drakvuf performance query

[rishiesh bhide]

Hi,
I had query reagrding drakvuf (4.8) performance. Did some basic performance testing by finding average Mutex creation time taken by guest with & without drakvuf (syscall).

I found that drakvuf adds around 0.2-0.4 millisecond overhead just for dispatching a callback. Even though I write empty callback handler function & trap only NtCreateMutant, it causes 30x performance degradation over standalone VM for mutex creation API ( 0.3ms taken vs 0.01 std time ).

Is this expected overhead?

[Tamas K Lengyel]

Yes, that's expected. The majority of the overhead doesn't usually
come from the callback itself (though it can if it's doing something
time-consuming), but from the fact that multiple VM exits and VM
entries have to be performed for the event to reach the callback.
Doing those context switches between the guest VM, Xen and the dom0
domain is what adds the most of the overhead.

> --
> You received this message because you are subscribed to the Google Groups "vmitools" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to vmitools+u...@googlegroups.com.
> To post to this group, send email to vmit...@googlegroups.com.
> Visit this group at https://groups.google.com/group/vmitools.
> For more options, visit https://groups.google.com/d/optout.