Posted Jan 31, 2007 8:24 UTC (Wed) by vmlinuz (subscriber, #24)
[Link]
You miss the point of the lrustyguestvisor. The answer to your question is that the performance doesn't matter because of the puppies. Look at all the cute puppies! Ahem... He's quite a persuasive speaker, is our Rusty...
A slightly more serious answer is that performance isn't the number one goal - it's not bad, but it's not going to be heavily tuned for performance if it makes things more complicated. At this point, it's simple enough that I believe it doesn't do SMP yet, although support for an SMP host kernel has apparently just been added.
lguest performance
Posted Feb 2, 2007 12:13 UTC (Fri) by rusty (subscriber, #26)
[Link]
> So what's performance like?
Poor. Last I did a guest kernel compile, it was twice as slow as running in the host (mind you, the guest only had 512M of ram, the host had 3G, so some can be explained away).
That said, there are plenty of low hanging fruit left, and progress is being made quickly. Balancing small code vs. performance is going to be the trick, but I'd hope to get within 25% of native on all reasonable benchmarks. Of course, for some things it's there already, but context-switch-heavy loads really show the weakness of the shadow page table implementation, for example.