.66525/ijictt145
BibTeX
@article{Lee2026,
author = {Lee, James and Martin, Jia},
title = {Phase-Aware Dual-Signal Allocation forPredictive Cold-Start Mitigation in Serverless Computing},
journal = {International Journal of Information and Communication Technology Trends},
year = {2026},
volume = {6},
number = {2},
pages = {356--367},
doi = {10.66525/ijictt145}
}
EUVIMED