Abstract: A solid state disk (SSD) caches disk-based volumes in a heterogeneous storage system, improving the overall storage-system performance. The hottest data blocks are identified based on two factors: the frequency of access, and temporal locality. Temporal locality is computed using a logarithmic system time. IO latency is reduced by migrating these hottest data blocks from hard-disk-based volumes to the solid-state flash-memory disks. Some dedicated mapping metadata and a novel top-K B-tree structure are used to index the blocks. Data blocks are ranked by awarding a higher current value for recent accesses, but also by the frequency of accesses. A non-trivial value for accesses in the past is retained by accumulating the two factors over many time spans expressed as a logarithmic system time. Having two factors, access frequency and the logarithmic system time, provides for a more balanced caching system.

Abstract: A solid state disk (SSD) caches disk-based volumes in a heterogeneous storage system, improving the overall storage-system performance. The hottest data blocks are identified based on two factors: the frequency of access, and temporal locality. Temporal locality is computed using a logarithmic system time. IO latency is reduced by migrating these hottest data blocks from hard-disk-based volumes to the solid-state flash-memory disks. Some dedicated mapping metadata and a novel top-K B-tree structure are used to index the blocks. Data blocks are ranked by awarding a higher current value for recent accesses, but also by the frequency of accesses. A non-trivial value for accesses in the past is retained by accumulating the two factors over many time spans expressed as a logarithmic system time. Having two factors, access frequency and the logarithmic system time, provides for a more balanced caching system.

Abstract: A solid state disk (SSD) caches disk-based volumes in a heterogeneous storage system, improving the overall storage-system performance. The hottest data blocks are identified based on two factors: the frequency of access, and temporal locality. Temporal locality is computed using a logarithmic system time. IO latency is reduced by migrating these hottest data blocks from hard-disk-based volumes to the solid-state flash-memory disks. Some dedicated mapping metadata and a novel top-K B-tree structure are used to index the blocks. Data blocks are ranked by awarding a higher current value for recent accesses, but also by the frequency of accesses. A non-trivial value for accesses in the past is retained by accumulating the two factors over many time spans expressed as a logarithmic system time. Having two factors, access frequency and the logarithmic system time, provides for a more balanced caching system.