Hacker News Reader

Comment by PaulHoule | Hacker News Reader

I think of Xen as a dying technology.

Lately IBM passed around a paper aimed at Amazon that showed how badly paravirtual machines behave compared to bare metal.

What they didn't note is that newer Intel chips have serious support for hardware virtualization, and that with hardware virtualizaton you get near bare metal performance. And they didn't note that the new instance types on AWS support hardware virtualization.

Paravirtualization was an important technology for a while because it was the fastest sort of virtualization ten years ago, but now it is the slowest.

View on HN

Paravirtualization bad, hardware virtualization good, is not an accurate view. You actually want a hybrid model to get the best performance. That's why KVM developers have put so much effort into the paravirtualized virtio drivers, for example. Xen has put lots of work into taking advantage of new hardware virtualization technology where it makes sense and continuing paravirtualization where it makes sense. See the two links below for great details.

http://blog.xen.org/index.php/2012/10/23/the-paravirtualizat...

http://blog.xen.org/index.php/2012/10/31/the-paravirtualizat...

Very informative, especially the second page, thanks for linking that!

If you have any other suggested reading re Xen / Virtualization I'd be very interested in them.

Try http://www.brendangregg.com/blog/2014-05-07/what-color-is-yo... and http://www.brendangregg.com/blog/2014-05-09/xen-feature-dete... .

I'm not sure if you're already aware and have concerns with it, but Xen offers several levels, ranging from full HVM to PV:

http://www.brendangregg.com/blog/2014-05-07/what-color-is-yo...

Xen and KVM development is funded by Intel and others. After virtualization algorithms (paravirtualization) are proven in software, they move into hardware. This has been taking place for almost a decade and continues. There is no conflict between S/W & H/W virtualization.

I haven't seen that paper but in my experience with thousands of Xen servers, the CPU overhead is less than 5% (typically about 3% or so), not sure what parameters are discussed when talking about how badly VM behave (obv. there's I/O contention for several VMs on a server etc)

Reply on news.ycombinator.com