Any of you OpenVZ providers using kernel limiters to sleep process per container depending on how much iop they are requesting? We've been hacking at 2.6.32 (OVZ) and messing with the cgroups and so far have been able to run 100 VZ containers on a subpar node without anyone experiencing iolag.

Anyone else played with this?

As much time as shared providers spend on cpu limiting users it's amazing the IO end of things has been ignored.