Abstract: In a compaction scheme for a log structured merge (LSM) tree, a storage device is configured to: receive a first user data piece from a host; generate a first meta data piece for a highest level of the LSM tree, corresponding to the first user data piece; when the highest level exceeds a set storage limit, trigger a compaction process on the first and second meta data pieces to generate compacted meta data pieces excluding overlapping meta data elements of the second meta data pieces overlapped with the first meta data piece. Through a garbage collection, victim user data elements corresponding to the overlapping meta data elements are deleted.

Abstract: A training system includes a dynamic random access memory (DRAM) configured to buffer training data; a central processing unit (CPU) coupled to the DRAM and configured to downsample the training data and provide the DRAM with the downsampled training data; a computational storage consisting of a solid-state drive (SSD) and field-programmable gate array (FPGA) and configured to perform dimensionality reduction on the downsampled training data to generate training data batches; and a graphic processing unit (GPU) configured to perform training on the training data batches.