Il 10/11/2013 11:31, Michael S. Tsirkin ha scritto: > > Reported-by: Luiz Capitulino <lcapitul...@redhat.com> > > Signed-off-by: Paolo Bonzini <pbonz...@redhat.com> > > So this causes a 12% performance regression on some TCG > tests, I think we should look into a smarter > datastructure to solve the issues.
It causes a 12% performance regression in a single testcase where KVM has a 150x performance regression. This says a lot about the relevance of the testcase. In any case, I have patches to avoid the regression. For 1.7 we can just revert the patches, for 1.8 we can apply this patch together with the optimizations that avoid introducing a regression. Paolo