Linux guest, vmware and the lazzy clock mystery

Last week we were having one issue in our virtual machines with lazy clocks. The guest, rh with 2.6.9-67 kernel, in a vmware 3.02 ESX host, was having troubles of inconsistency in their clock. Sometimes the time ran very slowly. We apply all the workarounds that vmware explain here. The difficulty didn't disappear.

The difference was so big that even the ntp daemon didn't sync the clocks. The reason was that ntpd was failing to sync the clocks because was drifted outside it's acceptable limit.

After look a little, we found that the problem was in the configuration of the virtual machine. We make a pool of 4 CPU for balance the load between three different guest. VMware was calibrating the CPU depending of their use and the variation in CPU was so bigger that the ntpd could manage it.

Angel Rey proposed to change the Reservation and the Limit in the CPU Resources config of the virtual Machine. The trouble disappears.

Here there are a good PDF explaining how VMWare manage these concerns from Mark Fei.