Originally Posted by FractalGrid
I was thinking either to give the guests an OS disk on the SSD or using it for swap. The Redhat document seemed to suggest using it for swap, but I thought I wanted to avoid any swapping of the host.
The RedHat virtualization guide has a few bright areas, and many that are ... not so ... bright? It could be called the "dev with a laptop" problem. They write about what they know and can try. The problem is they don't do anything else than spinning up a few virtual OS for testing software. This is utterly irrelevant when you're supposed to actually run servers.
- You can slice the SSD into many swap spaces for the virtual machines
- if you use ZFS you could use it as mixed ZIL + L2ARC device. L2ARC is simply cache, ZIL speeds up the FS log. Advantage here is that ZFS has really smart code that will automagically detach an SSD that has too many errors coming up.
- if you wanna go an uncommon route, use it as a "FlashCache"* device. Flash Cache has a mode called "write-around-read" which is perfect as a poor mans tiered storage. Writes go directly to disk avoiding wear on the SSD, but commonly read stuff is cached to the SSD.
From your introduction I guess you can handle building a kernel module
(This is a project by FaceBook which they are using in production for quite some time. So less bleeding edge than ZFS on Linux in a way, although one could argue whats more robust: Linux port of modern unix-grade filesystem or a facebook patch to linux)