Based on kernel version 4.8. Page generated on 2016-10-06 23:19 EST.
1 Using numa=fake and CPUSets for Resource Management 2 Written by David Rientjes <firstname.lastname@example.org> 3 4 This document describes how the numa=fake x86_64 command-line option can be used 5 in conjunction with cpusets for coarse memory management. Using this feature, 6 you can create fake NUMA nodes that represent contiguous chunks of memory and 7 assign them to cpusets and their attached tasks. This is a way of limiting the 8 amount of system memory that are available to a certain class of tasks. 9 10 For more information on the features of cpusets, see 11 Documentation/cgroup-v1/cpusets.txt. 12 There are a number of different configurations you can use for your needs. For 13 more information on the numa=fake command line option and its various ways of 14 configuring fake nodes, see Documentation/x86/x86_64/boot-options.txt. 15 16 For the purposes of this introduction, we'll assume a very primitive NUMA 17 emulation setup of "numa=fake=4*512,". This will split our system memory into 18 four equal chunks of 512M each that we can now use to assign to cpusets. As 19 you become more familiar with using this combination for resource control, 20 you'll determine a better setup to minimize the number of nodes you have to deal 21 with. 22 23 A machine may be split as follows with "numa=fake=4*512," as reported by dmesg: 24 25 Faking node 0 at 0000000000000000-0000000020000000 (512MB) 26 Faking node 1 at 0000000020000000-0000000040000000 (512MB) 27 Faking node 2 at 0000000040000000-0000000060000000 (512MB) 28 Faking node 3 at 0000000060000000-0000000080000000 (512MB) 29 ... 30 On node 0 totalpages: 130975 31 On node 1 totalpages: 131072 32 On node 2 totalpages: 131072 33 On node 3 totalpages: 131072 34 35 Now following the instructions for mounting the cpusets filesystem from 36 Documentation/cgroup-v1/cpusets.txt, you can assign fake nodes (i.e. contiguous memory 37 address spaces) to individual cpusets: 38 39 [root@xroads /]# mkdir exampleset 40 [root@xroads /]# mount -t cpuset none exampleset 41 [root@xroads /]# mkdir exampleset/ddset 42 [root@xroads /]# cd exampleset/ddset 43 [root@xroads /exampleset/ddset]# echo 0-1 > cpus 44 [root@xroads /exampleset/ddset]# echo 0-1 > mems 45 46 Now this cpuset, 'ddset', will only allowed access to fake nodes 0 and 1 for 47 memory allocations (1G). 48 49 You can now assign tasks to these cpusets to limit the memory resources 50 available to them according to the fake nodes assigned as mems: 51 52 [root@xroads /exampleset/ddset]# echo $$ > tasks 53 [root@xroads /exampleset/ddset]# dd if=/dev/zero of=tmp bs=1024 count=1G 54  13425 55 56 Notice the difference between the system memory usage as reported by 57 /proc/meminfo between the restricted cpuset case above and the unrestricted 58 case (i.e. running the same 'dd' command without assigning it to a fake NUMA 59 cpuset): 60 Unrestricted Restricted 61 MemTotal: 3091900 kB 3091900 kB 62 MemFree: 42113 kB 1513236 kB 63 64 This allows for coarse memory management for the tasks you assign to particular 65 cpusets. Since cpusets can form a hierarchy, you can create some pretty 66 interesting combinations of use-cases for various classes of tasks for your 67 memory management needs.