KNOWAC: I/O Prefetch via Accumulated Knowledge

Abstract

Abstract-The lasting memory-wall problem combined with the newly emerged big-data problem makes data access delay the first citizen of performance optimizations of cluster computing. Reduction of data access delay, however, is application dependent. It depends on the data access behaviors of the underlying applications. Therefore, leaning and understanding data access behaviors is a must for effective data access optimizations. Modern microprocessors are equipped with hardware data prefetchers, which predict data access patterns and prefetch data for CPU. However, memory systems in design do not have the capability to understand data access behaviors for performance optimizations. In this study, we propose a novel approach, named KNOWAC, to collect I/O information automatically through highlevel I/O libraries. KNOWAC accumulates I/O knowledge and reveals data usage patterns by exploring the collected highlevel I/O characteristics. The discovered data usage patterns can be used for different I/O optimizations. We apply KNOWAC to I/O prefetch under the framework of PnetCDF in this study. Experimental results on a real-world application show that KNOWAC is promising and has a true practical value in mitigating the I/O bottleneck.

Omnisc'IO: A Grammar-Based Approach to Spatial and Temporal I/O Patterns Prediction

“…Most of the work on spatial I/O patterns predication is done to assist I/O prefetching using various approaches, including Markov models [8], speculative execution [29], and knowledge accumulation [5]. These studies predict the I/O behavior based on statistical methods, however, and therefore require either prior knowledge of the application or long runs before the model converges.…”

“…The effectiveness of such techniques strongly depends on a certain level of knowledge of the I/O access patterns: prefetching and caching indeed require the location of future accesses (i.e., spatial behavior), while I/O scheduling leverages estimations of I/O requests interarrival time (i.e., temporal behavior). The key challenges inherent in these techniques include the proper comprehension and exploitation of the application's I/O behavior within the I/O stack itself [7], [5]. Hence, modeling and predicting the applications I/O behavior are of utmost importance.…”

“…To alleviate the performance penalty of the I/O bottleneck in petascale systems, researchers have explored techniques such as prefetching, caching, and scheduling [4], [5], [6] to improve I/O performance. The effectiveness of such techniques strongly depends on a certain level of knowledge of the I/O access patterns: prefetching and caching indeed require the location of future accesses (i.e., spatial behavior), while I/O scheduling leverages estimations of I/O requests interarrival time (i.e., temporal behavior).…”

A Checkpoint of Research on Parallel I/O for High-Performance Computing

“…This information is used to guide prefetching to a local buffer. Similar approaches -access pattern detection to guide prefetching -are proposed by Patrick et al [81], He, Sun, and Thakur [32], Lu et al [65], and Tang et al [97].…”

“…The processes' access spatiality is also obtained from MPI-IO in the approach proposed by Song et al [93]. Similarly, He, Sun, and Thakur [32] present an approach to collect I/O information from the PnetCDF library. Semantic data access information is obtained by collecting high-level variables.…”