0

How would it affect performance of the whole lvm2 pool if different physical volumes have had different speed properties?

I'm speaking foremost in the AWS/EC2 context. If I have one physical volume which is 1TB (3000 IOps), the other is 100GB (300 IOps), would it degrade the whole LVM volume group to just 300 IOps?

Based on this answer it seems that LVM is:

really just a table lookup in a fairly small table that block X is actually block Y on device Z.

Hence if I write one bunch of small files, which just happen to be mapped to the speedier physical volume, the whole transaction must be quicker than if that it was mapped to the slower drive?

How does it work in general?

Community
  • 1
NarūnasK
  • 408

1 Answers1

0

Don't do it. Performance will be unpredictable, which is worse than being slow. PVs with the same size and speed is easier to understand.

Further, on linear volumes, it will use one PV at a time. Once extents start writing to the slow PV, that chunk of data will be slow.

There are ways to tier or cache on a Linux storage stack. For example, lvmcache and bcache.

All fast disk is simplest for consistent performance, at a higher cost.

John Mahowald
  • 36,071