This article details the author's experience creating an object store called blobd, optimized for speed with sub-millisecond read times and high upload rates. It discusses design choices, including using a hash-based index and direct I/O to bypass traditional filesystems. The open-source project aims to enhance performance for small object storage.

This article analyzes cloud hardware developments over the past decade, focusing on performance improvements in CPU, memory, network, and NVMe storage. While network bandwidth has significantly increased, gains in CPU and memory have stagnated, and NVMe performance in the cloud has not kept pace with on-premise hardware. The findings suggest a shift towards specialized hardware and software integration to maximize performance.

The Marginalia Search index has undergone significant redesign to enhance performance through new data structures optimized for modern hardware, increasing the index size from 350 million to 800 million documents. The article discusses the challenges faced in query performance and the implications of NVMe SSD characteristics, as well as the transition from B-trees to deterministic block-based skip lists for improved efficiency in document retrieval.

DeepNVMe has been updated to enhance I/O performance in deep learning applications by improving checkpointing with FastPersist and model inference with ZeRO-Inference. These advancements include support for CPU-only environments, offset-based I/O operations, and tensor data type casting, along with significant speedups facilitated by Gen5 NVMe SSDs. The updates aim to democratize access to large models and optimize I/O-bound workloads for various users.