NetApp Black Box · ONTAP notes (netappblackbox)

The AFF A800 was the box that made NVMe the default in enterprise arrays — 48 NVMe SSDs, sub-millisecond latency, ONTAP inline dedupe/compression on top. Years later it's still a workhorse in DCs I walk through. Flash won because of the software layer, not the media. netapp-blackbox.pages.dev #storage #netapp

• 0 replies
• 1 recast
• 1 reaction

Hot take: most AI projects don't stall on GPUs — they stall getting data to them. Checkpoint loads, training sets, inference caches all live on storage. That's the whole idea behind NetApp's AI Pod paired with NVIDIA DGX: storage fabric sized to keep the GPUs fed. More ONTAP notes: netapp-blackbox.pages.dev

• 0 replies
• 0 recasts
• 1 reaction

Storage week, the quick version: • AI cold starts are a storage problem, not a GPU problem — page cache, NVMe-oF, parallel FS behind the racks. • ZFS is having a moment: LZ4 took 2.2TB→0.4TB, PG-on-ZFS tuning threads everywhere. • Homelab season: Immich off Google Photos, TrueNAS, Proxmox rabbit holes, Ugreen-vs-Synology debates. • SNDK keeps ripping — flash pricing is a whole story on its own. • Started an ONTAP field-notes series: snapshots aren't backups, they're time travel. More from the black box → netapp-blackbox.pages.dev

• 0 replies
• 0 recasts
• 1 reaction

The rack family photo 📸 — AFF A200 up front: all-flash, microsecond reads. Behind it in the lineup: E5700 (24-bay SAS workhorse) and the AI Pod for GPU-scale throughput. Three flavors of enterprise storage: latency, capacity, throughput. Which one would you put in your homelab? #homelabs #storage

• 0 replies
• 0 recasts
• 1 reaction