aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/x86/x86_64/fake-numa-for-cpusets
diff options
context:
space:
mode:
Diffstat (limited to 'Documentation/x86/x86_64/fake-numa-for-cpusets')
-rw-r--r--Documentation/x86/x86_64/fake-numa-for-cpusets66
1 files changed, 66 insertions, 0 deletions
diff --git a/Documentation/x86/x86_64/fake-numa-for-cpusets b/Documentation/x86/x86_64/fake-numa-for-cpusets
new file mode 100644
index 000000000000..d1a985c5b00a
--- /dev/null
+++ b/Documentation/x86/x86_64/fake-numa-for-cpusets
@@ -0,0 +1,66 @@
1Using numa=fake and CPUSets for Resource Management
2Written by David Rientjes <rientjes@cs.washington.edu>
3
4This document describes how the numa=fake x86_64 command-line option can be used
5in conjunction with cpusets for coarse memory management. Using this feature,
6you can create fake NUMA nodes that represent contiguous chunks of memory and
7assign them to cpusets and their attached tasks. This is a way of limiting the
8amount of system memory that are available to a certain class of tasks.
9
10For more information on the features of cpusets, see Documentation/cpusets.txt.
11There are a number of different configurations you can use for your needs. For
12more information on the numa=fake command line option and its various ways of
13configuring fake nodes, see Documentation/x86_64/boot-options.txt.
14
15For the purposes of this introduction, we'll assume a very primitive NUMA
16emulation setup of "numa=fake=4*512,". This will split our system memory into
17four equal chunks of 512M each that we can now use to assign to cpusets. As
18you become more familiar with using this combination for resource control,
19you'll determine a better setup to minimize the number of nodes you have to deal
20with.
21
22A machine may be split as follows with "numa=fake=4*512," as reported by dmesg:
23
24 Faking node 0 at 0000000000000000-0000000020000000 (512MB)
25 Faking node 1 at 0000000020000000-0000000040000000 (512MB)
26 Faking node 2 at 0000000040000000-0000000060000000 (512MB)
27 Faking node 3 at 0000000060000000-0000000080000000 (512MB)
28 ...
29 On node 0 totalpages: 130975
30 On node 1 totalpages: 131072
31 On node 2 totalpages: 131072
32 On node 3 totalpages: 131072
33
34Now following the instructions for mounting the cpusets filesystem from
35Documentation/cpusets.txt, you can assign fake nodes (i.e. contiguous memory
36address spaces) to individual cpusets:
37
38 [root@xroads /]# mkdir exampleset
39 [root@xroads /]# mount -t cpuset none exampleset
40 [root@xroads /]# mkdir exampleset/ddset
41 [root@xroads /]# cd exampleset/ddset
42 [root@xroads /exampleset/ddset]# echo 0-1 > cpus
43 [root@xroads /exampleset/ddset]# echo 0-1 > mems
44
45Now this cpuset, 'ddset', will only allowed access to fake nodes 0 and 1 for
46memory allocations (1G).
47
48You can now assign tasks to these cpusets to limit the memory resources
49available to them according to the fake nodes assigned as mems:
50
51 [root@xroads /exampleset/ddset]# echo $$ > tasks
52 [root@xroads /exampleset/ddset]# dd if=/dev/zero of=tmp bs=1024 count=1G
53 [1] 13425
54
55Notice the difference between the system memory usage as reported by
56/proc/meminfo between the restricted cpuset case above and the unrestricted
57case (i.e. running the same 'dd' command without assigning it to a fake NUMA
58cpuset):
59 Unrestricted Restricted
60 MemTotal: 3091900 kB 3091900 kB
61 MemFree: 42113 kB 1513236 kB
62
63This allows for coarse memory management for the tasks you assign to particular
64cpusets. Since cpusets can form a hierarchy, you can create some pretty
65interesting combinations of use-cases for various classes of tasks for your
66memory management needs.