Optane P5800X Drives in RAID 1 Optimal Chunk Size?

What is the optimal chunk size for Optane drives with MDM? I’m setting up a mirror of two P5800X drives and I would like to leverage the best possible read / write performance for these drives. @wendell had shown in benchmarks that the SQLite performance of these drives is best in class. What settings would be best to get to there?

root@s76:~# cat /proc/mdstat
Personalities : [raid1] [linear] [multipath] [raid0] [raid6] [raid5] [raid4] [raid10] 
md1 : active raid1 nvme3n1[1] nvme2n1[0]
      390579264 blocks super 1.2 [2/2] [UU]
      [==>..................]  resync = 11.8% (46403584/390579264) finish=27.8min speed=205728K/sec
      bitmap: 3/3 pages [12KB], 65536KB chunk

md0 : active raid1 nvme0n1p2[0] nvme1n1p2[1]
      1951856640 blocks super 1.2 [2/2] [UU]
      bitmap: 0/15 pages [0KB], 65536KB chunk

md0 is two Samsung 980 Pros in RAID 1.
md1 is two Intel P5800X in RAID 1.

The chunk size of 65536KB seems VERY large for NAND and 3DXPoint drives? md0 was setup during the Debian install and I feel that doesn’t seem like a sensible default for this type of device.

A chunk size that large makes sense for random I/o that’s decently large where you want to minimize p
Latency.

Have to understand and or test your workload on different configs

My load is SQLite databases for PHP sessions and Rust applications that share data to the website front end via the SQLite databases. It’s lots of random reads and writes. I would ideally like to also put my /tmp dir on there as well as my swap as it’s way fast and it keeps it off my NAND drives. (If I’m swapping tho, something is wrong, I don’t even use 1GB of RAM on my server right now.)

Optimal chunk size will depend on db page size.
The default page size for SQLlite databases is 4kb (Compile-time Options). This is configurable as it is stored in the db file header ( 512b-64kb).

Any test results to report? :wink:

This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.