Hey guys,

I have a scenario where there are a large number (tens of millions) of small files - ~100KB of mostly PDF and JPGs that I am interested to backup on UNRAID. Data will be transferred via NFS. The folder structures is a set of 256 nested folders within 256 folders so 65,536 folders in total. The path for each file is recorded in a database.

 

Obviously HDD write speed for small files is very slow. Would the following be a valid strategy?:
1) Allows different folders at the top level to be written to different disks (to speed up writes since the writes will be randomly selected)

2) Use SSD for parity rather than HDD so it can keep up with up to 24x HDDs writing?

 

We are planning on migrating all of this data to a Ceph based object storage system but it will take some time to get that up and running.

 

Thanks in advance for any opinions or comments

It will partly depend on the parity SSD, not all SSDs can sustain high write speeds, still the array doesn't usually perform very well with simultaneous writes, and will a lot of small files it will always be slow, in part because you cannot bypass fuse, a pool using an exclusive share would likely be much faster.

Thanks for the reply. I've bought a 4TB Samsung 990 Pro which has sustained write of 1.7GBps apparently. I'm more concerned about IOPs that sustained write due to the nature of the small files.

 

Will test it out and report back.