I have a Linux server system installed on an SSD and an HDD for user data. As there is space left on the SSD, I want to use it as a read cache for the HDD.
Looking at the possibilities, I found:
-
dm-cache: Needs a lot of reads before caching shows performance improvements according to https://www.redhat.com/en/blog/improving-read-performance-dm-cache. I think in my use case this is not a good strategy.
-
lvmcache: Builds on dm-cache and I'd need to put the SSD and HDD into one LV. I'd like to keep the cache transparent, so I can easily put the HDD into another system without having to do LVM magic first.
-
Bcache: I'd need to format the HDD for Bcache. Not what I want.
-
Flashcache: Sounds like what I want (just switch it on), but not actively maintained anymore according to https://github.com/facebookarchive/flashcache.
-
EnhanceIO: Builds on Flashcache, but dead since 2015.
Is there anything similar to Flashcache or EnhanceIO, but still actively maintained?
Best Answer
That is a decent survey of Linux kernel block caches. Of those, I only consider lvmcache and bcache: integrated into kernel and documented by stable distributions.
Both require a format for metadata, that is not going to be easy to avoid.
Moving disks to a different system is relatively easy if the target system also has caching support. LVM capable distros will scan for volumes, if not automatically with a simple command. Similar for bcache.
Either way, mount the volumes by file system UUID or label, abstracting away from device numbering.
Test procedures for removing the cache in planned and unplanned scenarios. Such as bcache in detached passthrough mode. Once you can recover just the backing disk without its cache, no need to revert to a raw, pre-cache disk anymore.