Publikationen

Autor	Vanerio, Juan; Liu, Huiran; Zhang, Qi; Schmid, Stefan
Datum	2024
Art	Conference Paper
Abstrakt	Content Distribution Networks (CDNs) play a crucial role in efficiently delivering online content to end-users. In this paper, we initiate the study of CDN domain planning with flexible assignments of domains to Points of Presence (PoPs) within a CDN, with the objective of minimizing the cost of transmissions while providing sufficient resources to serve the communication demands. The problem is subject to practical constraints of network deployment such as a percentile-based billing model, PoP's bandwidth and committed rate limits, geographic locality and quota constraints and minimum per domain cache-hit-ratios. We formulate the problem as an offline optimization task with a nonlinear objective function and linear constraints, which becomes computationally intensive for medium-sized instances. The 95th percentile billing model, commonly used by service providers, contributes significantly to this non-linearity. To address this, we propose Baiji, a multi-algorithm approach lever- aging insights from our formulation. Our empirical evaluation of Baiji on two synthetic and one real-world workloads demonstrates its effectiveness in approaching the upper bound on system performance. Baiji provides high-quality solutions for CDN monthly planning, with tunable execution times (from tens of seconds up to four hours), making Baiji suitable for practical deployment in CDNs.
Konferenz	Networking Conference 2024
Url	https://publica.fraunhofer.de/handle/publica/485512