"all" results in cumulative stats for all devices. Both for disk and network (on Linux) this results in multiple times counting the same device:

• Network devices (e.g. eth0) may be part of other (stacked) devices like eth0.8 (for vlan's) or bridges.
• Disk devices (e.g. sda) may be part of other (stacked) devices like sda1 (partition) LVM devices and RAID devices.
  So because many things may be counted multiple times, the resulting cumulative count becomes somewhat meaningles.

The attached patch illustrates how this can be solved by allowing users to specify a regexp in the agent config to make the agent filter only useful devices.