Interesting idea, not sure how well a utility could empirically test without a hardcoded list of SKUs, but here are some thoughts. * One difference is PLP - power loss protection. An enterprise drive must IMHO offer this and I don’t know offhand of one that doesn’t. Client / desktops drives often don’t. Don’t know if this can be programmatically determined. Maybe for NVMe, if compliant, but doubtful for SAS/SATA. * Sustained performance. Some client drives (the industry term) may look okay — for the first 60 seconds or so, then we see what is sometimes called cliffing, where it saturates and performance drops precipitously. I dont’ know off hand what kind of threshold you might enforce here, this could perhaps be determined by trying against representative SKUs. * Endurance is important. Client drives are intended for much gentler duty cycles than most enterprise drives. For general Ceph usage I usually suggest a ~~ 1DWPD SKU; for archival / object purposes QLC drives with 0.3 - 1.0 DWPD work great.

On Jan 10, 2025, at 5:11 PM, Martin Konold <martin.konold@konsec.com> wrote:

Hi there,

it is well documented that Ceph performance is extremely poor with consumer ssd/nvme block devices.

Recommending enterprise or data center devices is IMHO not sufficient as these terms are not really standardized.

I propose to write a little system program which determines the properties of a device with regards to the system calls performed by ceph osd.

Can anyone hint me how such a program should look like? Simply some timing of fsync system calls? _______________________________________________ ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io