ZFS Datastore - Fragmentation with SSD ?

[TwiX]

Hi,

I have recently replaced all my HDD for 10 SSD for my ZFS Datastore.

This is now extremely fast specially verify jobs (even if we can discuss the interest of verifying snapshots on a ZFS pool).

The weird thing is that PBS reports fragmentation on the related pool :



AFAIK, on a SSD pool, there is no fragmentation ?

 

[aaron]

AFAIK, on a SSD pool, there is no fragmentation ?

Well, definitely something I would not worry about with SSDs, even if ZFS reports some fragmentation. It still sees that the data is fragmented across available space of the SSDs.

 

[TwiX]

grazie mille

 

[hvisage]

In the context of SSDs, all writes are "fragmented" by design by the storage controllers on/inside the NVMe/SSDs

In the context of ZFS it's the ZFS that was not able to find a contiguous block of size Y and had to split it in smaller chunks (ashift size) and spread that all over the storage. Something that is more frequent as your ZFS pool usage nears/goes over the ~10-20% free space available mark and ZFS then spends more time to find blocks closer together to each and splitting up blocks, so yes, there is a performance impact, just not direct I/O related, but suddenly instead of a single Y block transfer, you now issue Nx ashift size IO requests and increasing the IOPS needed