Why is Devicemapper (direct-lvm) unsuitable for PaaS-type work?

build 7392c3b/1.12.6
The following documentation has some great advice for picking a storage driver:


I especially like the part where it lists the pros and cons of each pertaining to production usage. I am investigating configuring an AMI linux production instance for performance. (https://docs.docker.com/engine/userguide/storagedriver/images/driver-pros-cons.png)
Why is Devicemapper (direct-lvm) unsuitable for PaaS-type work?
As an aside, I considered AUFS from the docs, but given AMI linux is CentOS like, that was infeasible.

For one, aufs can share memory between processes and I don’t think devicemapper can.

In general I’d say most with a lot of experience have hit issues with devicemapper stability and performance, especially if it is not meticulously configured (I haven’t run it in production so don’t take this as gospel truth, it’s just the “word on the street”).

I’d recommend highly aufs or overlay2 if your kernel is recent enough.

ah, thanks @nathanleclaire this is helpful. From what I checked on the instance, aufs was not in the proc file. I will explore this or overlay2.

as an aside though, Amazon Linux is CentOSy so I think aufs is not recommended per this link: https://docs.docker.com/engine/userguide/storagedriver/device-mapper-driver/#/an-alternative-to-aufs