There's a bunch of patches that export resource tracking per-pid
(which you can effectively map back to uid). However if you are
collecting aggregated stats, this requires polling from userspace in
case you have processes that die or the kernel oom killer kicks in and
you lose that stats.
What are people's thoughts on yet another stats in either /proc or
/sys to export resources (network, cpu, io) aggregated per uid. Or
does this already exist in 2.6.27 somewhere?
--
MIke Chan