Eve: A measurement-centric emulation environment for adaptive internet servers

Abstract

Emulation plays a central role in the performance evaluation, capacity planning, and workload characterization of servers and data centers. Emulation tools usually require developers to focus on mimicking application behavior as well as to deal with system-level details of managing the emulation. With the continuing increase in computing capacity and complexity, capturing the interactions between different parts of an emulation (e.g., clients' reactions to server reconfiguration) increases the complexity and overhead of emulation design. Furthermore, since the amount of measurement data can easily be huge, efficient data management is becoming a key requirement to the proper scalability of any emulation tool. In this paper, we propose Eve, an efficient emulation environment that provides rapid development of distributed and adaptive emulators. By incorporating in-path data processing and custom triggers into a distributed shared variable (DSV) core, Eve provides full and customizable control of how and when measurement data is moved from the source to the DSV, where the data is stored. Both functions simplify data management and minimize the overhead of frequent updates, thus enhancing the created emulator's scalability. They also simplify feedback monitoring and control when creating adaptive emulators. The capabilities of Eve are shown to allow emulation designers to focus on application behavior rather than on system-level details.