Methods to cluster transactions into utilization classes with similar workload characteristics
DocUID: 1995-003 Full Text:
PDF Author: Alexandros Labrinidis
Abstract: Knowledge of the workload intrinsic characteristics is essential for dynamic goal oriented workload control algorithms used to optimize the distributed online transaction processing (OLTP) system's performance behavior, for example through the use of transaction routing algorithms. ``Intrinsic characteristics'' are not dependent on arrival rates, and they include the average number of database accesses, the files accessed, the CPU demand and the average number of synchronization points. CLUE is an environment for clustering transactions according to their workload intrinsic characteristics. It uses execution traces from distributed OLTP systems in order to cluster transactions with high data affinity in utilization classes. HALC is a simple, fast, heuristic algorithm that was developed to cope with the large volume of trace data. A Test Suite Generator was developed in order to create synthetic trace files as input to CLUE. Validation of CLUE's correctness has been made through the use of synthetic trace files. HALC's speed and quality of clustering were evaluated in comparison with the ISODATA and Bond Energy algorithms on real traces. Results have shown that HALC is exceptionally fast and that the quality of the clustering is always really good.
Published In: Institute of Computer Science, FORTH, Technical Report No. 135
Place Published: Greece
Year Published: 1995
Project: Others Subject Area: Others
Publication Type: Technical Report
Sponsor: Others