I am pleased to share my performance results only after many months of planning. My previous configuration was limited to 2TB SATA SSDs in RAIDZ1, which was bottlenecking my workloads severely. The new configuration utilizes four NVMe drives in ZFS mirror pairs.
CrystalDiskMark results are attached for reference only:
Sequential Read: 3,200 MB/s (previous: 520 MB/s)
Sequential Write: 2,800 MB/s (previous: 480 MB/s)
4K Random Read QD32: 580K IOPS
4K Random Write QD32: 520K IOPS
The drives themselves were decommissioned units from Walmart self-checkout machines. They are 256GB each, enterprise-grade with high endurance ratings. Kindly note that firmware is locked to a custom signature, but standard NVMe operations function normally.
I am happy to answer questions from fellow members regarding this configuration.
Transparency: I need to be honest here. We ran locked firmware NVMe at my last gig and had a blameless postmortem when three drives dropped offline simultaneously. The vendor signature check failed after a power event, drives refused to initialize, error budget burned in one afternoon.
Your CDM numbers look healthy. But transparency means asking: what's the plan when these decide they're still in a self-checkout and refuse your writes?
Sorry, total noob here but how did you get four NVMe drives in one server?? My motherboard only has two slots. Is there some kind of expansion card?? Sorry if this is obvious!!
¡those numbers are very impressive! Jajaja but ¿the walmart firmware? I had a friend who worked inventory and he said those machines run hot, very hot, is very concerning for flash endurance. He saw them die in eighteen months in the field
I wrote a long paragraph about the ethics but I deleted it. Short version: scavenging is scavenging, the drives dont care where they sleep. Jajaja. But watch the smart data closely
Post a reply
You need an account to reply.
Log in or
register to join the conversation.
