The expansion of memory disaggregation, driven by data-centric applications, increases heterogeneity in memory systems. This shift enables the use of inexpensive, yet lifetime-limited, flash memory to be used as a memory expansion module. We argue that TRIM should be introduced into memory management systems to effectively respond to this transition. In this position paper, we explore the potential adoption of flash memory as memory expansion and present an analytical model that offers a straightforward yet rigorous evaluation of TRIM's effectiveness. Using this model and characteristics extracted from real-world workloads, we evaluate the effectiveness of TRIM in scalable memory systems and prove its necessity.

This paper proposes a novel photo storage system called NDPipe, which accelerates the performance of training and inference for image data by leveraging near-data processing in photo storage servers. NDPipe distributes storage servers with inexpensive commodity GPUs in a data center and uses their collective intelligence to perform inference and training near image data. By efficiently partitioning deep neural network (DNN) models and exploiting the data parallelism of many storage servers, NDPipe can achieve high training throughput with low synchronization costs. NDPipe optimizes the near-data processing engine to maximally utilize system components in each storage server. Our results show that, given the same energy budget, NDPipe exhibits 1.39× higher inference throughput and 2.64× faster training speed than typical photo storage systems.

We present BigKV, a key-value cache specifically designed for caching large objects in an all-flash array (AFA). The design of BigKV is centered around the unique property of a cache: since it contains a copy of the data, exact bookkeeping of what is in the cache is not critical for correctness. By ignoring hash collisions, approximating metadata information, and allowing data loss from failures, BigKV significantly increases the cache hit ratio and keeps more useful objects in the system. Experiments on a real AFA show that our design increases the throughput by 3.1× on average and reduces the average and tail latency by 57% and 81%, respectively.