On Mon 07-12-15 16:06:18, Ankur Khurana wrote: > Hi, > > Recently, I got to know about the control groups on linux and found it > interesting and started experimenting with it. I tried cgroups for > various subsystems and it worked as I would expect it to work. Feeling > empowered with the use of control groups, I went one step further and > tried to get resource isolation on thread level. I used java as a > choice of programming language and tried to do resource allocation to > different threads. It worked really fine for cpu and blkio but results > for memory were somewhat interesting. Memory controller accounting is per process (resp. per mm_struct which is shared between threads in the same process or between processes created by clone(CLONE_VM) in general). The memory is accounted to the owner of the mm_struct which is usually the group leader of the process but this might get more muddy when the mm_struct is shared between processes which are not in the same thread group (see mm_update_next_owner). The reason is that the memory is inherently shared between threads. The implementation is also much simpler and straightforward because the charging operates on the mm_struct rather than a task same as other operations which are specific to user address space. HTH -- Michal Hocko SUSE Labs -- To unsubscribe from this list: send the line "unsubscribe cgroups" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
