Random IOPS through a SATA interface is orders of magnitude faster on NAND vs a spinning disk. A SATA SSD and a spinning disk might as well be two completely different categories of device for any non-sequential workload.

▲

kube-system 2 days ago | parent [-]

If you need NAND for the performance, why use a SATA controller? NVME is more available, faster, and cheaper.

▲

aidenn0 2 days ago | parent [-]

How do you put 16 NVME drives on a system with 24 total PCIe lanes?

▲

kube-system 2 days ago | parent | next [-]

You can do it with PCIE switches, but you shouldn't. If you need 16 NVME drives, you need to put them in an appropriate host system.

To properly design a computing solution, first you define the requirements, and then you select components with specifications that will fulfill your requirements.

If you try to work backwards, you are destined for failure. Dropping big cash on 16 high-capacity SSDs just to ham-jam them in an old system is a really really dumb idea, especially if you're concerned about IOPS.

▲

aidenn0 a day ago | parent [-]

Are you not aware that the region between "saturating 64 PCIe channels" and "spinning disks" contains a huge space of performance, much of which might be acceptable?

	▲	kube-system a day ago \| parent [-]
		Yes, but you can exceed those specifications using newer and cheaper hardware.

▲

justsomehnguy 2 days ago | parent | prev [-]

Tri-mode HBA/RAID, eg https://www.broadcom.com/products/storage/host-bus-adapters/...