Swpsparrot : an Application Environment for Data - Intensive Computingdouglas

Volume 6, Number 3, pp. 9 18. http://www.s pe.org ISSN © 2005 SWPS PARROT: AN APPLICATION ENVIRONMENT FOR DATA-INTENSIVE COMPUTING DOUGLAS THAIN AND MIRON LIVNY∗ Abstra t. Distributed omputing ontinues to be an alphabet-soup of servi es and proto ols for managing omputation and storage. To live in this environment, appli ations require middleware that an transparently adapt standard interfa es to new distributed systems; su h middleware is known as an interposition agent. In this paper, we present several lessons learned about interposition agents via a progressive study of design possibilities. Although performan e is an important on ern, we pay spe ial attention to less tangible issues su h as portability, reliability, and ompatibility. We begin with a omparison of seven methods of interposition and sele t one method, the debugger trap, that is the slowest but also the most reliable. Using this method, we implement a omplete interposition agent, Parrot, that spli es existing remote I/O systems into the namespa e of standard appli ations. The primary design problem of Parrot is the mapping of xed appli ation semanti s into the semanti s of the available I/O systems. We o er a detailed dis ussion of how errors and other unexpe ted onditions must be arefully managed in order to keep this mapping inta t. We on lude with a evaluation of the performan e of the I/O proto ols employed by Parrot, and use an Andrew-like ben hmark to demonstrate that semanti di eren es have onsequen es in performan e.1