Workload Analysis and Caching Strategies for Search Advertising Systems
To maximize profit and connect users to relevant products and services, search advertising systems use sophisticated machine learning algorithms to estimate the revenue expectations of thousands of matching ad listings per query. These machine learning computations constitute a substantial part of the operating cost, e.g., 10% to 30% of the total gross revenues. It is desirable to cache and reuse previous computation results to reduce this cost, but caching introduces approximation which comes with potential revenue loss. To maximize cost savings while minimizing the overall revenue impact, an intelligent refresh policy is required to decide when to refresh the cached computation results.
Our workload analysis and cache simulations show that caching is preferable yet challenging for search advertising systems. In this talk, I present two different caching designs: one uses heuristics and one uses fast machine learning algorithm to build the caching strategies. Evaluation shows that a traditional cache design reduces the total computation cost the most, but also introduces huge negative revenue and net profit impact. On the other hand, our proposed cache designs minimize the negative revenue impact while keeping the amount of cost savings, and improves the total net profit.
Presented in Partial Fulfillment of the CSD Speaking Skills Requirement