I am running a private cloud (opennebula, KVM) where I am running VMs both directly using LVM Logical Volumes and using disk images on a file system (one is faster, the other is easier to migrate), all on top of raid5.
I managed to procure some SSDs that I can put into my servers, and I would like to use them as cache, most probably using DM-cache (as bcache needs a disk reformat and flashcache is still not mainline).
Now, as I would like to cache multiple LVs, I could create separate metadata and cache partitions for each one, or I could create a new LV, put dm-cache on top of that and then use it as a PV and move the previous LVs there one by one.
So the resulting layering would be:
md1
== sda+sdb+sdc+sddPV0
== md1LV0
<- PV0 (I intend to leave the host OS's volumes uncached)PV1
== LV0 + dmcache- (
LV1
,LV2
,LV3
, ...) <- PV1
So my question are:
- Does this make sense? :)
- Is there a significant (>5%) performance penalty for an additional LVM layer?
- Is there an inherent incompatibility issue in this setup that I should be aware of?
0 Answers