1 .. _cpusets: 1 .. _cpusets:
2 2
3 ======= 3 =======
4 CPUSETS 4 CPUSETS
5 ======= 5 =======
6 6
7 Copyright (C) 2004 BULL SA. 7 Copyright (C) 2004 BULL SA.
8 8
9 Written by Simon.Derr@bull.net 9 Written by Simon.Derr@bull.net
10 10
11 - Portions Copyright (c) 2004-2006 Silicon Gra 11 - Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
12 - Modified by Paul Jackson <pj@sgi.com> 12 - Modified by Paul Jackson <pj@sgi.com>
13 - Modified by Christoph Lameter <cl@linux.com> 13 - Modified by Christoph Lameter <cl@linux.com>
14 - Modified by Paul Menage <menage@google.com> 14 - Modified by Paul Menage <menage@google.com>
15 - Modified by Hidetoshi Seto <seto.hidetoshi@jp 15 - Modified by Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
16 16
17 .. CONTENTS: 17 .. CONTENTS:
18 18
19 1. Cpusets 19 1. Cpusets
20 1.1 What are cpusets ? 20 1.1 What are cpusets ?
21 1.2 Why are cpusets needed ? 21 1.2 Why are cpusets needed ?
22 1.3 How are cpusets implemented ? 22 1.3 How are cpusets implemented ?
23 1.4 What are exclusive cpusets ? 23 1.4 What are exclusive cpusets ?
24 1.5 What is memory_pressure ? 24 1.5 What is memory_pressure ?
25 1.6 What is memory spread ? 25 1.6 What is memory spread ?
26 1.7 What is sched_load_balance ? 26 1.7 What is sched_load_balance ?
27 1.8 What is sched_relax_domain_level ? 27 1.8 What is sched_relax_domain_level ?
28 1.9 How do I use cpusets ? 28 1.9 How do I use cpusets ?
29 2. Usage Examples and Syntax 29 2. Usage Examples and Syntax
30 2.1 Basic Usage 30 2.1 Basic Usage
31 2.2 Adding/removing cpus 31 2.2 Adding/removing cpus
32 2.3 Setting flags 32 2.3 Setting flags
33 2.4 Attaching processes 33 2.4 Attaching processes
34 3. Questions 34 3. Questions
35 4. Contact 35 4. Contact
36 36
37 1. Cpusets 37 1. Cpusets
38 ========== 38 ==========
39 39
40 1.1 What are cpusets ? 40 1.1 What are cpusets ?
41 ---------------------- 41 ----------------------
42 42
43 Cpusets provide a mechanism for assigning a se 43 Cpusets provide a mechanism for assigning a set of CPUs and Memory
44 Nodes to a set of tasks. In this document "M 44 Nodes to a set of tasks. In this document "Memory Node" refers to
45 an on-line node that contains memory. 45 an on-line node that contains memory.
46 46
47 Cpusets constrain the CPU and Memory placement 47 Cpusets constrain the CPU and Memory placement of tasks to only
48 the resources within a task's current cpuset. 48 the resources within a task's current cpuset. They form a nested
49 hierarchy visible in a virtual file system. T 49 hierarchy visible in a virtual file system. These are the essential
50 hooks, beyond what is already present, require 50 hooks, beyond what is already present, required to manage dynamic
51 job placement on large systems. 51 job placement on large systems.
52 52
53 Cpusets use the generic cgroup subsystem descr 53 Cpusets use the generic cgroup subsystem described in
54 Documentation/admin-guide/cgroup-v1/cgroups.rs 54 Documentation/admin-guide/cgroup-v1/cgroups.rst.
55 55
56 Requests by a task, using the sched_setaffinit 56 Requests by a task, using the sched_setaffinity(2) system call to
57 include CPUs in its CPU affinity mask, and usi 57 include CPUs in its CPU affinity mask, and using the mbind(2) and
58 set_mempolicy(2) system calls to include Memor 58 set_mempolicy(2) system calls to include Memory Nodes in its memory
59 policy, are both filtered through that task's 59 policy, are both filtered through that task's cpuset, filtering out any
60 CPUs or Memory Nodes not in that cpuset. The 60 CPUs or Memory Nodes not in that cpuset. The scheduler will not
61 schedule a task on a CPU that is not allowed i 61 schedule a task on a CPU that is not allowed in its cpus_allowed
62 vector, and the kernel page allocator will not 62 vector, and the kernel page allocator will not allocate a page on a
63 node that is not allowed in the requesting tas 63 node that is not allowed in the requesting task's mems_allowed vector.
64 64
65 User level code may create and destroy cpusets 65 User level code may create and destroy cpusets by name in the cgroup
66 virtual file system, manage the attributes and 66 virtual file system, manage the attributes and permissions of these
67 cpusets and which CPUs and Memory Nodes are as 67 cpusets and which CPUs and Memory Nodes are assigned to each cpuset,
68 specify and query to which cpuset a task is as 68 specify and query to which cpuset a task is assigned, and list the
69 task pids assigned to a cpuset. 69 task pids assigned to a cpuset.
70 70
71 71
72 1.2 Why are cpusets needed ? 72 1.2 Why are cpusets needed ?
73 ---------------------------- 73 ----------------------------
74 74
75 The management of large computer systems, with 75 The management of large computer systems, with many processors (CPUs),
76 complex memory cache hierarchies and multiple 76 complex memory cache hierarchies and multiple Memory Nodes having
77 non-uniform access times (NUMA) presents addit 77 non-uniform access times (NUMA) presents additional challenges for
78 the efficient scheduling and memory placement 78 the efficient scheduling and memory placement of processes.
79 79
80 Frequently more modest sized systems can be op 80 Frequently more modest sized systems can be operated with adequate
81 efficiency just by letting the operating syste 81 efficiency just by letting the operating system automatically share
82 the available CPU and Memory resources amongst 82 the available CPU and Memory resources amongst the requesting tasks.
83 83
84 But larger systems, which benefit more from ca 84 But larger systems, which benefit more from careful processor and
85 memory placement to reduce memory access times 85 memory placement to reduce memory access times and contention,
86 and which typically represent a larger investm 86 and which typically represent a larger investment for the customer,
87 can benefit from explicitly placing jobs on pr 87 can benefit from explicitly placing jobs on properly sized subsets of
88 the system. 88 the system.
89 89
90 This can be especially valuable on: 90 This can be especially valuable on:
91 91
92 * Web Servers running multiple instances o 92 * Web Servers running multiple instances of the same web application,
93 * Servers running different applications ( 93 * Servers running different applications (for instance, a web server
94 and a database), or 94 and a database), or
95 * NUMA systems running large HPC applicati 95 * NUMA systems running large HPC applications with demanding
96 performance characteristics. 96 performance characteristics.
97 97
98 These subsets, or "soft partitions" must be ab 98 These subsets, or "soft partitions" must be able to be dynamically
99 adjusted, as the job mix changes, without impa 99 adjusted, as the job mix changes, without impacting other concurrently
100 executing jobs. The location of the running jo 100 executing jobs. The location of the running jobs pages may also be moved
101 when the memory locations are changed. 101 when the memory locations are changed.
102 102
103 The kernel cpuset patch provides the minimum e 103 The kernel cpuset patch provides the minimum essential kernel
104 mechanisms required to efficiently implement s 104 mechanisms required to efficiently implement such subsets. It
105 leverages existing CPU and Memory Placement fa 105 leverages existing CPU and Memory Placement facilities in the Linux
106 kernel to avoid any additional impact on the c 106 kernel to avoid any additional impact on the critical scheduler or
107 memory allocator code. 107 memory allocator code.
108 108
109 109
110 1.3 How are cpusets implemented ? 110 1.3 How are cpusets implemented ?
111 --------------------------------- 111 ---------------------------------
112 112
113 Cpusets provide a Linux kernel mechanism to co 113 Cpusets provide a Linux kernel mechanism to constrain which CPUs and
114 Memory Nodes are used by a process or set of p 114 Memory Nodes are used by a process or set of processes.
115 115
116 The Linux kernel already has a pair of mechani 116 The Linux kernel already has a pair of mechanisms to specify on which
117 CPUs a task may be scheduled (sched_setaffinit 117 CPUs a task may be scheduled (sched_setaffinity) and on which Memory
118 Nodes it may obtain memory (mbind, set_mempoli 118 Nodes it may obtain memory (mbind, set_mempolicy).
119 119
120 Cpusets extends these two mechanisms as follow 120 Cpusets extends these two mechanisms as follows:
121 121
122 - Cpusets are sets of allowed CPUs and Memory 122 - Cpusets are sets of allowed CPUs and Memory Nodes, known to the
123 kernel. 123 kernel.
124 - Each task in the system is attached to a cp 124 - Each task in the system is attached to a cpuset, via a pointer
125 in the task structure to a reference counte 125 in the task structure to a reference counted cgroup structure.
126 - Calls to sched_setaffinity are filtered to 126 - Calls to sched_setaffinity are filtered to just those CPUs
127 allowed in that task's cpuset. 127 allowed in that task's cpuset.
128 - Calls to mbind and set_mempolicy are filter 128 - Calls to mbind and set_mempolicy are filtered to just
129 those Memory Nodes allowed in that task's c 129 those Memory Nodes allowed in that task's cpuset.
130 - The root cpuset contains all the systems CP 130 - The root cpuset contains all the systems CPUs and Memory
131 Nodes. 131 Nodes.
132 - For any cpuset, one can define child cpuset 132 - For any cpuset, one can define child cpusets containing a subset
133 of the parents CPU and Memory Node resource 133 of the parents CPU and Memory Node resources.
134 - The hierarchy of cpusets can be mounted at 134 - The hierarchy of cpusets can be mounted at /dev/cpuset, for
135 browsing and manipulation from user space. 135 browsing and manipulation from user space.
136 - A cpuset may be marked exclusive, which ens 136 - A cpuset may be marked exclusive, which ensures that no other
137 cpuset (except direct ancestors and descend 137 cpuset (except direct ancestors and descendants) may contain
138 any overlapping CPUs or Memory Nodes. 138 any overlapping CPUs or Memory Nodes.
139 - You can list all the tasks (by pid) attache 139 - You can list all the tasks (by pid) attached to any cpuset.
140 140
141 The implementation of cpusets requires a few, 141 The implementation of cpusets requires a few, simple hooks
142 into the rest of the kernel, none in performan 142 into the rest of the kernel, none in performance critical paths:
143 143
144 - in init/main.c, to initialize the root cpus 144 - in init/main.c, to initialize the root cpuset at system boot.
145 - in fork and exit, to attach and detach a ta 145 - in fork and exit, to attach and detach a task from its cpuset.
146 - in sched_setaffinity, to mask the requested 146 - in sched_setaffinity, to mask the requested CPUs by what's
147 allowed in that task's cpuset. 147 allowed in that task's cpuset.
148 - in sched.c migrate_live_tasks(), to keep mi 148 - in sched.c migrate_live_tasks(), to keep migrating tasks within
149 the CPUs allowed by their cpuset, if possib 149 the CPUs allowed by their cpuset, if possible.
150 - in the mbind and set_mempolicy system calls 150 - in the mbind and set_mempolicy system calls, to mask the requested
151 Memory Nodes by what's allowed in that task 151 Memory Nodes by what's allowed in that task's cpuset.
152 - in page_alloc.c, to restrict memory to allo 152 - in page_alloc.c, to restrict memory to allowed nodes.
153 - in vmscan.c, to restrict page recovery to t 153 - in vmscan.c, to restrict page recovery to the current cpuset.
154 154
155 You should mount the "cgroup" filesystem type 155 You should mount the "cgroup" filesystem type in order to enable
156 browsing and modifying the cpusets presently k 156 browsing and modifying the cpusets presently known to the kernel. No
157 new system calls are added for cpusets - all s 157 new system calls are added for cpusets - all support for querying and
158 modifying cpusets is via this cpuset file syst 158 modifying cpusets is via this cpuset file system.
159 159
160 The /proc/<pid>/status file for each task has 160 The /proc/<pid>/status file for each task has four added lines,
161 displaying the task's cpus_allowed (on which C 161 displaying the task's cpus_allowed (on which CPUs it may be scheduled)
162 and mems_allowed (on which Memory Nodes it may 162 and mems_allowed (on which Memory Nodes it may obtain memory),
163 in the two formats seen in the following examp 163 in the two formats seen in the following example::
164 164
165 Cpus_allowed: ffffffff,ffffffff,ffffffff,f 165 Cpus_allowed: ffffffff,ffffffff,ffffffff,ffffffff
166 Cpus_allowed_list: 0-127 166 Cpus_allowed_list: 0-127
167 Mems_allowed: ffffffff,ffffffff 167 Mems_allowed: ffffffff,ffffffff
168 Mems_allowed_list: 0-63 168 Mems_allowed_list: 0-63
169 169
170 Each cpuset is represented by a directory in t 170 Each cpuset is represented by a directory in the cgroup file system
171 containing (on top of the standard cgroup file 171 containing (on top of the standard cgroup files) the following
172 files describing that cpuset: 172 files describing that cpuset:
173 173
174 - cpuset.cpus: list of CPUs in that cpuset 174 - cpuset.cpus: list of CPUs in that cpuset
175 - cpuset.mems: list of Memory Nodes in that c 175 - cpuset.mems: list of Memory Nodes in that cpuset
176 - cpuset.memory_migrate flag: if set, move pa 176 - cpuset.memory_migrate flag: if set, move pages to cpusets nodes
177 - cpuset.cpu_exclusive flag: is cpu placement 177 - cpuset.cpu_exclusive flag: is cpu placement exclusive?
178 - cpuset.mem_exclusive flag: is memory placem 178 - cpuset.mem_exclusive flag: is memory placement exclusive?
179 - cpuset.mem_hardwall flag: is memory alloca 179 - cpuset.mem_hardwall flag: is memory allocation hardwalled
180 - cpuset.memory_pressure: measure of how much 180 - cpuset.memory_pressure: measure of how much paging pressure in cpuset
181 - cpuset.memory_spread_page flag: if set, spr 181 - cpuset.memory_spread_page flag: if set, spread page cache evenly on allowed nodes
182 - cpuset.memory_spread_slab flag: OBSOLETE. D !! 182 - cpuset.memory_spread_slab flag: if set, spread slab cache evenly on allowed nodes
183 - cpuset.sched_load_balance flag: if set, loa 183 - cpuset.sched_load_balance flag: if set, load balance within CPUs on that cpuset
184 - cpuset.sched_relax_domain_level: the search 184 - cpuset.sched_relax_domain_level: the searching range when migrating tasks
185 185
186 In addition, only the root cpuset has the foll 186 In addition, only the root cpuset has the following file:
187 187
188 - cpuset.memory_pressure_enabled flag: comput 188 - cpuset.memory_pressure_enabled flag: compute memory_pressure?
189 189
190 New cpusets are created using the mkdir system 190 New cpusets are created using the mkdir system call or shell
191 command. The properties of a cpuset, such as 191 command. The properties of a cpuset, such as its flags, allowed
192 CPUs and Memory Nodes, and attached tasks, are 192 CPUs and Memory Nodes, and attached tasks, are modified by writing
193 to the appropriate file in that cpusets direct 193 to the appropriate file in that cpusets directory, as listed above.
194 194
195 The named hierarchical structure of nested cpu 195 The named hierarchical structure of nested cpusets allows partitioning
196 a large system into nested, dynamically change 196 a large system into nested, dynamically changeable, "soft-partitions".
197 197
198 The attachment of each task, automatically inh 198 The attachment of each task, automatically inherited at fork by any
199 children of that task, to a cpuset allows orga 199 children of that task, to a cpuset allows organizing the work load
200 on a system into related sets of tasks such th 200 on a system into related sets of tasks such that each set is constrained
201 to using the CPUs and Memory Nodes of a partic 201 to using the CPUs and Memory Nodes of a particular cpuset. A task
202 may be re-attached to any other cpuset, if all 202 may be re-attached to any other cpuset, if allowed by the permissions
203 on the necessary cpuset file system directorie 203 on the necessary cpuset file system directories.
204 204
205 Such management of a system "in the large" int 205 Such management of a system "in the large" integrates smoothly with
206 the detailed placement done on individual task 206 the detailed placement done on individual tasks and memory regions
207 using the sched_setaffinity, mbind and set_mem 207 using the sched_setaffinity, mbind and set_mempolicy system calls.
208 208
209 The following rules apply to each cpuset: 209 The following rules apply to each cpuset:
210 210
211 - Its CPUs and Memory Nodes must be a subset 211 - Its CPUs and Memory Nodes must be a subset of its parents.
212 - It can't be marked exclusive unless its par 212 - It can't be marked exclusive unless its parent is.
213 - If its cpu or memory is exclusive, they may 213 - If its cpu or memory is exclusive, they may not overlap any sibling.
214 214
215 These rules, and the natural hierarchy of cpus 215 These rules, and the natural hierarchy of cpusets, enable efficient
216 enforcement of the exclusive guarantee, withou 216 enforcement of the exclusive guarantee, without having to scan all
217 cpusets every time any of them change to ensur 217 cpusets every time any of them change to ensure nothing overlaps a
218 exclusive cpuset. Also, the use of a Linux vi 218 exclusive cpuset. Also, the use of a Linux virtual file system (vfs)
219 to represent the cpuset hierarchy provides for 219 to represent the cpuset hierarchy provides for a familiar permission
220 and name space for cpusets, with a minimum of 220 and name space for cpusets, with a minimum of additional kernel code.
221 221
222 The cpus and mems files in the root (top_cpuse 222 The cpus and mems files in the root (top_cpuset) cpuset are
223 read-only. The cpus file automatically tracks 223 read-only. The cpus file automatically tracks the value of
224 cpu_online_mask using a CPU hotplug notifier, 224 cpu_online_mask using a CPU hotplug notifier, and the mems file
225 automatically tracks the value of node_states[ 225 automatically tracks the value of node_states[N_MEMORY]--i.e.,
226 nodes with memory--using the cpuset_track_onli 226 nodes with memory--using the cpuset_track_online_nodes() hook.
227 227
228 The cpuset.effective_cpus and cpuset.effective 228 The cpuset.effective_cpus and cpuset.effective_mems files are
229 normally read-only copies of cpuset.cpus and c 229 normally read-only copies of cpuset.cpus and cpuset.mems files
230 respectively. If the cpuset cgroup filesystem 230 respectively. If the cpuset cgroup filesystem is mounted with the
231 special "cpuset_v2_mode" option, the behavior 231 special "cpuset_v2_mode" option, the behavior of these files will become
232 similar to the corresponding files in cpuset v 232 similar to the corresponding files in cpuset v2. In other words, hotplug
233 events will not change cpuset.cpus and cpuset. 233 events will not change cpuset.cpus and cpuset.mems. Those events will
234 only affect cpuset.effective_cpus and cpuset.e 234 only affect cpuset.effective_cpus and cpuset.effective_mems which show
235 the actual cpus and memory nodes that are curr 235 the actual cpus and memory nodes that are currently used by this cpuset.
236 See Documentation/admin-guide/cgroup-v2.rst fo 236 See Documentation/admin-guide/cgroup-v2.rst for more information about
237 cpuset v2 behavior. 237 cpuset v2 behavior.
238 238
239 239
240 1.4 What are exclusive cpusets ? 240 1.4 What are exclusive cpusets ?
241 -------------------------------- 241 --------------------------------
242 242
243 If a cpuset is cpu or mem exclusive, no other 243 If a cpuset is cpu or mem exclusive, no other cpuset, other than
244 a direct ancestor or descendant, may share any 244 a direct ancestor or descendant, may share any of the same CPUs or
245 Memory Nodes. 245 Memory Nodes.
246 246
247 A cpuset that is cpuset.mem_exclusive *or* cpu 247 A cpuset that is cpuset.mem_exclusive *or* cpuset.mem_hardwall is "hardwalled",
248 i.e. it restricts kernel allocations for page, 248 i.e. it restricts kernel allocations for page, buffer and other data
249 commonly shared by the kernel across multiple 249 commonly shared by the kernel across multiple users. All cpusets,
250 whether hardwalled or not, restrict allocation 250 whether hardwalled or not, restrict allocations of memory for user
251 space. This enables configuring a system so t 251 space. This enables configuring a system so that several independent
252 jobs can share common kernel data, such as fil 252 jobs can share common kernel data, such as file system pages, while
253 isolating each job's user allocation in its ow 253 isolating each job's user allocation in its own cpuset. To do this,
254 construct a large mem_exclusive cpuset to hold 254 construct a large mem_exclusive cpuset to hold all the jobs, and
255 construct child, non-mem_exclusive cpusets for 255 construct child, non-mem_exclusive cpusets for each individual job.
256 Only a small amount of typical kernel memory, 256 Only a small amount of typical kernel memory, such as requests from
257 interrupt handlers, is allowed to be taken out 257 interrupt handlers, is allowed to be taken outside even a
258 mem_exclusive cpuset. 258 mem_exclusive cpuset.
259 259
260 260
261 1.5 What is memory_pressure ? 261 1.5 What is memory_pressure ?
262 ----------------------------- 262 -----------------------------
263 The memory_pressure of a cpuset provides a sim 263 The memory_pressure of a cpuset provides a simple per-cpuset metric
264 of the rate that the tasks in a cpuset are att 264 of the rate that the tasks in a cpuset are attempting to free up in
265 use memory on the nodes of the cpuset to satis 265 use memory on the nodes of the cpuset to satisfy additional memory
266 requests. 266 requests.
267 267
268 This enables batch managers monitoring jobs ru 268 This enables batch managers monitoring jobs running in dedicated
269 cpusets to efficiently detect what level of me 269 cpusets to efficiently detect what level of memory pressure that job
270 is causing. 270 is causing.
271 271
272 This is useful both on tightly managed systems 272 This is useful both on tightly managed systems running a wide mix of
273 submitted jobs, which may choose to terminate 273 submitted jobs, which may choose to terminate or re-prioritize jobs that
274 are trying to use more memory than allowed on 274 are trying to use more memory than allowed on the nodes assigned to them,
275 and with tightly coupled, long running, massiv 275 and with tightly coupled, long running, massively parallel scientific
276 computing jobs that will dramatically fail to 276 computing jobs that will dramatically fail to meet required performance
277 goals if they start to use more memory than al 277 goals if they start to use more memory than allowed to them.
278 278
279 This mechanism provides a very economical way 279 This mechanism provides a very economical way for the batch manager
280 to monitor a cpuset for signs of memory pressu 280 to monitor a cpuset for signs of memory pressure. It's up to the
281 batch manager or other user code to decide wha 281 batch manager or other user code to decide what to do about it and
282 take action. 282 take action.
283 283
284 ==> 284 ==>
285 Unless this feature is enabled by writing 285 Unless this feature is enabled by writing "1" to the special file
286 /dev/cpuset/memory_pressure_enabled, the h 286 /dev/cpuset/memory_pressure_enabled, the hook in the rebalance
287 code of __alloc_pages() for this metric re 287 code of __alloc_pages() for this metric reduces to simply noticing
288 that the cpuset_memory_pressure_enabled fl 288 that the cpuset_memory_pressure_enabled flag is zero. So only
289 systems that enable this feature will comp 289 systems that enable this feature will compute the metric.
290 290
291 Why a per-cpuset, running average: 291 Why a per-cpuset, running average:
292 292
293 Because this meter is per-cpuset, rather t 293 Because this meter is per-cpuset, rather than per-task or mm,
294 the system load imposed by a batch schedul 294 the system load imposed by a batch scheduler monitoring this
295 metric is sharply reduced on large systems 295 metric is sharply reduced on large systems, because a scan of
296 the tasklist can be avoided on each set of 296 the tasklist can be avoided on each set of queries.
297 297
298 Because this meter is a running average, i 298 Because this meter is a running average, instead of an accumulating
299 counter, a batch scheduler can detect memo 299 counter, a batch scheduler can detect memory pressure with a
300 single read, instead of having to read and 300 single read, instead of having to read and accumulate results
301 for a period of time. 301 for a period of time.
302 302
303 Because this meter is per-cpuset rather th 303 Because this meter is per-cpuset rather than per-task or mm,
304 the batch scheduler can obtain the key inf 304 the batch scheduler can obtain the key information, memory
305 pressure in a cpuset, with a single read, 305 pressure in a cpuset, with a single read, rather than having to
306 query and accumulate results over all the 306 query and accumulate results over all the (dynamically changing)
307 set of tasks in the cpuset. 307 set of tasks in the cpuset.
308 308
309 A per-cpuset simple digital filter (requires a 309 A per-cpuset simple digital filter (requires a spinlock and 3 words
310 of data per-cpuset) is kept, and updated by an 310 of data per-cpuset) is kept, and updated by any task attached to that
311 cpuset, if it enters the synchronous (direct) 311 cpuset, if it enters the synchronous (direct) page reclaim code.
312 312
313 A per-cpuset file provides an integer number r 313 A per-cpuset file provides an integer number representing the recent
314 (half-life of 10 seconds) rate of direct page 314 (half-life of 10 seconds) rate of direct page reclaims caused by
315 the tasks in the cpuset, in units of reclaims 315 the tasks in the cpuset, in units of reclaims attempted per second,
316 times 1000. 316 times 1000.
317 317
318 318
319 1.6 What is memory spread ? 319 1.6 What is memory spread ?
320 --------------------------- 320 ---------------------------
321 There are two boolean flag files per cpuset th 321 There are two boolean flag files per cpuset that control where the
322 kernel allocates pages for the file system buf 322 kernel allocates pages for the file system buffers and related in
323 kernel data structures. They are called 'cpus 323 kernel data structures. They are called 'cpuset.memory_spread_page' and
324 'cpuset.memory_spread_slab'. 324 'cpuset.memory_spread_slab'.
325 325
326 If the per-cpuset boolean flag file 'cpuset.me 326 If the per-cpuset boolean flag file 'cpuset.memory_spread_page' is set, then
327 the kernel will spread the file system buffers 327 the kernel will spread the file system buffers (page cache) evenly
328 over all the nodes that the faulting task is a 328 over all the nodes that the faulting task is allowed to use, instead
329 of preferring to put those pages on the node w 329 of preferring to put those pages on the node where the task is running.
330 330
331 If the per-cpuset boolean flag file 'cpuset.me 331 If the per-cpuset boolean flag file 'cpuset.memory_spread_slab' is set,
332 then the kernel will spread some file system r 332 then the kernel will spread some file system related slab caches,
333 such as for inodes and dentries evenly over al 333 such as for inodes and dentries evenly over all the nodes that the
334 faulting task is allowed to use, instead of pr 334 faulting task is allowed to use, instead of preferring to put those
335 pages on the node where the task is running. 335 pages on the node where the task is running.
336 336
337 The setting of these flags does not affect ano 337 The setting of these flags does not affect anonymous data segment or
338 stack segment pages of a task. 338 stack segment pages of a task.
339 339
340 By default, both kinds of memory spreading are 340 By default, both kinds of memory spreading are off, and memory
341 pages are allocated on the node local to where 341 pages are allocated on the node local to where the task is running,
342 except perhaps as modified by the task's NUMA 342 except perhaps as modified by the task's NUMA mempolicy or cpuset
343 configuration, so long as sufficient free memo 343 configuration, so long as sufficient free memory pages are available.
344 344
345 When new cpusets are created, they inherit the 345 When new cpusets are created, they inherit the memory spread settings
346 of their parent. 346 of their parent.
347 347
348 Setting memory spreading causes allocations fo 348 Setting memory spreading causes allocations for the affected page
349 or slab caches to ignore the task's NUMA mempo 349 or slab caches to ignore the task's NUMA mempolicy and be spread
350 instead. Tasks using mbind() or set_mempoli 350 instead. Tasks using mbind() or set_mempolicy() calls to set NUMA
351 mempolicies will not notice any change in thes 351 mempolicies will not notice any change in these calls as a result of
352 their containing task's memory spread settings 352 their containing task's memory spread settings. If memory spreading
353 is turned off, then the currently specified NU 353 is turned off, then the currently specified NUMA mempolicy once again
354 applies to memory page allocations. 354 applies to memory page allocations.
355 355
356 Both 'cpuset.memory_spread_page' and 'cpuset.m 356 Both 'cpuset.memory_spread_page' and 'cpuset.memory_spread_slab' are boolean flag
357 files. By default they contain "0", meaning t 357 files. By default they contain "0", meaning that the feature is off
358 for that cpuset. If a "1" is written to that 358 for that cpuset. If a "1" is written to that file, then that turns
359 the named feature on. 359 the named feature on.
360 360
361 The implementation is simple. 361 The implementation is simple.
362 362
363 Setting the flag 'cpuset.memory_spread_page' t 363 Setting the flag 'cpuset.memory_spread_page' turns on a per-process flag
364 PFA_SPREAD_PAGE for each task that is in that 364 PFA_SPREAD_PAGE for each task that is in that cpuset or subsequently
365 joins that cpuset. The page allocation calls 365 joins that cpuset. The page allocation calls for the page cache
366 is modified to perform an inline check for thi 366 is modified to perform an inline check for this PFA_SPREAD_PAGE task
367 flag, and if set, a call to a new routine cpus 367 flag, and if set, a call to a new routine cpuset_mem_spread_node()
368 returns the node to prefer for the allocation. 368 returns the node to prefer for the allocation.
369 369
370 Similarly, setting 'cpuset.memory_spread_slab' 370 Similarly, setting 'cpuset.memory_spread_slab' turns on the flag
371 PFA_SPREAD_SLAB, and appropriately marked slab 371 PFA_SPREAD_SLAB, and appropriately marked slab caches will allocate
372 pages from the node returned by cpuset_mem_spr 372 pages from the node returned by cpuset_mem_spread_node().
373 373
374 The cpuset_mem_spread_node() routine is also s 374 The cpuset_mem_spread_node() routine is also simple. It uses the
375 value of a per-task rotor cpuset_mem_spread_ro 375 value of a per-task rotor cpuset_mem_spread_rotor to select the next
376 node in the current task's mems_allowed to pre 376 node in the current task's mems_allowed to prefer for the allocation.
377 377
378 This memory placement policy is also known (in 378 This memory placement policy is also known (in other contexts) as
379 round-robin or interleave. 379 round-robin or interleave.
380 380
381 This policy can provide substantial improvemen 381 This policy can provide substantial improvements for jobs that need
382 to place thread local data on the correspondin 382 to place thread local data on the corresponding node, but that need
383 to access large file system data sets that nee 383 to access large file system data sets that need to be spread across
384 the several nodes in the jobs cpuset in order 384 the several nodes in the jobs cpuset in order to fit. Without this
385 policy, especially for jobs that might have on 385 policy, especially for jobs that might have one thread reading in the
386 data set, the memory allocation across the nod 386 data set, the memory allocation across the nodes in the jobs cpuset
387 can become very uneven. 387 can become very uneven.
388 388
389 1.7 What is sched_load_balance ? 389 1.7 What is sched_load_balance ?
390 -------------------------------- 390 --------------------------------
391 391
392 The kernel scheduler (kernel/sched/core.c) aut 392 The kernel scheduler (kernel/sched/core.c) automatically load balances
393 tasks. If one CPU is underutilized, kernel co 393 tasks. If one CPU is underutilized, kernel code running on that
394 CPU will look for tasks on other more overload 394 CPU will look for tasks on other more overloaded CPUs and move those
395 tasks to itself, within the constraints of suc 395 tasks to itself, within the constraints of such placement mechanisms
396 as cpusets and sched_setaffinity. 396 as cpusets and sched_setaffinity.
397 397
398 The algorithmic cost of load balancing and its 398 The algorithmic cost of load balancing and its impact on key shared
399 kernel data structures such as the task list i 399 kernel data structures such as the task list increases more than
400 linearly with the number of CPUs being balance 400 linearly with the number of CPUs being balanced. So the scheduler
401 has support to partition the systems CPUs into 401 has support to partition the systems CPUs into a number of sched
402 domains such that it only load balances within 402 domains such that it only load balances within each sched domain.
403 Each sched domain covers some subset of the CP 403 Each sched domain covers some subset of the CPUs in the system;
404 no two sched domains overlap; some CPUs might 404 no two sched domains overlap; some CPUs might not be in any sched
405 domain and hence won't be load balanced. 405 domain and hence won't be load balanced.
406 406
407 Put simply, it costs less to balance between t 407 Put simply, it costs less to balance between two smaller sched domains
408 than one big one, but doing so means that over 408 than one big one, but doing so means that overloads in one of the
409 two domains won't be load balanced to the othe 409 two domains won't be load balanced to the other one.
410 410
411 By default, there is one sched domain covering 411 By default, there is one sched domain covering all CPUs, including those
412 marked isolated using the kernel boot time "is 412 marked isolated using the kernel boot time "isolcpus=" argument. However,
413 the isolated CPUs will not participate in load 413 the isolated CPUs will not participate in load balancing, and will not
414 have tasks running on them unless explicitly a 414 have tasks running on them unless explicitly assigned.
415 415
416 This default load balancing across all CPUs is 416 This default load balancing across all CPUs is not well suited for
417 the following two situations: 417 the following two situations:
418 418
419 1) On large systems, load balancing across ma 419 1) On large systems, load balancing across many CPUs is expensive.
420 If the system is managed using cpusets to 420 If the system is managed using cpusets to place independent jobs
421 on separate sets of CPUs, full load balanc 421 on separate sets of CPUs, full load balancing is unnecessary.
422 2) Systems supporting realtime on some CPUs n 422 2) Systems supporting realtime on some CPUs need to minimize
423 system overhead on those CPUs, including a 423 system overhead on those CPUs, including avoiding task load
424 balancing if that is not needed. 424 balancing if that is not needed.
425 425
426 When the per-cpuset flag "cpuset.sched_load_ba 426 When the per-cpuset flag "cpuset.sched_load_balance" is enabled (the default
427 setting), it requests that all the CPUs in tha 427 setting), it requests that all the CPUs in that cpusets allowed 'cpuset.cpus'
428 be contained in a single sched domain, ensurin 428 be contained in a single sched domain, ensuring that load balancing
429 can move a task (not otherwised pinned, as by 429 can move a task (not otherwised pinned, as by sched_setaffinity)
430 from any CPU in that cpuset to any other. 430 from any CPU in that cpuset to any other.
431 431
432 When the per-cpuset flag "cpuset.sched_load_ba 432 When the per-cpuset flag "cpuset.sched_load_balance" is disabled, then the
433 scheduler will avoid load balancing across the 433 scheduler will avoid load balancing across the CPUs in that cpuset,
434 --except-- in so far as is necessary because s 434 --except-- in so far as is necessary because some overlapping cpuset
435 has "sched_load_balance" enabled. 435 has "sched_load_balance" enabled.
436 436
437 So, for example, if the top cpuset has the fla 437 So, for example, if the top cpuset has the flag "cpuset.sched_load_balance"
438 enabled, then the scheduler will have one sche 438 enabled, then the scheduler will have one sched domain covering all
439 CPUs, and the setting of the "cpuset.sched_loa 439 CPUs, and the setting of the "cpuset.sched_load_balance" flag in any other
440 cpusets won't matter, as we're already fully l 440 cpusets won't matter, as we're already fully load balancing.
441 441
442 Therefore in the above two situations, the top 442 Therefore in the above two situations, the top cpuset flag
443 "cpuset.sched_load_balance" should be disabled 443 "cpuset.sched_load_balance" should be disabled, and only some of the smaller,
444 child cpusets have this flag enabled. 444 child cpusets have this flag enabled.
445 445
446 When doing this, you don't usually want to lea 446 When doing this, you don't usually want to leave any unpinned tasks in
447 the top cpuset that might use non-trivial amou 447 the top cpuset that might use non-trivial amounts of CPU, as such tasks
448 may be artificially constrained to some subset 448 may be artificially constrained to some subset of CPUs, depending on
449 the particulars of this flag setting in descen 449 the particulars of this flag setting in descendant cpusets. Even if
450 such a task could use spare CPU cycles in some 450 such a task could use spare CPU cycles in some other CPUs, the kernel
451 scheduler might not consider the possibility o 451 scheduler might not consider the possibility of load balancing that
452 task to that underused CPU. 452 task to that underused CPU.
453 453
454 Of course, tasks pinned to a particular CPU ca 454 Of course, tasks pinned to a particular CPU can be left in a cpuset
455 that disables "cpuset.sched_load_balance" as t 455 that disables "cpuset.sched_load_balance" as those tasks aren't going anywhere
456 else anyway. 456 else anyway.
457 457
458 There is an impedance mismatch here, between c 458 There is an impedance mismatch here, between cpusets and sched domains.
459 Cpusets are hierarchical and nest. Sched doma 459 Cpusets are hierarchical and nest. Sched domains are flat; they don't
460 overlap and each CPU is in at most one sched d 460 overlap and each CPU is in at most one sched domain.
461 461
462 It is necessary for sched domains to be flat b 462 It is necessary for sched domains to be flat because load balancing
463 across partially overlapping sets of CPUs woul 463 across partially overlapping sets of CPUs would risk unstable dynamics
464 that would be beyond our understanding. So if 464 that would be beyond our understanding. So if each of two partially
465 overlapping cpusets enables the flag 'cpuset.s 465 overlapping cpusets enables the flag 'cpuset.sched_load_balance', then we
466 form a single sched domain that is a superset 466 form a single sched domain that is a superset of both. We won't move
467 a task to a CPU outside its cpuset, but the sc 467 a task to a CPU outside its cpuset, but the scheduler load balancing
468 code might waste some compute cycles consideri 468 code might waste some compute cycles considering that possibility.
469 469
470 This mismatch is why there is not a simple one 470 This mismatch is why there is not a simple one-to-one relation
471 between which cpusets have the flag "cpuset.sc 471 between which cpusets have the flag "cpuset.sched_load_balance" enabled,
472 and the sched domain configuration. If a cpus 472 and the sched domain configuration. If a cpuset enables the flag, it
473 will get balancing across all its CPUs, but if 473 will get balancing across all its CPUs, but if it disables the flag,
474 it will only be assured of no load balancing i 474 it will only be assured of no load balancing if no other overlapping
475 cpuset enables the flag. 475 cpuset enables the flag.
476 476
477 If two cpusets have partially overlapping 'cpu 477 If two cpusets have partially overlapping 'cpuset.cpus' allowed, and only
478 one of them has this flag enabled, then the ot 478 one of them has this flag enabled, then the other may find its
479 tasks only partially load balanced, just on th 479 tasks only partially load balanced, just on the overlapping CPUs.
480 This is just the general case of the top_cpuse 480 This is just the general case of the top_cpuset example given a few
481 paragraphs above. In the general case, as in 481 paragraphs above. In the general case, as in the top cpuset case,
482 don't leave tasks that might use non-trivial a 482 don't leave tasks that might use non-trivial amounts of CPU in
483 such partially load balanced cpusets, as they 483 such partially load balanced cpusets, as they may be artificially
484 constrained to some subset of the CPUs allowed 484 constrained to some subset of the CPUs allowed to them, for lack of
485 load balancing to the other CPUs. 485 load balancing to the other CPUs.
486 486
487 CPUs in "cpuset.isolcpus" were excluded from l 487 CPUs in "cpuset.isolcpus" were excluded from load balancing by the
488 isolcpus= kernel boot option, and will never b 488 isolcpus= kernel boot option, and will never be load balanced regardless
489 of the value of "cpuset.sched_load_balance" in 489 of the value of "cpuset.sched_load_balance" in any cpuset.
490 490
491 1.7.1 sched_load_balance implementation detail 491 1.7.1 sched_load_balance implementation details.
492 ---------------------------------------------- 492 ------------------------------------------------
493 493
494 The per-cpuset flag 'cpuset.sched_load_balance 494 The per-cpuset flag 'cpuset.sched_load_balance' defaults to enabled (contrary
495 to most cpuset flags.) When enabled for a cpu 495 to most cpuset flags.) When enabled for a cpuset, the kernel will
496 ensure that it can load balance across all the 496 ensure that it can load balance across all the CPUs in that cpuset
497 (makes sure that all the CPUs in the cpus_allo 497 (makes sure that all the CPUs in the cpus_allowed of that cpuset are
498 in the same sched domain.) 498 in the same sched domain.)
499 499
500 If two overlapping cpusets both have 'cpuset.s 500 If two overlapping cpusets both have 'cpuset.sched_load_balance' enabled,
501 then they will be (must be) both in the same s 501 then they will be (must be) both in the same sched domain.
502 502
503 If, as is the default, the top cpuset has 'cpu 503 If, as is the default, the top cpuset has 'cpuset.sched_load_balance' enabled,
504 then by the above that means there is a single 504 then by the above that means there is a single sched domain covering
505 the whole system, regardless of any other cpus 505 the whole system, regardless of any other cpuset settings.
506 506
507 The kernel commits to user space that it will 507 The kernel commits to user space that it will avoid load balancing
508 where it can. It will pick as fine a granular 508 where it can. It will pick as fine a granularity partition of sched
509 domains as it can while still providing load b 509 domains as it can while still providing load balancing for any set
510 of CPUs allowed to a cpuset having 'cpuset.sch 510 of CPUs allowed to a cpuset having 'cpuset.sched_load_balance' enabled.
511 511
512 The internal kernel cpuset to scheduler interf 512 The internal kernel cpuset to scheduler interface passes from the
513 cpuset code to the scheduler code a partition 513 cpuset code to the scheduler code a partition of the load balanced
514 CPUs in the system. This partition is a set of 514 CPUs in the system. This partition is a set of subsets (represented
515 as an array of struct cpumask) of CPUs, pairwi 515 as an array of struct cpumask) of CPUs, pairwise disjoint, that cover
516 all the CPUs that must be load balanced. 516 all the CPUs that must be load balanced.
517 517
518 The cpuset code builds a new such partition an 518 The cpuset code builds a new such partition and passes it to the
519 scheduler sched domain setup code, to have the 519 scheduler sched domain setup code, to have the sched domains rebuilt
520 as necessary, whenever: 520 as necessary, whenever:
521 521
522 - the 'cpuset.sched_load_balance' flag of a c 522 - the 'cpuset.sched_load_balance' flag of a cpuset with non-empty CPUs changes,
523 - or CPUs come or go from a cpuset with this 523 - or CPUs come or go from a cpuset with this flag enabled,
524 - or 'cpuset.sched_relax_domain_level' value 524 - or 'cpuset.sched_relax_domain_level' value of a cpuset with non-empty CPUs
525 and with this flag enabled changes, 525 and with this flag enabled changes,
526 - or a cpuset with non-empty CPUs and with th 526 - or a cpuset with non-empty CPUs and with this flag enabled is removed,
527 - or a cpu is offlined/onlined. 527 - or a cpu is offlined/onlined.
528 528
529 This partition exactly defines what sched doma 529 This partition exactly defines what sched domains the scheduler should
530 setup - one sched domain for each element (str 530 setup - one sched domain for each element (struct cpumask) in the
531 partition. 531 partition.
532 532
533 The scheduler remembers the currently active s 533 The scheduler remembers the currently active sched domain partitions.
534 When the scheduler routine partition_sched_dom 534 When the scheduler routine partition_sched_domains() is invoked from
535 the cpuset code to update these sched domains, 535 the cpuset code to update these sched domains, it compares the new
536 partition requested with the current, and upda 536 partition requested with the current, and updates its sched domains,
537 removing the old and adding the new, for each 537 removing the old and adding the new, for each change.
538 538
539 539
540 1.8 What is sched_relax_domain_level ? 540 1.8 What is sched_relax_domain_level ?
541 -------------------------------------- 541 --------------------------------------
542 542
543 In sched domain, the scheduler migrates tasks 543 In sched domain, the scheduler migrates tasks in 2 ways; periodic load
544 balance on tick, and at time of some schedule 544 balance on tick, and at time of some schedule events.
545 545
546 When a task is woken up, scheduler try to move 546 When a task is woken up, scheduler try to move the task on idle CPU.
547 For example, if a task A running on CPU X acti 547 For example, if a task A running on CPU X activates another task B
548 on the same CPU X, and if CPU Y is X's sibling 548 on the same CPU X, and if CPU Y is X's sibling and performing idle,
549 then scheduler migrate task B to CPU Y so that 549 then scheduler migrate task B to CPU Y so that task B can start on
550 CPU Y without waiting task A on CPU X. 550 CPU Y without waiting task A on CPU X.
551 551
552 And if a CPU run out of tasks in its runqueue, 552 And if a CPU run out of tasks in its runqueue, the CPU try to pull
553 extra tasks from other busy CPUs to help them 553 extra tasks from other busy CPUs to help them before it is going to
554 be idle. 554 be idle.
555 555
556 Of course it takes some searching cost to find 556 Of course it takes some searching cost to find movable tasks and/or
557 idle CPUs, the scheduler might not search all 557 idle CPUs, the scheduler might not search all CPUs in the domain
558 every time. In fact, in some architectures, t 558 every time. In fact, in some architectures, the searching ranges on
559 events are limited in the same socket or node 559 events are limited in the same socket or node where the CPU locates,
560 while the load balance on tick searches all. 560 while the load balance on tick searches all.
561 561
562 For example, assume CPU Z is relatively far fr 562 For example, assume CPU Z is relatively far from CPU X. Even if CPU Z
563 is idle while CPU X and the siblings are busy, 563 is idle while CPU X and the siblings are busy, scheduler can't migrate
564 woken task B from X to Z since it is out of it 564 woken task B from X to Z since it is out of its searching range.
565 As the result, task B on CPU X need to wait ta 565 As the result, task B on CPU X need to wait task A or wait load balance
566 on the next tick. For some applications in sp 566 on the next tick. For some applications in special situation, waiting
567 1 tick may be too long. 567 1 tick may be too long.
568 568
569 The 'cpuset.sched_relax_domain_level' file all 569 The 'cpuset.sched_relax_domain_level' file allows you to request changing
570 this searching range as you like. This file t 570 this searching range as you like. This file takes int value which
571 indicates size of searching range in levels ap !! 571 indicates size of searching range in levels ideally as follows,
572 otherwise initial value -1 that indicates the 572 otherwise initial value -1 that indicates the cpuset has no request.
573 573
574 ====== ======================================= 574 ====== ===========================================================
575 -1 no request. use system default or follo 575 -1 no request. use system default or follow request of others.
576 0 no search. 576 0 no search.
577 1 search siblings (hyperthreads in a core 577 1 search siblings (hyperthreads in a core).
578 2 search cores in a package. 578 2 search cores in a package.
579 3 search cpus in a node [= system wide on 579 3 search cpus in a node [= system wide on non-NUMA system]
580 4 search nodes in a chunk of node [on NUM 580 4 search nodes in a chunk of node [on NUMA system]
581 5 search system wide [on NUMA system] 581 5 search system wide [on NUMA system]
582 ====== ======================================= 582 ====== ===========================================================
583 <<
584 Not all levels can be present and values can c <<
585 system architecture and kernel configuration. <<
586 /sys/kernel/debug/sched/domains/cpu*/domain*/ <<
587 details. <<
588 583
589 The system default is architecture dependent. 584 The system default is architecture dependent. The system default
590 can be changed using the relax_domain_level= b 585 can be changed using the relax_domain_level= boot parameter.
591 586
592 This file is per-cpuset and affect the sched d 587 This file is per-cpuset and affect the sched domain where the cpuset
593 belongs to. Therefore if the flag 'cpuset.sch 588 belongs to. Therefore if the flag 'cpuset.sched_load_balance' of a cpuset
594 is disabled, then 'cpuset.sched_relax_domain_l 589 is disabled, then 'cpuset.sched_relax_domain_level' have no effect since
595 there is no sched domain belonging the cpuset. 590 there is no sched domain belonging the cpuset.
596 591
597 If multiple cpusets are overlapping and hence 592 If multiple cpusets are overlapping and hence they form a single sched
598 domain, the largest value among those is used. 593 domain, the largest value among those is used. Be careful, if one
599 requests 0 and others are -1 then 0 is used. 594 requests 0 and others are -1 then 0 is used.
600 595
601 Note that modifying this file will have both g 596 Note that modifying this file will have both good and bad effects,
602 and whether it is acceptable or not depends on 597 and whether it is acceptable or not depends on your situation.
603 Don't modify this file if you are not sure. 598 Don't modify this file if you are not sure.
604 599
605 If your situation is: 600 If your situation is:
606 601
607 - The migration costs between each cpu can be 602 - The migration costs between each cpu can be assumed considerably
608 small(for you) due to your special applicat 603 small(for you) due to your special application's behavior or
609 special hardware support for CPU cache etc. 604 special hardware support for CPU cache etc.
610 - The searching cost doesn't have impact(for 605 - The searching cost doesn't have impact(for you) or you can make
611 the searching cost enough small by managing 606 the searching cost enough small by managing cpuset to compact etc.
612 - The latency is required even it sacrifices 607 - The latency is required even it sacrifices cache hit rate etc.
613 then increasing 'sched_relax_domain_level' 608 then increasing 'sched_relax_domain_level' would benefit you.
614 609
615 610
616 1.9 How do I use cpusets ? 611 1.9 How do I use cpusets ?
617 -------------------------- 612 --------------------------
618 613
619 In order to minimize the impact of cpusets on 614 In order to minimize the impact of cpusets on critical kernel
620 code, such as the scheduler, and due to the fa 615 code, such as the scheduler, and due to the fact that the kernel
621 does not support one task updating the memory 616 does not support one task updating the memory placement of another
622 task directly, the impact on a task of changin 617 task directly, the impact on a task of changing its cpuset CPU
623 or Memory Node placement, or of changing to wh 618 or Memory Node placement, or of changing to which cpuset a task
624 is attached, is subtle. 619 is attached, is subtle.
625 620
626 If a cpuset has its Memory Nodes modified, the 621 If a cpuset has its Memory Nodes modified, then for each task attached
627 to that cpuset, the next time that the kernel 622 to that cpuset, the next time that the kernel attempts to allocate
628 a page of memory for that task, the kernel wil 623 a page of memory for that task, the kernel will notice the change
629 in the task's cpuset, and update its per-task 624 in the task's cpuset, and update its per-task memory placement to
630 remain within the new cpusets memory placement 625 remain within the new cpusets memory placement. If the task was using
631 mempolicy MPOL_BIND, and the nodes to which it 626 mempolicy MPOL_BIND, and the nodes to which it was bound overlap with
632 its new cpuset, then the task will continue to 627 its new cpuset, then the task will continue to use whatever subset
633 of MPOL_BIND nodes are still allowed in the ne 628 of MPOL_BIND nodes are still allowed in the new cpuset. If the task
634 was using MPOL_BIND and now none of its MPOL_B 629 was using MPOL_BIND and now none of its MPOL_BIND nodes are allowed
635 in the new cpuset, then the task will be essen 630 in the new cpuset, then the task will be essentially treated as if it
636 was MPOL_BIND bound to the new cpuset (even th 631 was MPOL_BIND bound to the new cpuset (even though its NUMA placement,
637 as queried by get_mempolicy(), doesn't change) 632 as queried by get_mempolicy(), doesn't change). If a task is moved
638 from one cpuset to another, then the kernel wi 633 from one cpuset to another, then the kernel will adjust the task's
639 memory placement, as above, the next time that 634 memory placement, as above, the next time that the kernel attempts
640 to allocate a page of memory for that task. 635 to allocate a page of memory for that task.
641 636
642 If a cpuset has its 'cpuset.cpus' modified, th 637 If a cpuset has its 'cpuset.cpus' modified, then each task in that cpuset
643 will have its allowed CPU placement changed im 638 will have its allowed CPU placement changed immediately. Similarly,
644 if a task's pid is written to another cpuset's 639 if a task's pid is written to another cpuset's 'tasks' file, then its
645 allowed CPU placement is changed immediately. 640 allowed CPU placement is changed immediately. If such a task had been
646 bound to some subset of its cpuset using the s 641 bound to some subset of its cpuset using the sched_setaffinity() call,
647 the task will be allowed to run on any CPU all 642 the task will be allowed to run on any CPU allowed in its new cpuset,
648 negating the effect of the prior sched_setaffi 643 negating the effect of the prior sched_setaffinity() call.
649 644
650 In summary, the memory placement of a task who 645 In summary, the memory placement of a task whose cpuset is changed is
651 updated by the kernel, on the next allocation 646 updated by the kernel, on the next allocation of a page for that task,
652 and the processor placement is updated immedia 647 and the processor placement is updated immediately.
653 648
654 Normally, once a page is allocated (given a ph 649 Normally, once a page is allocated (given a physical page
655 of main memory) then that page stays on whatev 650 of main memory) then that page stays on whatever node it
656 was allocated, so long as it remains allocated 651 was allocated, so long as it remains allocated, even if the
657 cpusets memory placement policy 'cpuset.mems' 652 cpusets memory placement policy 'cpuset.mems' subsequently changes.
658 If the cpuset flag file 'cpuset.memory_migrate 653 If the cpuset flag file 'cpuset.memory_migrate' is set true, then when
659 tasks are attached to that cpuset, any pages t 654 tasks are attached to that cpuset, any pages that task had
660 allocated to it on nodes in its previous cpuse 655 allocated to it on nodes in its previous cpuset are migrated
661 to the task's new cpuset. The relative placeme 656 to the task's new cpuset. The relative placement of the page within
662 the cpuset is preserved during these migration 657 the cpuset is preserved during these migration operations if possible.
663 For example if the page was on the second vali 658 For example if the page was on the second valid node of the prior cpuset
664 then the page will be placed on the second val 659 then the page will be placed on the second valid node of the new cpuset.
665 660
666 Also if 'cpuset.memory_migrate' is set true, t 661 Also if 'cpuset.memory_migrate' is set true, then if that cpuset's
667 'cpuset.mems' file is modified, pages allocate 662 'cpuset.mems' file is modified, pages allocated to tasks in that
668 cpuset, that were on nodes in the previous set 663 cpuset, that were on nodes in the previous setting of 'cpuset.mems',
669 will be moved to nodes in the new setting of ' 664 will be moved to nodes in the new setting of 'mems.'
670 Pages that were not in the task's prior cpuset 665 Pages that were not in the task's prior cpuset, or in the cpuset's
671 prior 'cpuset.mems' setting, will not be moved 666 prior 'cpuset.mems' setting, will not be moved.
672 667
673 There is an exception to the above. If hotplu 668 There is an exception to the above. If hotplug functionality is used
674 to remove all the CPUs that are currently assi 669 to remove all the CPUs that are currently assigned to a cpuset,
675 then all the tasks in that cpuset will be move 670 then all the tasks in that cpuset will be moved to the nearest ancestor
676 with non-empty cpus. But the moving of some ( 671 with non-empty cpus. But the moving of some (or all) tasks might fail if
677 cpuset is bound with another cgroup subsystem 672 cpuset is bound with another cgroup subsystem which has some restrictions
678 on task attaching. In this failing case, thos 673 on task attaching. In this failing case, those tasks will stay
679 in the original cpuset, and the kernel will au 674 in the original cpuset, and the kernel will automatically update
680 their cpus_allowed to allow all online CPUs. 675 their cpus_allowed to allow all online CPUs. When memory hotplug
681 functionality for removing Memory Nodes is ava 676 functionality for removing Memory Nodes is available, a similar exception
682 is expected to apply there as well. In genera 677 is expected to apply there as well. In general, the kernel prefers to
683 violate cpuset placement, over starving a task 678 violate cpuset placement, over starving a task that has had all
684 its allowed CPUs or Memory Nodes taken offline 679 its allowed CPUs or Memory Nodes taken offline.
685 680
686 There is a second exception to the above. GFP 681 There is a second exception to the above. GFP_ATOMIC requests are
687 kernel internal allocations that must be satis 682 kernel internal allocations that must be satisfied, immediately.
688 The kernel may drop some request, in rare case 683 The kernel may drop some request, in rare cases even panic, if a
689 GFP_ATOMIC alloc fails. If the request cannot 684 GFP_ATOMIC alloc fails. If the request cannot be satisfied within
690 the current task's cpuset, then we relax the c 685 the current task's cpuset, then we relax the cpuset, and look for
691 memory anywhere we can find it. It's better t 686 memory anywhere we can find it. It's better to violate the cpuset
692 than stress the kernel. 687 than stress the kernel.
693 688
694 To start a new job that is to be contained wit 689 To start a new job that is to be contained within a cpuset, the steps are:
695 690
696 1) mkdir /sys/fs/cgroup/cpuset 691 1) mkdir /sys/fs/cgroup/cpuset
697 2) mount -t cgroup -ocpuset cpuset /sys/fs/cg 692 2) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset
698 3) Create the new cpuset by doing mkdir's and 693 3) Create the new cpuset by doing mkdir's and write's (or echo's) in
699 the /sys/fs/cgroup/cpuset virtual file sys 694 the /sys/fs/cgroup/cpuset virtual file system.
700 4) Start a task that will be the "founding fa 695 4) Start a task that will be the "founding father" of the new job.
701 5) Attach that task to the new cpuset by writ 696 5) Attach that task to the new cpuset by writing its pid to the
702 /sys/fs/cgroup/cpuset tasks file for that 697 /sys/fs/cgroup/cpuset tasks file for that cpuset.
703 6) fork, exec or clone the job tasks from thi 698 6) fork, exec or clone the job tasks from this founding father task.
704 699
705 For example, the following sequence of command 700 For example, the following sequence of commands will setup a cpuset
706 named "Charlie", containing just CPUs 2 and 3, 701 named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
707 and then start a subshell 'sh' in that cpuset: 702 and then start a subshell 'sh' in that cpuset::
708 703
709 mount -t cgroup -ocpuset cpuset /sys/fs/cgro 704 mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset
710 cd /sys/fs/cgroup/cpuset 705 cd /sys/fs/cgroup/cpuset
711 mkdir Charlie 706 mkdir Charlie
712 cd Charlie 707 cd Charlie
713 /bin/echo 2-3 > cpuset.cpus 708 /bin/echo 2-3 > cpuset.cpus
714 /bin/echo 1 > cpuset.mems 709 /bin/echo 1 > cpuset.mems
715 /bin/echo $$ > tasks 710 /bin/echo $$ > tasks
716 sh 711 sh
717 # The subshell 'sh' is now running in cpuset 712 # The subshell 'sh' is now running in cpuset Charlie
718 # The next line should display '/Charlie' 713 # The next line should display '/Charlie'
719 cat /proc/self/cpuset 714 cat /proc/self/cpuset
720 715
721 There are ways to query or modify cpusets: 716 There are ways to query or modify cpusets:
722 717
723 - via the cpuset file system directly, using 718 - via the cpuset file system directly, using the various cd, mkdir, echo,
724 cat, rmdir commands from the shell, or thei 719 cat, rmdir commands from the shell, or their equivalent from C.
725 - via the C library libcpuset. 720 - via the C library libcpuset.
726 - via the C library libcgroup. 721 - via the C library libcgroup.
727 (https://github.com/libcgroup/libcgroup/) 722 (https://github.com/libcgroup/libcgroup/)
728 - via the python application cset. 723 - via the python application cset.
729 (http://code.google.com/p/cpuset/) 724 (http://code.google.com/p/cpuset/)
730 725
731 The sched_setaffinity calls can also be done a 726 The sched_setaffinity calls can also be done at the shell prompt using
732 SGI's runon or Robert Love's taskset. The mbi 727 SGI's runon or Robert Love's taskset. The mbind and set_mempolicy
733 calls can be done at the shell prompt using th 728 calls can be done at the shell prompt using the numactl command
734 (part of Andi Kleen's numa package). 729 (part of Andi Kleen's numa package).
735 730
736 2. Usage Examples and Syntax 731 2. Usage Examples and Syntax
737 ============================ 732 ============================
738 733
739 2.1 Basic Usage 734 2.1 Basic Usage
740 --------------- 735 ---------------
741 736
742 Creating, modifying, using the cpusets can be 737 Creating, modifying, using the cpusets can be done through the cpuset
743 virtual filesystem. 738 virtual filesystem.
744 739
745 To mount it, type: 740 To mount it, type:
746 # mount -t cgroup -o cpuset cpuset /sys/fs/cgr 741 # mount -t cgroup -o cpuset cpuset /sys/fs/cgroup/cpuset
747 742
748 Then under /sys/fs/cgroup/cpuset you can find 743 Then under /sys/fs/cgroup/cpuset you can find a tree that corresponds to the
749 tree of the cpusets in the system. For instanc 744 tree of the cpusets in the system. For instance, /sys/fs/cgroup/cpuset
750 is the cpuset that holds the whole system. 745 is the cpuset that holds the whole system.
751 746
752 If you want to create a new cpuset under /sys/ 747 If you want to create a new cpuset under /sys/fs/cgroup/cpuset::
753 748
754 # cd /sys/fs/cgroup/cpuset 749 # cd /sys/fs/cgroup/cpuset
755 # mkdir my_cpuset 750 # mkdir my_cpuset
756 751
757 Now you want to do something with this cpuset: 752 Now you want to do something with this cpuset::
758 753
759 # cd my_cpuset 754 # cd my_cpuset
760 755
761 In this directory you can find several files:: 756 In this directory you can find several files::
762 757
763 # ls 758 # ls
764 cgroup.clone_children cpuset.memory_pressur 759 cgroup.clone_children cpuset.memory_pressure
765 cgroup.event_control cpuset.memory_spread_ 760 cgroup.event_control cpuset.memory_spread_page
766 cgroup.procs cpuset.memory_spread_ 761 cgroup.procs cpuset.memory_spread_slab
767 cpuset.cpu_exclusive cpuset.mems 762 cpuset.cpu_exclusive cpuset.mems
768 cpuset.cpus cpuset.sched_load_bal 763 cpuset.cpus cpuset.sched_load_balance
769 cpuset.mem_exclusive cpuset.sched_relax_do 764 cpuset.mem_exclusive cpuset.sched_relax_domain_level
770 cpuset.mem_hardwall notify_on_release 765 cpuset.mem_hardwall notify_on_release
771 cpuset.memory_migrate tasks 766 cpuset.memory_migrate tasks
772 767
773 Reading them will give you information about t 768 Reading them will give you information about the state of this cpuset:
774 the CPUs and Memory Nodes it can use, the proc 769 the CPUs and Memory Nodes it can use, the processes that are using
775 it, its properties. By writing to these files 770 it, its properties. By writing to these files you can manipulate
776 the cpuset. 771 the cpuset.
777 772
778 Set some flags:: 773 Set some flags::
779 774
780 # /bin/echo 1 > cpuset.cpu_exclusive 775 # /bin/echo 1 > cpuset.cpu_exclusive
781 776
782 Add some cpus:: 777 Add some cpus::
783 778
784 # /bin/echo 0-7 > cpuset.cpus 779 # /bin/echo 0-7 > cpuset.cpus
785 780
786 Add some mems:: 781 Add some mems::
787 782
788 # /bin/echo 0-7 > cpuset.mems 783 # /bin/echo 0-7 > cpuset.mems
789 784
790 Now attach your shell to this cpuset:: 785 Now attach your shell to this cpuset::
791 786
792 # /bin/echo $$ > tasks 787 # /bin/echo $$ > tasks
793 788
794 You can also create cpusets inside your cpuset 789 You can also create cpusets inside your cpuset by using mkdir in this
795 directory:: 790 directory::
796 791
797 # mkdir my_sub_cs 792 # mkdir my_sub_cs
798 793
799 To remove a cpuset, just use rmdir:: 794 To remove a cpuset, just use rmdir::
800 795
801 # rmdir my_sub_cs 796 # rmdir my_sub_cs
802 797
803 This will fail if the cpuset is in use (has cp 798 This will fail if the cpuset is in use (has cpusets inside, or has
804 processes attached). 799 processes attached).
805 800
806 Note that for legacy reasons, the "cpuset" fil 801 Note that for legacy reasons, the "cpuset" filesystem exists as a
807 wrapper around the cgroup filesystem. 802 wrapper around the cgroup filesystem.
808 803
809 The command:: 804 The command::
810 805
811 mount -t cpuset X /sys/fs/cgroup/cpuset 806 mount -t cpuset X /sys/fs/cgroup/cpuset
812 807
813 is equivalent to:: 808 is equivalent to::
814 809
815 mount -t cgroup -ocpuset,noprefix X /sys/fs/ 810 mount -t cgroup -ocpuset,noprefix X /sys/fs/cgroup/cpuset
816 echo "/sbin/cpuset_release_agent" > /sys/fs/ 811 echo "/sbin/cpuset_release_agent" > /sys/fs/cgroup/cpuset/release_agent
817 812
818 2.2 Adding/removing cpus 813 2.2 Adding/removing cpus
819 ------------------------ 814 ------------------------
820 815
821 This is the syntax to use when writing in the 816 This is the syntax to use when writing in the cpus or mems files
822 in cpuset directories:: 817 in cpuset directories::
823 818
824 # /bin/echo 1-4 > cpuset.cpus -> set 819 # /bin/echo 1-4 > cpuset.cpus -> set cpus list to cpus 1,2,3,4
825 # /bin/echo 1,2,3,4 > cpuset.cpus -> set 820 # /bin/echo 1,2,3,4 > cpuset.cpus -> set cpus list to cpus 1,2,3,4
826 821
827 To add a CPU to a cpuset, write the new list o 822 To add a CPU to a cpuset, write the new list of CPUs including the
828 CPU to be added. To add 6 to the above cpuset: 823 CPU to be added. To add 6 to the above cpuset::
829 824
830 # /bin/echo 1-4,6 > cpuset.cpus -> set 825 # /bin/echo 1-4,6 > cpuset.cpus -> set cpus list to cpus 1,2,3,4,6
831 826
832 Similarly to remove a CPU from a cpuset, write 827 Similarly to remove a CPU from a cpuset, write the new list of CPUs
833 without the CPU to be removed. 828 without the CPU to be removed.
834 829
835 To remove all the CPUs:: 830 To remove all the CPUs::
836 831
837 # /bin/echo "" > cpuset.cpus -> cle 832 # /bin/echo "" > cpuset.cpus -> clear cpus list
838 833
839 2.3 Setting flags 834 2.3 Setting flags
840 ----------------- 835 -----------------
841 836
842 The syntax is very simple:: 837 The syntax is very simple::
843 838
844 # /bin/echo 1 > cpuset.cpu_exclusive -> set 839 # /bin/echo 1 > cpuset.cpu_exclusive -> set flag 'cpuset.cpu_exclusive'
845 # /bin/echo 0 > cpuset.cpu_exclusive -> uns 840 # /bin/echo 0 > cpuset.cpu_exclusive -> unset flag 'cpuset.cpu_exclusive'
846 841
847 2.4 Attaching processes 842 2.4 Attaching processes
848 ----------------------- 843 -----------------------
849 844
850 :: 845 ::
851 846
852 # /bin/echo PID > tasks 847 # /bin/echo PID > tasks
853 848
854 Note that it is PID, not PIDs. You can only at 849 Note that it is PID, not PIDs. You can only attach ONE task at a time.
855 If you have several tasks to attach, you have 850 If you have several tasks to attach, you have to do it one after another::
856 851
857 # /bin/echo PID1 > tasks 852 # /bin/echo PID1 > tasks
858 # /bin/echo PID2 > tasks 853 # /bin/echo PID2 > tasks
859 ... 854 ...
860 # /bin/echo PIDn > tasks 855 # /bin/echo PIDn > tasks
861 856
862 857
863 3. Questions 858 3. Questions
864 ============ 859 ============
865 860
866 Q: 861 Q:
867 what's up with this '/bin/echo' ? 862 what's up with this '/bin/echo' ?
868 863
869 A: 864 A:
870 bash's builtin 'echo' command does not chec 865 bash's builtin 'echo' command does not check calls to write() against
871 errors. If you use it in the cpuset file sy 866 errors. If you use it in the cpuset file system, you won't be
872 able to tell whether a command succeeded or 867 able to tell whether a command succeeded or failed.
873 868
874 Q: 869 Q:
875 When I attach processes, only the first of 870 When I attach processes, only the first of the line gets really attached !
876 871
877 A: 872 A:
878 We can only return one error code per call 873 We can only return one error code per call to write(). So you should also
879 put only ONE pid. 874 put only ONE pid.
880 875
881 4. Contact 876 4. Contact
882 ========== 877 ==========
883 878
884 Web: http://www.bullopensource.org/cpuset 879 Web: http://www.bullopensource.org/cpuset
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.