When putting together our VPS system, we spent a lot of time looking at XEN. Most VPS providers use XEN. XEN has great tools.
However, we eventually settled on using KVM because it was conceptually simpler, had better I/O performance (according to hdparm and ab), but most importantly because XEN’s future began to look bleak. XEN might have better tools now, we figured, but KVM will get better attention.
Most of our VPS customers so far have moved from XEN-based solutions, and greatly appreciate the performance difference. I wanted to note some of the things I’ve noticed helping them move from their current providers:
- XEN hypervisor’d kernels don’t boot under KVM (hint: rsync –exclude /boot)
- Installs are a lot slower on KVM than on XEN (perhaps it’s a problem with CDROM emulation)
- fsync performance is lower on KVM and XEN; noatime helps a lot
- Lots of providers seem to create swap as a file inside the guest. This appears to be done to simplify accounting, but we get enormous performance gains by not doing this (even under XEN)
- CPU tasks are a lot faster under XEN. If you’re compiling things all day, XEN (or a regular virtual hosting account) might make more sense.
The KVM Forum reported lots of really interesting developments that our VPS customers will be able to take advantage of, that XEN-based VPS users will simply miss out on, at least until xenner becomes stable.