| |||||||||||||
ELCE11: Sandboxing for automotive LinuxELCE11: Sandboxing for automotive LinuxPosted Nov 3, 2011 23:00 UTC (Thu) by martinfick (subscriber, #4455)In reply to: ELCE11: Sandboxing for automotive Linux by jimparis Parent article: ELCE11: Sandboxing for automotive Linux
> If done right, hardware-level virtualization (ie KVM) should have only minor performance implications, especially on a platform where things like network and file I/O are already quite slow and shouldn't be affected by a small CPU overhead.
I think that is dreaming. OS level virtualization can handle 1000s of guests, do you think KVM "done right" could even handle 100?
(Log in to post comments)
ELCE11: Sandboxing for automotive Linux Posted Nov 4, 2011 19:59 UTC (Fri) by jimparis (subscriber, #38647) [Link]
When you start talking about 100 or 1000 guests, the limiting factors to full virtualization quickly become I/O bandwidth, scheduler pressure, RAM, etc. Virtualizing exactly 1 guest is an entirely different problem, especially if the primary goals are security and trying to mix two dissimilar systems. So yeah, I do think that "KVM done right" is far better for isolating a single Android instance than trying to modify both the host and guest to coexist. That's not to say that OS level virtualization doesn't have its uses.
|
|||||||||||||