blob: 47e568a9370afa28f703acf9ce55f88a8ca52519 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001 CPUSETS
2 -------
3
4Copyright (C) 2004 BULL SA.
5Written by Simon.Derr@bull.net
6
Christoph Lameterb4fb3762006-03-14 19:50:20 -08007Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.
Linus Torvalds1da177e2005-04-16 15:20:36 -07008Modified by Paul Jackson <pj@sgi.com>
Christoph Lameterb4fb3762006-03-14 19:50:20 -08009Modified by Christoph Lameter <clameter@sgi.com>
Paul Menage8793d852007-10-18 23:39:39 -070010Modified by Paul Menage <menage@google.com>
Hidetoshi Seto4d5f3552008-04-15 14:03:17 +090011Modified by Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012
13CONTENTS:
14=========
15
161. Cpusets
17 1.1 What are cpusets ?
18 1.2 Why are cpusets needed ?
19 1.3 How are cpusets implemented ?
Paul Jacksonbd5e09c2006-01-08 01:01:50 -080020 1.4 What are exclusive cpusets ?
Paul Menage8793d852007-10-18 23:39:39 -070021 1.5 What is memory_pressure ?
22 1.6 What is memory spread ?
Paul Jackson029190c2007-10-18 23:40:20 -070023 1.7 What is sched_load_balance ?
Hidetoshi Seto4d5f3552008-04-15 14:03:17 +090024 1.8 What is sched_relax_domain_level ?
25 1.9 How do I use cpusets ?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262. Usage Examples and Syntax
27 2.1 Basic Usage
28 2.2 Adding/removing cpus
29 2.3 Setting flags
30 2.4 Attaching processes
313. Questions
324. Contact
33
341. Cpusets
35==========
36
371.1 What are cpusets ?
38----------------------
39
40Cpusets provide a mechanism for assigning a set of CPUs and Memory
Christoph Lameter0e1e7c72007-10-16 01:25:38 -070041Nodes to a set of tasks. In this document "Memory Node" refers to
42an on-line node that contains memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
44Cpusets constrain the CPU and Memory placement of tasks to only
45the resources within a tasks current cpuset. They form a nested
46hierarchy visible in a virtual file system. These are the essential
47hooks, beyond what is already present, required to manage dynamic
48job placement on large systems.
49
Paul Menage8793d852007-10-18 23:39:39 -070050Cpusets use the generic cgroup subsystem described in
51Documentation/cgroup.txt.
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Paul Menage8793d852007-10-18 23:39:39 -070053Requests by a task, using the sched_setaffinity(2) system call to
54include CPUs in its CPU affinity mask, and using the mbind(2) and
55set_mempolicy(2) system calls to include Memory Nodes in its memory
56policy, are both filtered through that tasks cpuset, filtering out any
57CPUs or Memory Nodes not in that cpuset. The scheduler will not
58schedule a task on a CPU that is not allowed in its cpus_allowed
59vector, and the kernel page allocator will not allocate a page on a
60node that is not allowed in the requesting tasks mems_allowed vector.
61
62User level code may create and destroy cpusets by name in the cgroup
Linus Torvalds1da177e2005-04-16 15:20:36 -070063virtual file system, manage the attributes and permissions of these
64cpusets and which CPUs and Memory Nodes are assigned to each cpuset,
65specify and query to which cpuset a task is assigned, and list the
66task pids assigned to a cpuset.
67
68
691.2 Why are cpusets needed ?
70----------------------------
71
72The management of large computer systems, with many processors (CPUs),
73complex memory cache hierarchies and multiple Memory Nodes having
74non-uniform access times (NUMA) presents additional challenges for
75the efficient scheduling and memory placement of processes.
76
77Frequently more modest sized systems can be operated with adequate
78efficiency just by letting the operating system automatically share
79the available CPU and Memory resources amongst the requesting tasks.
80
81But larger systems, which benefit more from careful processor and
82memory placement to reduce memory access times and contention,
83and which typically represent a larger investment for the customer,
Jean Delvare33430dc2005-10-30 15:02:20 -080084can benefit from explicitly placing jobs on properly sized subsets of
Linus Torvalds1da177e2005-04-16 15:20:36 -070085the system.
86
87This can be especially valuable on:
88
89 * Web Servers running multiple instances of the same web application,
90 * Servers running different applications (for instance, a web server
91 and a database), or
92 * NUMA systems running large HPC applications with demanding
93 performance characteristics.
94
95These subsets, or "soft partitions" must be able to be dynamically
96adjusted, as the job mix changes, without impacting other concurrently
Christoph Lameterb4fb3762006-03-14 19:50:20 -080097executing jobs. The location of the running jobs pages may also be moved
98when the memory locations are changed.
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100The kernel cpuset patch provides the minimum essential kernel
101mechanisms required to efficiently implement such subsets. It
102leverages existing CPU and Memory Placement facilities in the Linux
103kernel to avoid any additional impact on the critical scheduler or
104memory allocator code.
105
106
1071.3 How are cpusets implemented ?
108---------------------------------
109
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800110Cpusets provide a Linux kernel mechanism to constrain which CPUs and
111Memory Nodes are used by a process or set of processes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
113The Linux kernel already has a pair of mechanisms to specify on which
114CPUs a task may be scheduled (sched_setaffinity) and on which Memory
115Nodes it may obtain memory (mbind, set_mempolicy).
116
117Cpusets extends these two mechanisms as follows:
118
119 - Cpusets are sets of allowed CPUs and Memory Nodes, known to the
120 kernel.
121 - Each task in the system is attached to a cpuset, via a pointer
Paul Menage8793d852007-10-18 23:39:39 -0700122 in the task structure to a reference counted cgroup structure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 - Calls to sched_setaffinity are filtered to just those CPUs
124 allowed in that tasks cpuset.
125 - Calls to mbind and set_mempolicy are filtered to just
126 those Memory Nodes allowed in that tasks cpuset.
127 - The root cpuset contains all the systems CPUs and Memory
128 Nodes.
129 - For any cpuset, one can define child cpusets containing a subset
130 of the parents CPU and Memory Node resources.
131 - The hierarchy of cpusets can be mounted at /dev/cpuset, for
132 browsing and manipulation from user space.
133 - A cpuset may be marked exclusive, which ensures that no other
134 cpuset (except direct ancestors and descendents) may contain
135 any overlapping CPUs or Memory Nodes.
136 - You can list all the tasks (by pid) attached to any cpuset.
137
138The implementation of cpusets requires a few, simple hooks
139into the rest of the kernel, none in performance critical paths:
140
Paul Jackson864913f2006-01-11 02:01:38 +0100141 - in init/main.c, to initialize the root cpuset at system boot.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 - in fork and exit, to attach and detach a task from its cpuset.
143 - in sched_setaffinity, to mask the requested CPUs by what's
144 allowed in that tasks cpuset.
145 - in sched.c migrate_all_tasks(), to keep migrating tasks within
146 the CPUs allowed by their cpuset, if possible.
147 - in the mbind and set_mempolicy system calls, to mask the requested
148 Memory Nodes by what's allowed in that tasks cpuset.
Paul Jackson864913f2006-01-11 02:01:38 +0100149 - in page_alloc.c, to restrict memory to allowed nodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 - in vmscan.c, to restrict page recovery to the current cpuset.
151
Paul Menage8793d852007-10-18 23:39:39 -0700152You should mount the "cgroup" filesystem type in order to enable
153browsing and modifying the cpusets presently known to the kernel. No
154new system calls are added for cpusets - all support for querying and
155modifying cpusets is via this cpuset file system.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
Paul Jackson985ee7f2008-07-04 10:00:01 -0700157The /proc/<pid>/status file for each task has four added lines,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158displaying the tasks cpus_allowed (on which CPUs it may be scheduled)
159and mems_allowed (on which Memory Nodes it may obtain memory),
Paul Jackson985ee7f2008-07-04 10:00:01 -0700160in the two formats seen in the following example:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161
162 Cpus_allowed: ffffffff,ffffffff,ffffffff,ffffffff
Paul Jackson985ee7f2008-07-04 10:00:01 -0700163 Cpus_allowed_list: 0-127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 Mems_allowed: ffffffff,ffffffff
Paul Jackson985ee7f2008-07-04 10:00:01 -0700165 Mems_allowed_list: 0-63
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Paul Menage8793d852007-10-18 23:39:39 -0700167Each cpuset is represented by a directory in the cgroup file system
168containing (on top of the standard cgroup files) the following
169files describing that cpuset:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 - cpus: list of CPUs in that cpuset
172 - mems: list of Memory Nodes in that cpuset
Paul Jackson45b07ef2006-01-08 01:00:56 -0800173 - memory_migrate flag: if set, move pages to cpusets nodes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 - cpu_exclusive flag: is cpu placement exclusive?
175 - mem_exclusive flag: is memory placement exclusive?
Paul Menage78608362008-04-29 01:00:26 -0700176 - mem_hardwall flag: is memory allocation hardwalled
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800177 - memory_pressure: measure of how much paging pressure in cpuset
178
179In addition, the root cpuset only has the following file:
180 - memory_pressure_enabled flag: compute memory_pressure?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
182New cpusets are created using the mkdir system call or shell
183command. The properties of a cpuset, such as its flags, allowed
184CPUs and Memory Nodes, and attached tasks, are modified by writing
185to the appropriate file in that cpusets directory, as listed above.
186
187The named hierarchical structure of nested cpusets allows partitioning
188a large system into nested, dynamically changeable, "soft-partitions".
189
190The attachment of each task, automatically inherited at fork by any
191children of that task, to a cpuset allows organizing the work load
192on a system into related sets of tasks such that each set is constrained
193to using the CPUs and Memory Nodes of a particular cpuset. A task
194may be re-attached to any other cpuset, if allowed by the permissions
195on the necessary cpuset file system directories.
196
197Such management of a system "in the large" integrates smoothly with
198the detailed placement done on individual tasks and memory regions
199using the sched_setaffinity, mbind and set_mempolicy system calls.
200
201The following rules apply to each cpuset:
202
203 - Its CPUs and Memory Nodes must be a subset of its parents.
Miao Xie6a7d68e2008-06-05 22:45:54 -0700204 - It can't be marked exclusive unless its parent is.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 - If its cpu or memory is exclusive, they may not overlap any sibling.
206
207These rules, and the natural hierarchy of cpusets, enable efficient
208enforcement of the exclusive guarantee, without having to scan all
209cpusets every time any of them change to ensure nothing overlaps a
210exclusive cpuset. Also, the use of a Linux virtual file system (vfs)
211to represent the cpuset hierarchy provides for a familiar permission
212and name space for cpusets, with a minimum of additional kernel code.
213
Paul Jackson38837fc2006-09-29 02:01:16 -0700214The cpus and mems files in the root (top_cpuset) cpuset are
215read-only. The cpus file automatically tracks the value of
216cpu_online_map using a CPU hotplug notifier, and the mems file
KOSAKI Motohiro0b720372008-02-23 15:23:41 -0800217automatically tracks the value of node_states[N_HIGH_MEMORY]--i.e.,
Christoph Lameter0e1e7c72007-10-16 01:25:38 -0700218nodes with memory--using the cpuset_track_online_nodes() hook.
Paul Jackson4c4d50f2006-08-27 01:23:51 -0700219
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800220
2211.4 What are exclusive cpusets ?
222--------------------------------
223
224If a cpuset is cpu or mem exclusive, no other cpuset, other than
225a direct ancestor or descendent, may share any of the same CPUs or
226Memory Nodes.
227
Paul Menage78608362008-04-29 01:00:26 -0700228A cpuset that is mem_exclusive *or* mem_hardwall is "hardwalled",
229i.e. it restricts kernel allocations for page, buffer and other data
230commonly shared by the kernel across multiple users. All cpusets,
231whether hardwalled or not, restrict allocations of memory for user
232space. This enables configuring a system so that several independent
233jobs can share common kernel data, such as file system pages, while
234isolating each job's user allocation in its own cpuset. To do this,
235construct a large mem_exclusive cpuset to hold all the jobs, and
236construct child, non-mem_exclusive cpusets for each individual job.
237Only a small amount of typical kernel memory, such as requests from
238interrupt handlers, is allowed to be taken outside even a
239mem_exclusive cpuset.
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800240
241
Paul Menage8793d852007-10-18 23:39:39 -07002421.5 What is memory_pressure ?
Paul Jacksonbd5e09c2006-01-08 01:01:50 -0800243-----------------------------
244The memory_pressure of a cpuset provides a simple per-cpuset metric
245of the rate that the tasks in a cpuset are attempting to free up in
246use memory on the nodes of the cpuset to satisfy additional memory
247requests.
248
249This enables batch managers monitoring jobs running in dedicated
250cpusets to efficiently detect what level of memory pressure that job
251is causing.
252
253This is useful both on tightly managed systems running a wide mix of
254submitted jobs, which may choose to terminate or re-prioritize jobs that
255are trying to use more memory than allowed on the nodes assigned them,
256and with tightly coupled, long running, massively parallel scientific
257computing jobs that will dramatically fail to meet required performance
258goals if they start to use more memory than allowed to them.
259
260This mechanism provides a very economical way for the batch manager
261to monitor a cpuset for signs of memory pressure. It's up to the
262batch manager or other user code to decide what to do about it and
263take action.
264
265==> Unless this feature is enabled by writing "1" to the special file
266 /dev/cpuset/memory_pressure_enabled, the hook in the rebalance
267 code of __alloc_pages() for this metric reduces to simply noticing
268 that the cpuset_memory_pressure_enabled flag is zero. So only
269 systems that enable this feature will compute the metric.
270
271Why a per-cpuset, running average:
272
273 Because this meter is per-cpuset, rather than per-task or mm,
274 the system load imposed by a batch scheduler monitoring this
275 metric is sharply reduced on large systems, because a scan of
276 the tasklist can be avoided on each set of queries.
277
278 Because this meter is a running average, instead of an accumulating
279 counter, a batch scheduler can detect memory pressure with a
280 single read, instead of having to read and accumulate results
281 for a period of time.
282
283 Because this meter is per-cpuset rather than per-task or mm,
284 the batch scheduler can obtain the key information, memory
285 pressure in a cpuset, with a single read, rather than having to
286 query and accumulate results over all the (dynamically changing)
287 set of tasks in the cpuset.
288
289A per-cpuset simple digital filter (requires a spinlock and 3 words
290of data per-cpuset) is kept, and updated by any task attached to that
291cpuset, if it enters the synchronous (direct) page reclaim code.
292
293A per-cpuset file provides an integer number representing the recent
294(half-life of 10 seconds) rate of direct page reclaims caused by
295the tasks in the cpuset, in units of reclaims attempted per second,
296times 1000.
297
298
Paul Menage8793d852007-10-18 23:39:39 -07002991.6 What is memory spread ?
Paul Jackson825a46a2006-03-24 03:16:03 -0800300---------------------------
301There are two boolean flag files per cpuset that control where the
302kernel allocates pages for the file system buffers and related in
303kernel data structures. They are called 'memory_spread_page' and
304'memory_spread_slab'.
305
306If the per-cpuset boolean flag file 'memory_spread_page' is set, then
307the kernel will spread the file system buffers (page cache) evenly
308over all the nodes that the faulting task is allowed to use, instead
309of preferring to put those pages on the node where the task is running.
310
311If the per-cpuset boolean flag file 'memory_spread_slab' is set,
312then the kernel will spread some file system related slab caches,
313such as for inodes and dentries evenly over all the nodes that the
314faulting task is allowed to use, instead of preferring to put those
315pages on the node where the task is running.
316
317The setting of these flags does not affect anonymous data segment or
318stack segment pages of a task.
319
320By default, both kinds of memory spreading are off, and memory
321pages are allocated on the node local to where the task is running,
322except perhaps as modified by the tasks NUMA mempolicy or cpuset
323configuration, so long as sufficient free memory pages are available.
324
325When new cpusets are created, they inherit the memory spread settings
326of their parent.
327
328Setting memory spreading causes allocations for the affected page
329or slab caches to ignore the tasks NUMA mempolicy and be spread
330instead. Tasks using mbind() or set_mempolicy() calls to set NUMA
331mempolicies will not notice any change in these calls as a result of
332their containing tasks memory spread settings. If memory spreading
333is turned off, then the currently specified NUMA mempolicy once again
334applies to memory page allocations.
335
336Both 'memory_spread_page' and 'memory_spread_slab' are boolean flag
337files. By default they contain "0", meaning that the feature is off
338for that cpuset. If a "1" is written to that file, then that turns
339the named feature on.
340
341The implementation is simple.
342
343Setting the flag 'memory_spread_page' turns on a per-process flag
344PF_SPREAD_PAGE for each task that is in that cpuset or subsequently
345joins that cpuset. The page allocation calls for the page cache
346is modified to perform an inline check for this PF_SPREAD_PAGE task
347flag, and if set, a call to a new routine cpuset_mem_spread_node()
348returns the node to prefer for the allocation.
349
Miao Xie6a7d68e2008-06-05 22:45:54 -0700350Similarly, setting 'memory_spread_slab' turns on the flag
Paul Jackson825a46a2006-03-24 03:16:03 -0800351PF_SPREAD_SLAB, and appropriately marked slab caches will allocate
352pages from the node returned by cpuset_mem_spread_node().
353
354The cpuset_mem_spread_node() routine is also simple. It uses the
355value of a per-task rotor cpuset_mem_spread_rotor to select the next
356node in the current tasks mems_allowed to prefer for the allocation.
357
358This memory placement policy is also known (in other contexts) as
359round-robin or interleave.
360
361This policy can provide substantial improvements for jobs that need
362to place thread local data on the corresponding node, but that need
363to access large file system data sets that need to be spread across
364the several nodes in the jobs cpuset in order to fit. Without this
365policy, especially for jobs that might have one thread reading in the
366data set, the memory allocation across the nodes in the jobs cpuset
367can become very uneven.
368
Paul Jackson029190c2007-10-18 23:40:20 -07003691.7 What is sched_load_balance ?
370--------------------------------
Paul Jackson825a46a2006-03-24 03:16:03 -0800371
Paul Jackson029190c2007-10-18 23:40:20 -0700372The kernel scheduler (kernel/sched.c) automatically load balances
373tasks. If one CPU is underutilized, kernel code running on that
374CPU will look for tasks on other more overloaded CPUs and move those
375tasks to itself, within the constraints of such placement mechanisms
376as cpusets and sched_setaffinity.
377
378The algorithmic cost of load balancing and its impact on key shared
379kernel data structures such as the task list increases more than
380linearly with the number of CPUs being balanced. So the scheduler
381has support to partition the systems CPUs into a number of sched
382domains such that it only load balances within each sched domain.
383Each sched domain covers some subset of the CPUs in the system;
384no two sched domains overlap; some CPUs might not be in any sched
385domain and hence won't be load balanced.
386
387Put simply, it costs less to balance between two smaller sched domains
388than one big one, but doing so means that overloads in one of the
389two domains won't be load balanced to the other one.
390
391By default, there is one sched domain covering all CPUs, except those
392marked isolated using the kernel boot time "isolcpus=" argument.
393
394This default load balancing across all CPUs is not well suited for
395the following two situations:
396 1) On large systems, load balancing across many CPUs is expensive.
397 If the system is managed using cpusets to place independent jobs
398 on separate sets of CPUs, full load balancing is unnecessary.
399 2) Systems supporting realtime on some CPUs need to minimize
400 system overhead on those CPUs, including avoiding task load
401 balancing if that is not needed.
402
403When the per-cpuset flag "sched_load_balance" is enabled (the default
404setting), it requests that all the CPUs in that cpusets allowed 'cpus'
405be contained in a single sched domain, ensuring that load balancing
406can move a task (not otherwised pinned, as by sched_setaffinity)
407from any CPU in that cpuset to any other.
408
409When the per-cpuset flag "sched_load_balance" is disabled, then the
410scheduler will avoid load balancing across the CPUs in that cpuset,
411--except-- in so far as is necessary because some overlapping cpuset
412has "sched_load_balance" enabled.
413
414So, for example, if the top cpuset has the flag "sched_load_balance"
415enabled, then the scheduler will have one sched domain covering all
416CPUs, and the setting of the "sched_load_balance" flag in any other
417cpusets won't matter, as we're already fully load balancing.
418
419Therefore in the above two situations, the top cpuset flag
420"sched_load_balance" should be disabled, and only some of the smaller,
421child cpusets have this flag enabled.
422
423When doing this, you don't usually want to leave any unpinned tasks in
424the top cpuset that might use non-trivial amounts of CPU, as such tasks
425may be artificially constrained to some subset of CPUs, depending on
426the particulars of this flag setting in descendent cpusets. Even if
427such a task could use spare CPU cycles in some other CPUs, the kernel
428scheduler might not consider the possibility of load balancing that
429task to that underused CPU.
430
431Of course, tasks pinned to a particular CPU can be left in a cpuset
432that disables "sched_load_balance" as those tasks aren't going anywhere
433else anyway.
434
435There is an impedance mismatch here, between cpusets and sched domains.
436Cpusets are hierarchical and nest. Sched domains are flat; they don't
437overlap and each CPU is in at most one sched domain.
438
439It is necessary for sched domains to be flat because load balancing
440across partially overlapping sets of CPUs would risk unstable dynamics
441that would be beyond our understanding. So if each of two partially
442overlapping cpusets enables the flag 'sched_load_balance', then we
443form a single sched domain that is a superset of both. We won't move
444a task to a CPU outside it cpuset, but the scheduler load balancing
445code might waste some compute cycles considering that possibility.
446
447This mismatch is why there is not a simple one-to-one relation
448between which cpusets have the flag "sched_load_balance" enabled,
449and the sched domain configuration. If a cpuset enables the flag, it
450will get balancing across all its CPUs, but if it disables the flag,
451it will only be assured of no load balancing if no other overlapping
452cpuset enables the flag.
453
454If two cpusets have partially overlapping 'cpus' allowed, and only
455one of them has this flag enabled, then the other may find its
456tasks only partially load balanced, just on the overlapping CPUs.
457This is just the general case of the top_cpuset example given a few
458paragraphs above. In the general case, as in the top cpuset case,
459don't leave tasks that might use non-trivial amounts of CPU in
460such partially load balanced cpusets, as they may be artificially
461constrained to some subset of the CPUs allowed to them, for lack of
462load balancing to the other CPUs.
463
4641.7.1 sched_load_balance implementation details.
465------------------------------------------------
466
467The per-cpuset flag 'sched_load_balance' defaults to enabled (contrary
468to most cpuset flags.) When enabled for a cpuset, the kernel will
469ensure that it can load balance across all the CPUs in that cpuset
470(makes sure that all the CPUs in the cpus_allowed of that cpuset are
471in the same sched domain.)
472
473If two overlapping cpusets both have 'sched_load_balance' enabled,
474then they will be (must be) both in the same sched domain.
475
476If, as is the default, the top cpuset has 'sched_load_balance' enabled,
477then by the above that means there is a single sched domain covering
478the whole system, regardless of any other cpuset settings.
479
480The kernel commits to user space that it will avoid load balancing
481where it can. It will pick as fine a granularity partition of sched
482domains as it can while still providing load balancing for any set
483of CPUs allowed to a cpuset having 'sched_load_balance' enabled.
484
485The internal kernel cpuset to scheduler interface passes from the
486cpuset code to the scheduler code a partition of the load balanced
487CPUs in the system. This partition is a set of subsets (represented
488as an array of cpumask_t) of CPUs, pairwise disjoint, that cover all
489the CPUs that must be load balanced.
490
491Whenever the 'sched_load_balance' flag changes, or CPUs come or go
492from a cpuset with this flag enabled, or a cpuset with this flag
493enabled is removed, the cpuset code builds a new such partition and
494passes it to the scheduler sched domain setup code, to have the sched
495domains rebuilt as necessary.
496
497This partition exactly defines what sched domains the scheduler should
498setup - one sched domain for each element (cpumask_t) in the partition.
499
500The scheduler remembers the currently active sched domain partitions.
501When the scheduler routine partition_sched_domains() is invoked from
502the cpuset code to update these sched domains, it compares the new
503partition requested with the current, and updates its sched domains,
504removing the old and adding the new, for each change.
505
Hidetoshi Seto4d5f3552008-04-15 14:03:17 +0900506
5071.8 What is sched_relax_domain_level ?
508--------------------------------------
509
510In sched domain, the scheduler migrates tasks in 2 ways; periodic load
511balance on tick, and at time of some schedule events.
512
513When a task is woken up, scheduler try to move the task on idle CPU.
514For example, if a task A running on CPU X activates another task B
515on the same CPU X, and if CPU Y is X's sibling and performing idle,
516then scheduler migrate task B to CPU Y so that task B can start on
517CPU Y without waiting task A on CPU X.
518
519And if a CPU run out of tasks in its runqueue, the CPU try to pull
520extra tasks from other busy CPUs to help them before it is going to
521be idle.
522
523Of course it takes some searching cost to find movable tasks and/or
524idle CPUs, the scheduler might not search all CPUs in the domain
525everytime. In fact, in some architectures, the searching ranges on
526events are limited in the same socket or node where the CPU locates,
527while the load balance on tick searchs all.
528
529For example, assume CPU Z is relatively far from CPU X. Even if CPU Z
530is idle while CPU X and the siblings are busy, scheduler can't migrate
531woken task B from X to Z since it is out of its searching range.
532As the result, task B on CPU X need to wait task A or wait load balance
533on the next tick. For some applications in special situation, waiting
5341 tick may be too long.
535
536The 'sched_relax_domain_level' file allows you to request changing
537this searching range as you like. This file takes int value which
538indicates size of searching range in levels ideally as follows,
539otherwise initial value -1 that indicates the cpuset has no request.
540
541 -1 : no request. use system default or follow request of others.
542 0 : no search.
543 1 : search siblings (hyperthreads in a core).
544 2 : search cores in a package.
545 3 : search cpus in a node [= system wide on non-NUMA system]
546 ( 4 : search nodes in a chunk of node [on NUMA system] )
Li Zefan30e0e172008-05-13 10:27:17 +0800547 ( 5 : search system wide [on NUMA system] )
Hidetoshi Seto4d5f3552008-04-15 14:03:17 +0900548
Paul Jackson46b6d942008-07-04 10:00:09 -0700549The system default is architecture dependent. The system default
550can be changed using the relax_domain_level= boot parameter.
551
Hidetoshi Seto4d5f3552008-04-15 14:03:17 +0900552This file is per-cpuset and affect the sched domain where the cpuset
553belongs to. Therefore if the flag 'sched_load_balance' of a cpuset
554is disabled, then 'sched_relax_domain_level' have no effect since
555there is no sched domain belonging the cpuset.
556
557If multiple cpusets are overlapping and hence they form a single sched
558domain, the largest value among those is used. Be careful, if one
559requests 0 and others are -1 then 0 is used.
560
561Note that modifying this file will have both good and bad effects,
562and whether it is acceptable or not will be depend on your situation.
563Don't modify this file if you are not sure.
564
565If your situation is:
566 - The migration costs between each cpu can be assumed considerably
567 small(for you) due to your special application's behavior or
568 special hardware support for CPU cache etc.
569 - The searching cost doesn't have impact(for you) or you can make
570 the searching cost enough small by managing cpuset to compact etc.
571 - The latency is required even it sacrifices cache hit rate etc.
572then increasing 'sched_relax_domain_level' would benefit you.
573
574
5751.9 How do I use cpusets ?
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576--------------------------
577
578In order to minimize the impact of cpusets on critical kernel
579code, such as the scheduler, and due to the fact that the kernel
580does not support one task updating the memory placement of another
581task directly, the impact on a task of changing its cpuset CPU
582or Memory Node placement, or of changing to which cpuset a task
583is attached, is subtle.
584
585If a cpuset has its Memory Nodes modified, then for each task attached
586to that cpuset, the next time that the kernel attempts to allocate
587a page of memory for that task, the kernel will notice the change
588in the tasks cpuset, and update its per-task memory placement to
589remain within the new cpusets memory placement. If the task was using
590mempolicy MPOL_BIND, and the nodes to which it was bound overlap with
591its new cpuset, then the task will continue to use whatever subset
592of MPOL_BIND nodes are still allowed in the new cpuset. If the task
593was using MPOL_BIND and now none of its MPOL_BIND nodes are allowed
594in the new cpuset, then the task will be essentially treated as if it
595was MPOL_BIND bound to the new cpuset (even though its numa placement,
596as queried by get_mempolicy(), doesn't change). If a task is moved
597from one cpuset to another, then the kernel will adjust the tasks
598memory placement, as above, the next time that the kernel attempts
599to allocate a page of memory for that task.
600
Paul Jackson8f5aa262008-02-07 00:14:48 -0800601If a cpuset has its 'cpus' modified, then each task in that cpuset
602will have its allowed CPU placement changed immediately. Similarly,
603if a tasks pid is written to a cpusets 'tasks' file, in either its
604current cpuset or another cpuset, then its allowed CPU placement is
605changed immediately. If such a task had been bound to some subset
606of its cpuset using the sched_setaffinity() call, the task will be
607allowed to run on any CPU allowed in its new cpuset, negating the
608affect of the prior sched_setaffinity() call.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609
610In summary, the memory placement of a task whose cpuset is changed is
611updated by the kernel, on the next allocation of a page for that task,
612but the processor placement is not updated, until that tasks pid is
613rewritten to the 'tasks' file of its cpuset. This is done to avoid
614impacting the scheduler code in the kernel with a check for changes
615in a tasks processor placement.
616
Paul Jackson45b07ef2006-01-08 01:00:56 -0800617Normally, once a page is allocated (given a physical page
618of main memory) then that page stays on whatever node it
619was allocated, so long as it remains allocated, even if the
620cpusets memory placement policy 'mems' subsequently changes.
621If the cpuset flag file 'memory_migrate' is set true, then when
622tasks are attached to that cpuset, any pages that task had
623allocated to it on nodes in its previous cpuset are migrated
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800624to the tasks new cpuset. The relative placement of the page within
625the cpuset is preserved during these migration operations if possible.
626For example if the page was on the second valid node of the prior cpuset
627then the page will be placed on the second valid node of the new cpuset.
628
Paul Jackson45b07ef2006-01-08 01:00:56 -0800629Also if 'memory_migrate' is set true, then if that cpusets
630'mems' file is modified, pages allocated to tasks in that
631cpuset, that were on nodes in the previous setting of 'mems',
Christoph Lameterb4fb3762006-03-14 19:50:20 -0800632will be moved to nodes in the new setting of 'mems.'
633Pages that were not in the tasks prior cpuset, or in the cpusets
634prior 'mems' setting, will not be moved.
Paul Jackson45b07ef2006-01-08 01:00:56 -0800635
Tobias Klauserd533f672005-09-10 00:26:46 -0700636There is an exception to the above. If hotplug functionality is used
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637to remove all the CPUs that are currently assigned to a cpuset,
Li Zefan02499432008-09-13 02:33:09 -0700638then all the tasks in that cpuset will be moved to the nearest ancestor
639with non-empty cpus. But the moving of some (or all) tasks might fail if
640cpuset is bound with another cgroup subsystem which has some restrictions
641on task attaching. In this failing case, those tasks will stay
642in the original cpuset, and the kernel will automatically update
643their cpus_allowed to allow all online CPUs. When memory hotplug
644functionality for removing Memory Nodes is available, a similar exception
645is expected to apply there as well. In general, the kernel prefers to
646violate cpuset placement, over starving a task that has had all
647its allowed CPUs or Memory Nodes taken offline.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648
649There is a second exception to the above. GFP_ATOMIC requests are
650kernel internal allocations that must be satisfied, immediately.
651The kernel may drop some request, in rare cases even panic, if a
652GFP_ATOMIC alloc fails. If the request cannot be satisfied within
653the current tasks cpuset, then we relax the cpuset, and look for
654memory anywhere we can find it. It's better to violate the cpuset
655than stress the kernel.
656
657To start a new job that is to be contained within a cpuset, the steps are:
658
659 1) mkdir /dev/cpuset
Paul Menage8793d852007-10-18 23:39:39 -0700660 2) mount -t cgroup -ocpuset cpuset /dev/cpuset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 3) Create the new cpuset by doing mkdir's and write's (or echo's) in
662 the /dev/cpuset virtual file system.
663 4) Start a task that will be the "founding father" of the new job.
664 5) Attach that task to the new cpuset by writing its pid to the
665 /dev/cpuset tasks file for that cpuset.
666 6) fork, exec or clone the job tasks from this founding father task.
667
668For example, the following sequence of commands will setup a cpuset
669named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,
670and then start a subshell 'sh' in that cpuset:
671
Paul Menage8793d852007-10-18 23:39:39 -0700672 mount -t cgroup -ocpuset cpuset /dev/cpuset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 cd /dev/cpuset
674 mkdir Charlie
675 cd Charlie
676 /bin/echo 2-3 > cpus
677 /bin/echo 1 > mems
678 /bin/echo $$ > tasks
679 sh
680 # The subshell 'sh' is now running in cpuset Charlie
681 # The next line should display '/Charlie'
682 cat /proc/self/cpuset
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684In the future, a C library interface to cpusets will likely be
685available. For now, the only way to query or modify cpusets is
686via the cpuset file system, using the various cd, mkdir, echo, cat,
687rmdir commands from the shell, or their equivalent from C.
688
689The sched_setaffinity calls can also be done at the shell prompt using
690SGI's runon or Robert Love's taskset. The mbind and set_mempolicy
691calls can be done at the shell prompt using the numactl command
692(part of Andi Kleen's numa package).
693
6942. Usage Examples and Syntax
695============================
696
6972.1 Basic Usage
698---------------
699
700Creating, modifying, using the cpusets can be done through the cpuset
701virtual filesystem.
702
703To mount it, type:
Paul Menage8793d852007-10-18 23:39:39 -0700704# mount -t cgroup -o cpuset cpuset /dev/cpuset
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
706Then under /dev/cpuset you can find a tree that corresponds to the
707tree of the cpusets in the system. For instance, /dev/cpuset
708is the cpuset that holds the whole system.
709
710If you want to create a new cpuset under /dev/cpuset:
711# cd /dev/cpuset
712# mkdir my_cpuset
713
714Now you want to do something with this cpuset.
715# cd my_cpuset
716
717In this directory you can find several files:
718# ls
Miao Xie6a7d68e2008-06-05 22:45:54 -0700719cpu_exclusive memory_migrate mems tasks
720cpus memory_pressure notify_on_release
721mem_exclusive memory_spread_page sched_load_balance
722mem_hardwall memory_spread_slab sched_relax_domain_level
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
724Reading them will give you information about the state of this cpuset:
725the CPUs and Memory Nodes it can use, the processes that are using
726it, its properties. By writing to these files you can manipulate
727the cpuset.
728
729Set some flags:
730# /bin/echo 1 > cpu_exclusive
731
732Add some cpus:
733# /bin/echo 0-7 > cpus
734
Simon Horman2400ff72007-04-01 23:49:40 -0700735Add some mems:
736# /bin/echo 0-7 > mems
737
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738Now attach your shell to this cpuset:
739# /bin/echo $$ > tasks
740
741You can also create cpusets inside your cpuset by using mkdir in this
742directory.
743# mkdir my_sub_cs
744
745To remove a cpuset, just use rmdir:
746# rmdir my_sub_cs
747This will fail if the cpuset is in use (has cpusets inside, or has
748processes attached).
749
Paul Menage8793d852007-10-18 23:39:39 -0700750Note that for legacy reasons, the "cpuset" filesystem exists as a
751wrapper around the cgroup filesystem.
752
753The command
754
755mount -t cpuset X /dev/cpuset
756
757is equivalent to
758
759mount -t cgroup -ocpuset X /dev/cpuset
760echo "/sbin/cpuset_release_agent" > /dev/cpuset/release_agent
761
Linus Torvalds1da177e2005-04-16 15:20:36 -07007622.2 Adding/removing cpus
763------------------------
764
765This is the syntax to use when writing in the cpus or mems files
766in cpuset directories:
767
768# /bin/echo 1-4 > cpus -> set cpus list to cpus 1,2,3,4
769# /bin/echo 1,2,3,4 > cpus -> set cpus list to cpus 1,2,3,4
770
7712.3 Setting flags
772-----------------
773
774The syntax is very simple:
775
776# /bin/echo 1 > cpu_exclusive -> set flag 'cpu_exclusive'
777# /bin/echo 0 > cpu_exclusive -> unset flag 'cpu_exclusive'
778
7792.4 Attaching processes
780-----------------------
781
782# /bin/echo PID > tasks
783
784Note that it is PID, not PIDs. You can only attach ONE task at a time.
785If you have several tasks to attach, you have to do it one after another:
786
787# /bin/echo PID1 > tasks
788# /bin/echo PID2 > tasks
789 ...
790# /bin/echo PIDn > tasks
791
792
7933. Questions
794============
795
796Q: what's up with this '/bin/echo' ?
797A: bash's builtin 'echo' command does not check calls to write() against
798 errors. If you use it in the cpuset file system, you won't be
799 able to tell whether a command succeeded or failed.
800
801Q: When I attach processes, only the first of the line gets really attached !
802A: We can only return one error code per call to write(). So you should also
803 put only ONE pid.
804
8054. Contact
806==========
807
808Web: http://www.bullopensource.org/cpuset