danfra
Member OP
Sequential write collapse confirmed
Ran fio on my OVHcloud VPS this morning. 850 MB/s in February. 90 MB/s now. Same instance, same zone.
Tested with direct=1, bs=1M, numjobs=4. Not a noisy neighbor situation—this is sustained degradation over weeks.
Anyone else tracking this? I run these weekly for client baseline reports. First time I've seen this pattern on NVMe. I used https://github.com/masonr/yet-another-bench-script for the numbers.
mitigated 800Gbps before breakfast