On Wed, Sep 7, 2016 at 9:52 AM, Sander Kuusemets <sander.kuusemets at ut.ee> wrote: > Hello, > > You could do it with Cgroups, if your specified processes run or can be > made to run in group permissions. > > https://access.redhat.com/documentation/en-US/Red_Hat_Enterp > rise_Linux/6/html/Resource_Management_Guide/sec-cpu_and_ > memory-use_case.html > > Regards, > > -- > Sander Kuusemets > University of Tartu, High Performance Computing, IT Specialist > Skype: sander.kuusemets1 > +372 737 5694 > > hello, In the past I mainly used blkio controller in control groups, to limit disk I/O bandwidth of a process. Just to notice that the cpu controller also has the cfs_quota_us and cfs_period_us which can provide absolute limits, in respect with relative ones of the shares parameter. In Red Hat manual I only see the shares parameter, but you can find more info about the other two installing kernel-doc package and reading /usr/share/doc/kernel-doc-2.6.32/Documentation/scheduler/sched-bwc.txt A rapid test on a CentOS 6.5 system with 2 physical cpus, 8 cores each and HT enabled yum install libcgroup service start cgconfig cgcreate -g cpu:/half_cpu echo "2000" > /cgroup/cpu/half_cpu/cpu.cfs_quota_us echo "4000" > /cgroup/cpu/half_cpu/cpu.cfs_period_us You can tune latency/burst working on period value... see the doc I use a program that makes matrix multiplication (1024x1024 in example below) and fully uses one cpu ./bmm 1024 elapsed 10 seconds using top during execution I see line detail of this type around the 32 threads: Cpu18 :100.0%us, 0.0%sy, 0.0%ni, 0.0%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st cgexec -g cpu:half_cpu ./bmm 1024 elapsed 20 seconds using top I see line detail of this type during execution: Cpu0 : 49.8%us, 0.0%sy, 0.0%ni, 50.2%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st HIH, Gianluca