November 4, 2025
Fast, furious, and fileless
Building blobd: single-machine object store with sub-ms reads and 15 GB/s upload
Blazing-fast DIY storage thrills speed nerds, spooks safety crowd
TLDR: Blobd promises lightning-fast reads and 15 GB/s uploads by bypassing the usual layers. Commenters split: speed fans cheer, while skeptics slam missing S3 compatibility, worry about “YOLO” data safety, and ask for comparisons to standard filesystems. It’s a thrilling tech demo, but trust and tooling remain the big hurdles.
A developer built “blobd,” a homebrew storage engine that dodges the cloud and even skips the filesystem to hit sub-millisecond reads and eye-watering 15 GB/s uploads. The speed flex is real: tiny reads arrive almost instantly, while popular tools like MinIO take tens of milliseconds. But the comment section? Pure chaos. One chorus is hyped on the performance, another clutches backup plans like pearls. The hottest line: “YOLO your data” — the community’s way of saying this thing feels fast… and a little reckless.
Users pointed out familiar ancestors like Facebook’s Haystack and SeaweedFS, suggesting blobd isn’t a moon landing so much as a hot new remix. Others demanded real-world comparisons to tried-and-true filesystems — think XFS, ZFS, btrfs — before they crown a new king. The vibe: speed freaks vs. safety squad. Jokes flew about “sub-ms reads, sub-par backups,” and folks imagined the CTO asking for S3 (Amazon’s storage API) compatibility while the dev just whispers, “trust me.” It’s a classic internet showdown: blistering performance versus boring reliability, with everyone agreeing on one thing — if it really makes streaming snappier, users will feel it instantly, and they’ll never go back.
Key Points
- •Blobd is an open-source, single-machine object store optimized for low-latency random reads and small objects.
- •It prioritizes reads over writes and favors creates over updates and deletes, aiming for constant-time lookups without key enumeration.
- •The system uses a raw block device with three components—heap allocator, index, and journal—and employs memory mapping for on-disk metadata.
- •On eight 1.8 GB/s write NVMe SSDs, blobd saturates disk bandwidth during uploads, exceeding 15 GB/s total throughput.
- •Blobd achieves sub-millisecond TTFB for random 4K reads across objects, with example ~0.33 ms for 12 KB objects, outperforming MinIO in cited tests.