Wekaio __top__ -

In the mid-2010s, a quiet crisis was brewing in data centers. Flash storage—blazing-fast SSDs—had arrived, but the software running the servers couldn't keep up. It was like putting a Formula 1 engine inside a horse-drawn carriage. The hardware was ready to scream, but the operating system and traditional file systems were choking on the data.

They built the on a radical idea: a parallel, distributed architecture that runs on standard x86 servers. Instead of a dedicated storage appliance, Weka turned every server’s local flash into part of a giant, unified pool. It was a "shared-nothing" architecture that used NVMe-over-Fabrics (NVMe-oF) to connect everything at near-memory speeds. wekaio

Here was the magic trick: Weka separated the metadata (the "table of contents" telling where a file lives) from the data itself. Both were spread across all servers in the cluster. This meant there was no single "traffic cop" to get overwhelmed. When 10,000 servers asked for 10,000 different files, the system simply used all its brains at once. By 2018, WekaIO was ready. In independent benchmark tests (using the SPECsfs standard), the software achieved a stunning result: over 2 million I/O operations per second (IOPS) and 90 gigabytes per second of throughput from a single namespace. That was like downloading the entire Library of Congress in under a minute. In the mid-2010s, a quiet crisis was brewing in data centers