Automatic Fine-Grained Transaction Categorization for Multi-tier Applications

Abstract

Multi-tier architecture has become the industry standard for building Web applications. These applications feature in multiple categories of transactions. Properly categorizing the transactions and accurately characterizing the resource usage for each category is crucial for modeling the performance of multi-tier application. Existing studies either ignores the transaction categorization problem by simply using the URL path as the identifier of category, or require complex monitoring infrastructure. In this paper we propose a method called Transaction ICA, which automatically categorize transactions based on only widely available Web access log and aggregate resource utilization data. The method use URL path as the initial categorization setting, and iteratively split and merge categories based on estimated resource usage. The method incorporates regression based resource usage estimation technique and independent component analysis based request categorization technique. We validate the feasibility of our method using a synthetic 2-tier Web application. The experiments shows the method can correctly categorize transactions into coherent groups and give accurate per category resource demand, the result categorization is also more fine-grained than the one from existing method.