group_isolation is a bad thing on very low load, but helps a lot when the harddisk is very busy to give equal shares to the disk even on random io. we monitor the weighted diskstat and if a threshold is reached in the window size, group_isolation is enabled. this causes a better responsive system on heavy io, but still allows optimisation by the io scheduler on low io.
complain if sysfs is not mounted
safer detection of kernel threads per user io active group added trace debugging to the scheduler
we adjust the oom killer flags of the most important programs so the X server and the session programs won't be killed when oom runs through. added lua api for oom and ioprio
will be existing in debian branch only, as debian maintainer requested.
gnome has the same bad behaviour as kde...
returns the current position in the active pids list, so 1 is the current used program. minor cleanups
we can make help swaping out memory while still help to preventing lags after longer idle phases by using fine graduated swappiness
cleanup the whole xwatch plugin. it now searches the users env for XAUTHORITY files tries harder to connect. this should now work with gdm login managers that use a extra xauth file per session. removed the fork since seteuid is just doing the job fine. cleanup various other places
we now use nice hashtables for saving the environment, preparsed. we do this only on demand, same on the command line. export process groups to lua
ulatency.get_uid_stats returns a tuple (active, idle) for fast rule swtiches
we now track all sessions created/removed from console kit in the core. this tracking allows now to destinguish between active and ideling sessions depending the desktop ui notifies consolekit correctly. fixed bug when DEVELOP_MODE is on and started as root with sudo. now compiles with dbus off
ulatencyd now provides a dbus interface on the system bus that allows to query and manipulate the daemon while running. currently implemented methods are: - querying the system flags. allows a gui to notify the user when something bad is happening. Signal for this is pending. - set the active flag for processes. this allows the window manager for example to set the active pid and no hack through the xwatch plugin is required. - get informations about available scheduler configs and - allows with root priviliges to change the configuration while running. this can be used to switch to a game scheduler config for example extended the u_scheduler interface to provide more methods for this interface
if the parent pid changes for something to 0, i guess this can happen to kernel threads, they need to be attached to the root node
add some more asserts. prevent endless loop when proc is freed
env, commandline etc will be stored in u_proc and no longer be used from libproc. the values will read on demand. fix for bug #5