Due to high access performance and price-per-byte considerations, flash memory has been recommended for use as a mid-tier cache in a multi-tier storage system. However, previous studies related to flash-based mid-tier caching only considered the indirect use of flash memory via a flash translation layer, which causes expensive flash-based cache maintenance. This paper identifies the weaknesses of such indirect methods, with a focus on the cold-page migration problem. As improvements, we propose two novel approaches, an indirect approach called LPD (logical page drop) and a native approach called NFA (native flash access). The basic idea is to drop cold pages proactively so that the garbage collection overhead can be minimized. Our experiments demonstrate that both approaches, especially the native one, effectively improve the use of flash memory in the mid-tier cache. NFA reduces the number of garbage collections and block erasures by up to a factor of five and improves the mid-tier throughput by up to 66%. © 2012 Springer-Verlag.
CITATION STYLE
Ou, Y., Xu, J., & Härder, T. (2012). Towards an efficient flash-based mid-tier cache. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7446 LNCS, pp. 55–70). https://doi.org/10.1007/978-3-642-32600-4_6
Mendeley helps you to discover research relevant for your work.