Score:0

Why mdadm RAID does **NOT** boost reading performance?

br flag

I add 2 NVMe SSDs(Samsung 980 Pro) into my PC as tablespace of postgresql, while the OS boots from another SATA SSD(Samsung 860 EVO) as before.

These 2 nvme m.2 SSDs have a basic performace about reading 6GB/s and writing 600MB/s, that I got in Gnome-Disks.

I made a soft RAID from these two nvme by means of mdadm, and then tested again.

The writing rate reaches 1.3GB/s as expected, but the reading rate still is 6GB/s.

I'm wondering why the reading performance does NOT change?

Thanks!

CPU:AMD-5900X

Motherboard: Asus Pro WS X570-ACE

Chipset: AMD-X570

OS:Kubuntu-20.04.2

in flag
For the sake of completeness, are you using RAID0 or RAID1?
Score:4
in flag

6GB/s is a common max rate for disk controllers.

Very likely, the speed bottleneck is not the disks but the disk controller, so if both disks are on the same controller, you won't get faster than 6GB/s.

Looking at your motherboard specs, it has 4x SATA 6Gb/s ports, 2x 8GB/s M.2 slot. The specs on the nvme disk are max 7GB/s according to Samsung's web page.

It appears that the first M.2 slot should get 4 lanes, which would support ~8GB/s. The second M.2 slot only gets 2 lanes, and one of those is shared with a PCI slot, so it only gets both lanes if the PCI slot is empty.

Leon avatar
br flag
Thank you very much! Your info is very useful! Since I have only two SATA disk, if I connect them to SATA0-1 & SATA0-2, and disable rest SATA controller in BIOS, can I let more PCIe4 lanes be arranged to the second M.2 slot? Or is there a method can arrange lanes manually?
user10489 avatar
in flag
The lanes are hard wired to specific devices. Specific devices share them, and SATA is not mentioned as sharing lanes, so I doubt that would make a difference. You can't just juggle them. Read your manual, it mentions a specific PCI port that shares lanes with M.2_2, which is still half the speed of M.2_1
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.