885 lines · plain
1.. _cpusets:2 3=======4CPUSETS5=======6 7Copyright (C) 2004 BULL SA.8 9Written by Simon.Derr@bull.net10 11- Portions Copyright (c) 2004-2006 Silicon Graphics, Inc.12- Modified by Paul Jackson <pj@sgi.com>13- Modified by Christoph Lameter <cl@linux.com>14- Modified by Paul Menage <menage@google.com>15- Modified by Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>16 17.. CONTENTS:18 19 1. Cpusets20 1.1 What are cpusets ?21 1.2 Why are cpusets needed ?22 1.3 How are cpusets implemented ?23 1.4 What are exclusive cpusets ?24 1.5 What is memory_pressure ?25 1.6 What is memory spread ?26 1.7 What is sched_load_balance ?27 1.8 What is sched_relax_domain_level ?28 1.9 How do I use cpusets ?29 2. Usage Examples and Syntax30 2.1 Basic Usage31 2.2 Adding/removing cpus32 2.3 Setting flags33 2.4 Attaching processes34 3. Questions35 4. Contact36 371. Cpusets38==========39 401.1 What are cpusets ?41----------------------42 43Cpusets provide a mechanism for assigning a set of CPUs and Memory44Nodes to a set of tasks. In this document "Memory Node" refers to45an on-line node that contains memory.46 47Cpusets constrain the CPU and Memory placement of tasks to only48the resources within a task's current cpuset. They form a nested49hierarchy visible in a virtual file system. These are the essential50hooks, beyond what is already present, required to manage dynamic51job placement on large systems.52 53Cpusets use the generic cgroup subsystem described in54Documentation/admin-guide/cgroup-v1/cgroups.rst.55 56Requests by a task, using the sched_setaffinity(2) system call to57include CPUs in its CPU affinity mask, and using the mbind(2) and58set_mempolicy(2) system calls to include Memory Nodes in its memory59policy, are both filtered through that task's cpuset, filtering out any60CPUs or Memory Nodes not in that cpuset. The scheduler will not61schedule a task on a CPU that is not allowed in its cpus_allowed62vector, and the kernel page allocator will not allocate a page on a63node that is not allowed in the requesting task's mems_allowed vector.64 65User level code may create and destroy cpusets by name in the cgroup66virtual file system, manage the attributes and permissions of these67cpusets and which CPUs and Memory Nodes are assigned to each cpuset,68specify and query to which cpuset a task is assigned, and list the69task pids assigned to a cpuset.70 71 721.2 Why are cpusets needed ?73----------------------------74 75The management of large computer systems, with many processors (CPUs),76complex memory cache hierarchies and multiple Memory Nodes having77non-uniform access times (NUMA) presents additional challenges for78the efficient scheduling and memory placement of processes.79 80Frequently more modest sized systems can be operated with adequate81efficiency just by letting the operating system automatically share82the available CPU and Memory resources amongst the requesting tasks.83 84But larger systems, which benefit more from careful processor and85memory placement to reduce memory access times and contention,86and which typically represent a larger investment for the customer,87can benefit from explicitly placing jobs on properly sized subsets of88the system.89 90This can be especially valuable on:91 92 * Web Servers running multiple instances of the same web application,93 * Servers running different applications (for instance, a web server94 and a database), or95 * NUMA systems running large HPC applications with demanding96 performance characteristics.97 98These subsets, or "soft partitions" must be able to be dynamically99adjusted, as the job mix changes, without impacting other concurrently100executing jobs. The location of the running jobs pages may also be moved101when the memory locations are changed.102 103The kernel cpuset patch provides the minimum essential kernel104mechanisms required to efficiently implement such subsets. It105leverages existing CPU and Memory Placement facilities in the Linux106kernel to avoid any additional impact on the critical scheduler or107memory allocator code.108 109 1101.3 How are cpusets implemented ?111---------------------------------112 113Cpusets provide a Linux kernel mechanism to constrain which CPUs and114Memory Nodes are used by a process or set of processes.115 116The Linux kernel already has a pair of mechanisms to specify on which117CPUs a task may be scheduled (sched_setaffinity) and on which Memory118Nodes it may obtain memory (mbind, set_mempolicy).119 120Cpusets extends these two mechanisms as follows:121 122 - Cpusets are sets of allowed CPUs and Memory Nodes, known to the123 kernel.124 - Each task in the system is attached to a cpuset, via a pointer125 in the task structure to a reference counted cgroup structure.126 - Calls to sched_setaffinity are filtered to just those CPUs127 allowed in that task's cpuset.128 - Calls to mbind and set_mempolicy are filtered to just129 those Memory Nodes allowed in that task's cpuset.130 - The root cpuset contains all the systems CPUs and Memory131 Nodes.132 - For any cpuset, one can define child cpusets containing a subset133 of the parents CPU and Memory Node resources.134 - The hierarchy of cpusets can be mounted at /dev/cpuset, for135 browsing and manipulation from user space.136 - A cpuset may be marked exclusive, which ensures that no other137 cpuset (except direct ancestors and descendants) may contain138 any overlapping CPUs or Memory Nodes.139 - You can list all the tasks (by pid) attached to any cpuset.140 141The implementation of cpusets requires a few, simple hooks142into the rest of the kernel, none in performance critical paths:143 144 - in init/main.c, to initialize the root cpuset at system boot.145 - in fork and exit, to attach and detach a task from its cpuset.146 - in sched_setaffinity, to mask the requested CPUs by what's147 allowed in that task's cpuset.148 - in sched.c migrate_live_tasks(), to keep migrating tasks within149 the CPUs allowed by their cpuset, if possible.150 - in the mbind and set_mempolicy system calls, to mask the requested151 Memory Nodes by what's allowed in that task's cpuset.152 - in page_alloc.c, to restrict memory to allowed nodes.153 - in vmscan.c, to restrict page recovery to the current cpuset.154 155You should mount the "cgroup" filesystem type in order to enable156browsing and modifying the cpusets presently known to the kernel. No157new system calls are added for cpusets - all support for querying and158modifying cpusets is via this cpuset file system.159 160The /proc/<pid>/status file for each task has four added lines,161displaying the task's cpus_allowed (on which CPUs it may be scheduled)162and mems_allowed (on which Memory Nodes it may obtain memory),163in the two formats seen in the following example::164 165 Cpus_allowed: ffffffff,ffffffff,ffffffff,ffffffff166 Cpus_allowed_list: 0-127167 Mems_allowed: ffffffff,ffffffff168 Mems_allowed_list: 0-63169 170Each cpuset is represented by a directory in the cgroup file system171containing (on top of the standard cgroup files) the following172files describing that cpuset:173 174 - cpuset.cpus: list of CPUs in that cpuset175 - cpuset.mems: list of Memory Nodes in that cpuset176 - cpuset.memory_migrate flag: if set, move pages to cpusets nodes177 - cpuset.cpu_exclusive flag: is cpu placement exclusive?178 - cpuset.mem_exclusive flag: is memory placement exclusive?179 - cpuset.mem_hardwall flag: is memory allocation hardwalled180 - cpuset.memory_pressure: measure of how much paging pressure in cpuset181 - cpuset.memory_spread_page flag: if set, spread page cache evenly on allowed nodes182 - cpuset.memory_spread_slab flag: OBSOLETE. Doesn't have any function.183 - cpuset.sched_load_balance flag: if set, load balance within CPUs on that cpuset184 - cpuset.sched_relax_domain_level: the searching range when migrating tasks185 186In addition, only the root cpuset has the following file:187 188 - cpuset.memory_pressure_enabled flag: compute memory_pressure?189 190New cpusets are created using the mkdir system call or shell191command. The properties of a cpuset, such as its flags, allowed192CPUs and Memory Nodes, and attached tasks, are modified by writing193to the appropriate file in that cpusets directory, as listed above.194 195The named hierarchical structure of nested cpusets allows partitioning196a large system into nested, dynamically changeable, "soft-partitions".197 198The attachment of each task, automatically inherited at fork by any199children of that task, to a cpuset allows organizing the work load200on a system into related sets of tasks such that each set is constrained201to using the CPUs and Memory Nodes of a particular cpuset. A task202may be re-attached to any other cpuset, if allowed by the permissions203on the necessary cpuset file system directories.204 205Such management of a system "in the large" integrates smoothly with206the detailed placement done on individual tasks and memory regions207using the sched_setaffinity, mbind and set_mempolicy system calls.208 209The following rules apply to each cpuset:210 211 - Its CPUs and Memory Nodes must be a subset of its parents.212 - It can't be marked exclusive unless its parent is.213 - If its cpu or memory is exclusive, they may not overlap any sibling.214 215These rules, and the natural hierarchy of cpusets, enable efficient216enforcement of the exclusive guarantee, without having to scan all217cpusets every time any of them change to ensure nothing overlaps a218exclusive cpuset. Also, the use of a Linux virtual file system (vfs)219to represent the cpuset hierarchy provides for a familiar permission220and name space for cpusets, with a minimum of additional kernel code.221 222The cpus and mems files in the root (top_cpuset) cpuset are223read-only. The cpus file automatically tracks the value of224cpu_online_mask using a CPU hotplug notifier, and the mems file225automatically tracks the value of node_states[N_MEMORY]--i.e.,226nodes with memory--using the cpuset_track_online_nodes() hook.227 228The cpuset.effective_cpus and cpuset.effective_mems files are229normally read-only copies of cpuset.cpus and cpuset.mems files230respectively. If the cpuset cgroup filesystem is mounted with the231special "cpuset_v2_mode" option, the behavior of these files will become232similar to the corresponding files in cpuset v2. In other words, hotplug233events will not change cpuset.cpus and cpuset.mems. Those events will234only affect cpuset.effective_cpus and cpuset.effective_mems which show235the actual cpus and memory nodes that are currently used by this cpuset.236See Documentation/admin-guide/cgroup-v2.rst for more information about237cpuset v2 behavior.238 239 2401.4 What are exclusive cpusets ?241--------------------------------242 243If a cpuset is cpu or mem exclusive, no other cpuset, other than244a direct ancestor or descendant, may share any of the same CPUs or245Memory Nodes.246 247A cpuset that is cpuset.mem_exclusive *or* cpuset.mem_hardwall is "hardwalled",248i.e. it restricts kernel allocations for page, buffer and other data249commonly shared by the kernel across multiple users. All cpusets,250whether hardwalled or not, restrict allocations of memory for user251space. This enables configuring a system so that several independent252jobs can share common kernel data, such as file system pages, while253isolating each job's user allocation in its own cpuset. To do this,254construct a large mem_exclusive cpuset to hold all the jobs, and255construct child, non-mem_exclusive cpusets for each individual job.256Only a small amount of typical kernel memory, such as requests from257interrupt handlers, is allowed to be taken outside even a258mem_exclusive cpuset.259 260 2611.5 What is memory_pressure ?262-----------------------------263The memory_pressure of a cpuset provides a simple per-cpuset metric264of the rate that the tasks in a cpuset are attempting to free up in265use memory on the nodes of the cpuset to satisfy additional memory266requests.267 268This enables batch managers monitoring jobs running in dedicated269cpusets to efficiently detect what level of memory pressure that job270is causing.271 272This is useful both on tightly managed systems running a wide mix of273submitted jobs, which may choose to terminate or re-prioritize jobs that274are trying to use more memory than allowed on the nodes assigned to them,275and with tightly coupled, long running, massively parallel scientific276computing jobs that will dramatically fail to meet required performance277goals if they start to use more memory than allowed to them.278 279This mechanism provides a very economical way for the batch manager280to monitor a cpuset for signs of memory pressure. It's up to the281batch manager or other user code to decide what to do about it and282take action.283 284==>285 Unless this feature is enabled by writing "1" to the special file286 /dev/cpuset/memory_pressure_enabled, the hook in the rebalance287 code of __alloc_pages() for this metric reduces to simply noticing288 that the cpuset_memory_pressure_enabled flag is zero. So only289 systems that enable this feature will compute the metric.290 291Why a per-cpuset, running average:292 293 Because this meter is per-cpuset, rather than per-task or mm,294 the system load imposed by a batch scheduler monitoring this295 metric is sharply reduced on large systems, because a scan of296 the tasklist can be avoided on each set of queries.297 298 Because this meter is a running average, instead of an accumulating299 counter, a batch scheduler can detect memory pressure with a300 single read, instead of having to read and accumulate results301 for a period of time.302 303 Because this meter is per-cpuset rather than per-task or mm,304 the batch scheduler can obtain the key information, memory305 pressure in a cpuset, with a single read, rather than having to306 query and accumulate results over all the (dynamically changing)307 set of tasks in the cpuset.308 309A per-cpuset simple digital filter (requires a spinlock and 3 words310of data per-cpuset) is kept, and updated by any task attached to that311cpuset, if it enters the synchronous (direct) page reclaim code.312 313A per-cpuset file provides an integer number representing the recent314(half-life of 10 seconds) rate of direct page reclaims caused by315the tasks in the cpuset, in units of reclaims attempted per second,316times 1000.317 318 3191.6 What is memory spread ?320---------------------------321There are two boolean flag files per cpuset that control where the322kernel allocates pages for the file system buffers and related in323kernel data structures. They are called 'cpuset.memory_spread_page' and324'cpuset.memory_spread_slab'.325 326If the per-cpuset boolean flag file 'cpuset.memory_spread_page' is set, then327the kernel will spread the file system buffers (page cache) evenly328over all the nodes that the faulting task is allowed to use, instead329of preferring to put those pages on the node where the task is running.330 331If the per-cpuset boolean flag file 'cpuset.memory_spread_slab' is set,332then the kernel will spread some file system related slab caches,333such as for inodes and dentries evenly over all the nodes that the334faulting task is allowed to use, instead of preferring to put those335pages on the node where the task is running.336 337The setting of these flags does not affect anonymous data segment or338stack segment pages of a task.339 340By default, both kinds of memory spreading are off, and memory341pages are allocated on the node local to where the task is running,342except perhaps as modified by the task's NUMA mempolicy or cpuset343configuration, so long as sufficient free memory pages are available.344 345When new cpusets are created, they inherit the memory spread settings346of their parent.347 348Setting memory spreading causes allocations for the affected page349or slab caches to ignore the task's NUMA mempolicy and be spread350instead. Tasks using mbind() or set_mempolicy() calls to set NUMA351mempolicies will not notice any change in these calls as a result of352their containing task's memory spread settings. If memory spreading353is turned off, then the currently specified NUMA mempolicy once again354applies to memory page allocations.355 356Both 'cpuset.memory_spread_page' and 'cpuset.memory_spread_slab' are boolean flag357files. By default they contain "0", meaning that the feature is off358for that cpuset. If a "1" is written to that file, then that turns359the named feature on.360 361The implementation is simple.362 363Setting the flag 'cpuset.memory_spread_page' turns on a per-process flag364PFA_SPREAD_PAGE for each task that is in that cpuset or subsequently365joins that cpuset. The page allocation calls for the page cache366is modified to perform an inline check for this PFA_SPREAD_PAGE task367flag, and if set, a call to a new routine cpuset_mem_spread_node()368returns the node to prefer for the allocation.369 370Similarly, setting 'cpuset.memory_spread_slab' turns on the flag371PFA_SPREAD_SLAB, and appropriately marked slab caches will allocate372pages from the node returned by cpuset_mem_spread_node().373 374The cpuset_mem_spread_node() routine is also simple. It uses the375value of a per-task rotor cpuset_mem_spread_rotor to select the next376node in the current task's mems_allowed to prefer for the allocation.377 378This memory placement policy is also known (in other contexts) as379round-robin or interleave.380 381This policy can provide substantial improvements for jobs that need382to place thread local data on the corresponding node, but that need383to access large file system data sets that need to be spread across384the several nodes in the jobs cpuset in order to fit. Without this385policy, especially for jobs that might have one thread reading in the386data set, the memory allocation across the nodes in the jobs cpuset387can become very uneven.388 3891.7 What is sched_load_balance ?390--------------------------------391 392The kernel scheduler (kernel/sched/core.c) automatically load balances393tasks. If one CPU is underutilized, kernel code running on that394CPU will look for tasks on other more overloaded CPUs and move those395tasks to itself, within the constraints of such placement mechanisms396as cpusets and sched_setaffinity.397 398The algorithmic cost of load balancing and its impact on key shared399kernel data structures such as the task list increases more than400linearly with the number of CPUs being balanced. So the scheduler401has support to partition the systems CPUs into a number of sched402domains such that it only load balances within each sched domain.403Each sched domain covers some subset of the CPUs in the system;404no two sched domains overlap; some CPUs might not be in any sched405domain and hence won't be load balanced.406 407Put simply, it costs less to balance between two smaller sched domains408than one big one, but doing so means that overloads in one of the409two domains won't be load balanced to the other one.410 411By default, there is one sched domain covering all CPUs, including those412marked isolated using the kernel boot time "isolcpus=" argument. However,413the isolated CPUs will not participate in load balancing, and will not414have tasks running on them unless explicitly assigned.415 416This default load balancing across all CPUs is not well suited for417the following two situations:418 419 1) On large systems, load balancing across many CPUs is expensive.420 If the system is managed using cpusets to place independent jobs421 on separate sets of CPUs, full load balancing is unnecessary.422 2) Systems supporting realtime on some CPUs need to minimize423 system overhead on those CPUs, including avoiding task load424 balancing if that is not needed.425 426When the per-cpuset flag "cpuset.sched_load_balance" is enabled (the default427setting), it requests that all the CPUs in that cpusets allowed 'cpuset.cpus'428be contained in a single sched domain, ensuring that load balancing429can move a task (not otherwised pinned, as by sched_setaffinity)430from any CPU in that cpuset to any other.431 432When the per-cpuset flag "cpuset.sched_load_balance" is disabled, then the433scheduler will avoid load balancing across the CPUs in that cpuset,434--except-- in so far as is necessary because some overlapping cpuset435has "sched_load_balance" enabled.436 437So, for example, if the top cpuset has the flag "cpuset.sched_load_balance"438enabled, then the scheduler will have one sched domain covering all439CPUs, and the setting of the "cpuset.sched_load_balance" flag in any other440cpusets won't matter, as we're already fully load balancing.441 442Therefore in the above two situations, the top cpuset flag443"cpuset.sched_load_balance" should be disabled, and only some of the smaller,444child cpusets have this flag enabled.445 446When doing this, you don't usually want to leave any unpinned tasks in447the top cpuset that might use non-trivial amounts of CPU, as such tasks448may be artificially constrained to some subset of CPUs, depending on449the particulars of this flag setting in descendant cpusets. Even if450such a task could use spare CPU cycles in some other CPUs, the kernel451scheduler might not consider the possibility of load balancing that452task to that underused CPU.453 454Of course, tasks pinned to a particular CPU can be left in a cpuset455that disables "cpuset.sched_load_balance" as those tasks aren't going anywhere456else anyway.457 458There is an impedance mismatch here, between cpusets and sched domains.459Cpusets are hierarchical and nest. Sched domains are flat; they don't460overlap and each CPU is in at most one sched domain.461 462It is necessary for sched domains to be flat because load balancing463across partially overlapping sets of CPUs would risk unstable dynamics464that would be beyond our understanding. So if each of two partially465overlapping cpusets enables the flag 'cpuset.sched_load_balance', then we466form a single sched domain that is a superset of both. We won't move467a task to a CPU outside its cpuset, but the scheduler load balancing468code might waste some compute cycles considering that possibility.469 470This mismatch is why there is not a simple one-to-one relation471between which cpusets have the flag "cpuset.sched_load_balance" enabled,472and the sched domain configuration. If a cpuset enables the flag, it473will get balancing across all its CPUs, but if it disables the flag,474it will only be assured of no load balancing if no other overlapping475cpuset enables the flag.476 477If two cpusets have partially overlapping 'cpuset.cpus' allowed, and only478one of them has this flag enabled, then the other may find its479tasks only partially load balanced, just on the overlapping CPUs.480This is just the general case of the top_cpuset example given a few481paragraphs above. In the general case, as in the top cpuset case,482don't leave tasks that might use non-trivial amounts of CPU in483such partially load balanced cpusets, as they may be artificially484constrained to some subset of the CPUs allowed to them, for lack of485load balancing to the other CPUs.486 487CPUs in "cpuset.isolcpus" were excluded from load balancing by the488isolcpus= kernel boot option, and will never be load balanced regardless489of the value of "cpuset.sched_load_balance" in any cpuset.490 4911.7.1 sched_load_balance implementation details.492------------------------------------------------493 494The per-cpuset flag 'cpuset.sched_load_balance' defaults to enabled (contrary495to most cpuset flags.) When enabled for a cpuset, the kernel will496ensure that it can load balance across all the CPUs in that cpuset497(makes sure that all the CPUs in the cpus_allowed of that cpuset are498in the same sched domain.)499 500If two overlapping cpusets both have 'cpuset.sched_load_balance' enabled,501then they will be (must be) both in the same sched domain.502 503If, as is the default, the top cpuset has 'cpuset.sched_load_balance' enabled,504then by the above that means there is a single sched domain covering505the whole system, regardless of any other cpuset settings.506 507The kernel commits to user space that it will avoid load balancing508where it can. It will pick as fine a granularity partition of sched509domains as it can while still providing load balancing for any set510of CPUs allowed to a cpuset having 'cpuset.sched_load_balance' enabled.511 512The internal kernel cpuset to scheduler interface passes from the513cpuset code to the scheduler code a partition of the load balanced514CPUs in the system. This partition is a set of subsets (represented515as an array of struct cpumask) of CPUs, pairwise disjoint, that cover516all the CPUs that must be load balanced.517 518The cpuset code builds a new such partition and passes it to the519scheduler sched domain setup code, to have the sched domains rebuilt520as necessary, whenever:521 522 - 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 flag enabled,524 - or 'cpuset.sched_relax_domain_level' value of a cpuset with non-empty CPUs525 and with this flag enabled changes,526 - or a cpuset with non-empty CPUs and with this flag enabled is removed,527 - or a cpu is offlined/onlined.528 529This partition exactly defines what sched domains the scheduler should530setup - one sched domain for each element (struct cpumask) in the531partition.532 533The scheduler remembers the currently active sched domain partitions.534When the scheduler routine partition_sched_domains() is invoked from535the cpuset code to update these sched domains, it compares the new536partition requested with the current, and updates its sched domains,537removing the old and adding the new, for each change.538 539 5401.8 What is sched_relax_domain_level ?541--------------------------------------542 543In sched domain, the scheduler migrates tasks in 2 ways; periodic load544balance on tick, and at time of some schedule events.545 546When a task is woken up, scheduler try to move the task on idle CPU.547For example, if a task A running on CPU X activates another task B548on the same CPU X, and if CPU Y is X's sibling and performing idle,549then scheduler migrate task B to CPU Y so that task B can start on550CPU Y without waiting task A on CPU X.551 552And if a CPU run out of tasks in its runqueue, the CPU try to pull553extra tasks from other busy CPUs to help them before it is going to554be idle.555 556Of course it takes some searching cost to find movable tasks and/or557idle CPUs, the scheduler might not search all CPUs in the domain558every time. In fact, in some architectures, the searching ranges on559events are limited in the same socket or node where the CPU locates,560while the load balance on tick searches all.561 562For example, assume CPU Z is relatively far from CPU X. Even if CPU Z563is idle while CPU X and the siblings are busy, scheduler can't migrate564woken task B from X to Z since it is out of its searching range.565As the result, task B on CPU X need to wait task A or wait load balance566on the next tick. For some applications in special situation, waiting5671 tick may be too long.568 569The 'cpuset.sched_relax_domain_level' file allows you to request changing570this searching range as you like. This file takes int value which571indicates size of searching range in levels approximately as follows,572otherwise initial value -1 that indicates the cpuset has no request.573 574====== ===========================================================575 -1 no request. use system default or follow request of others.576 0 no search.577 1 search siblings (hyperthreads in a core).578 2 search cores in a package.579 3 search cpus in a node [= system wide on non-NUMA system]580 4 search nodes in a chunk of node [on NUMA system]581 5 search system wide [on NUMA system]582====== ===========================================================583 584Not all levels can be present and values can change depending on the585system architecture and kernel configuration. Check586/sys/kernel/debug/sched/domains/cpu*/domain*/ for system-specific587details.588 589The system default is architecture dependent. The system default590can be changed using the relax_domain_level= boot parameter.591 592This file is per-cpuset and affect the sched domain where the cpuset593belongs to. Therefore if the flag 'cpuset.sched_load_balance' of a cpuset594is disabled, then 'cpuset.sched_relax_domain_level' have no effect since595there is no sched domain belonging the cpuset.596 597If multiple cpusets are overlapping and hence they form a single sched598domain, the largest value among those is used. Be careful, if one599requests 0 and others are -1 then 0 is used.600 601Note that modifying this file will have both good and bad effects,602and whether it is acceptable or not depends on your situation.603Don't modify this file if you are not sure.604 605If your situation is:606 607 - The migration costs between each cpu can be assumed considerably608 small(for you) due to your special application's behavior or609 special hardware support for CPU cache etc.610 - The searching cost doesn't have impact(for you) or you can make611 the searching cost enough small by managing cpuset to compact etc.612 - The latency is required even it sacrifices cache hit rate etc.613 then increasing 'sched_relax_domain_level' would benefit you.614 615 6161.9 How do I use cpusets ?617--------------------------618 619In order to minimize the impact of cpusets on critical kernel620code, such as the scheduler, and due to the fact that the kernel621does not support one task updating the memory placement of another622task directly, the impact on a task of changing its cpuset CPU623or Memory Node placement, or of changing to which cpuset a task624is attached, is subtle.625 626If a cpuset has its Memory Nodes modified, then for each task attached627to that cpuset, the next time that the kernel attempts to allocate628a page of memory for that task, the kernel will notice the change629in the task's cpuset, and update its per-task memory placement to630remain within the new cpusets memory placement. If the task was using631mempolicy MPOL_BIND, and the nodes to which it was bound overlap with632its new cpuset, then the task will continue to use whatever subset633of MPOL_BIND nodes are still allowed in the new cpuset. If the task634was using MPOL_BIND and now none of its MPOL_BIND nodes are allowed635in the new cpuset, then the task will be essentially treated as if it636was MPOL_BIND bound to the new cpuset (even though its NUMA placement,637as queried by get_mempolicy(), doesn't change). If a task is moved638from one cpuset to another, then the kernel will adjust the task's639memory placement, as above, the next time that the kernel attempts640to allocate a page of memory for that task.641 642If a cpuset has its 'cpuset.cpus' modified, then each task in that cpuset643will have its allowed CPU placement changed immediately. Similarly,644if a task's pid is written to another cpuset's 'tasks' file, then its645allowed CPU placement is changed immediately. If such a task had been646bound to some subset of its cpuset using the sched_setaffinity() call,647the task will be allowed to run on any CPU allowed in its new cpuset,648negating the effect of the prior sched_setaffinity() call.649 650In summary, the memory placement of a task whose cpuset is changed is651updated by the kernel, on the next allocation of a page for that task,652and the processor placement is updated immediately.653 654Normally, once a page is allocated (given a physical page655of main memory) then that page stays on whatever node it656was allocated, so long as it remains allocated, even if the657cpusets memory placement policy 'cpuset.mems' subsequently changes.658If the cpuset flag file 'cpuset.memory_migrate' is set true, then when659tasks are attached to that cpuset, any pages that task had660allocated to it on nodes in its previous cpuset are migrated661to the task's new cpuset. The relative placement of the page within662the cpuset is preserved during these migration operations if possible.663For example if the page was on the second valid node of the prior cpuset664then the page will be placed on the second valid node of the new cpuset.665 666Also if 'cpuset.memory_migrate' is set true, then if that cpuset's667'cpuset.mems' file is modified, pages allocated to tasks in that668cpuset, that were on nodes in the previous setting of 'cpuset.mems',669will be moved to nodes in the new setting of 'mems.'670Pages that were not in the task's prior cpuset, or in the cpuset's671prior 'cpuset.mems' setting, will not be moved.672 673There is an exception to the above. If hotplug functionality is used674to remove all the CPUs that are currently assigned to a cpuset,675then all the tasks in that cpuset will be moved to the nearest ancestor676with non-empty cpus. But the moving of some (or all) tasks might fail if677cpuset is bound with another cgroup subsystem which has some restrictions678on task attaching. In this failing case, those tasks will stay679in the original cpuset, and the kernel will automatically update680their cpus_allowed to allow all online CPUs. When memory hotplug681functionality for removing Memory Nodes is available, a similar exception682is expected to apply there as well. In general, the kernel prefers to683violate cpuset placement, over starving a task that has had all684its allowed CPUs or Memory Nodes taken offline.685 686There is a second exception to the above. GFP_ATOMIC requests are687kernel internal allocations that must be satisfied, immediately.688The kernel may drop some request, in rare cases even panic, if a689GFP_ATOMIC alloc fails. If the request cannot be satisfied within690the current task's cpuset, then we relax the cpuset, and look for691memory anywhere we can find it. It's better to violate the cpuset692than stress the kernel.693 694To start a new job that is to be contained within a cpuset, the steps are:695 696 1) mkdir /sys/fs/cgroup/cpuset697 2) mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset698 3) Create the new cpuset by doing mkdir's and write's (or echo's) in699 the /sys/fs/cgroup/cpuset virtual file system.700 4) Start a task that will be the "founding father" of the new job.701 5) Attach that task to the new cpuset by writing its pid to the702 /sys/fs/cgroup/cpuset tasks file for that cpuset.703 6) fork, exec or clone the job tasks from this founding father task.704 705For example, the following sequence of commands will setup a cpuset706named "Charlie", containing just CPUs 2 and 3, and Memory Node 1,707and then start a subshell 'sh' in that cpuset::708 709 mount -t cgroup -ocpuset cpuset /sys/fs/cgroup/cpuset710 cd /sys/fs/cgroup/cpuset711 mkdir Charlie712 cd Charlie713 /bin/echo 2-3 > cpuset.cpus714 /bin/echo 1 > cpuset.mems715 /bin/echo $$ > tasks716 sh717 # The subshell 'sh' is now running in cpuset Charlie718 # The next line should display '/Charlie'719 cat /proc/self/cpuset720 721There are ways to query or modify cpusets:722 723 - via the cpuset file system directly, using the various cd, mkdir, echo,724 cat, rmdir commands from the shell, or their equivalent from C.725 - via the C library libcpuset.726 - via the C library libcgroup.727 (https://github.com/libcgroup/libcgroup/)728 - via the python application cset.729 (http://code.google.com/p/cpuset/)730 731The sched_setaffinity calls can also be done at the shell prompt using732SGI's runon or Robert Love's taskset. The mbind and set_mempolicy733calls can be done at the shell prompt using the numactl command734(part of Andi Kleen's numa package).735 7362. Usage Examples and Syntax737============================738 7392.1 Basic Usage740---------------741 742Creating, modifying, using the cpusets can be done through the cpuset743virtual filesystem.744 745To mount it, type:746# mount -t cgroup -o cpuset cpuset /sys/fs/cgroup/cpuset747 748Then under /sys/fs/cgroup/cpuset you can find a tree that corresponds to the749tree of the cpusets in the system. For instance, /sys/fs/cgroup/cpuset750is the cpuset that holds the whole system.751 752If you want to create a new cpuset under /sys/fs/cgroup/cpuset::753 754 # cd /sys/fs/cgroup/cpuset755 # mkdir my_cpuset756 757Now you want to do something with this cpuset::758 759 # cd my_cpuset760 761In this directory you can find several files::762 763 # ls764 cgroup.clone_children cpuset.memory_pressure765 cgroup.event_control cpuset.memory_spread_page766 cgroup.procs cpuset.memory_spread_slab767 cpuset.cpu_exclusive cpuset.mems768 cpuset.cpus cpuset.sched_load_balance769 cpuset.mem_exclusive cpuset.sched_relax_domain_level770 cpuset.mem_hardwall notify_on_release771 cpuset.memory_migrate tasks772 773Reading them will give you information about the state of this cpuset:774the CPUs and Memory Nodes it can use, the processes that are using775it, its properties. By writing to these files you can manipulate776the cpuset.777 778Set some flags::779 780 # /bin/echo 1 > cpuset.cpu_exclusive781 782Add some cpus::783 784 # /bin/echo 0-7 > cpuset.cpus785 786Add some mems::787 788 # /bin/echo 0-7 > cpuset.mems789 790Now attach your shell to this cpuset::791 792 # /bin/echo $$ > tasks793 794You can also create cpusets inside your cpuset by using mkdir in this795directory::796 797 # mkdir my_sub_cs798 799To remove a cpuset, just use rmdir::800 801 # rmdir my_sub_cs802 803This will fail if the cpuset is in use (has cpusets inside, or has804processes attached).805 806Note that for legacy reasons, the "cpuset" filesystem exists as a807wrapper around the cgroup filesystem.808 809The command::810 811 mount -t cpuset X /sys/fs/cgroup/cpuset812 813is equivalent to::814 815 mount -t cgroup -ocpuset,noprefix X /sys/fs/cgroup/cpuset816 echo "/sbin/cpuset_release_agent" > /sys/fs/cgroup/cpuset/release_agent817 8182.2 Adding/removing cpus819------------------------820 821This is the syntax to use when writing in the cpus or mems files822in cpuset directories::823 824 # /bin/echo 1-4 > cpuset.cpus -> set cpus list to cpus 1,2,3,4825 # /bin/echo 1,2,3,4 > cpuset.cpus -> set cpus list to cpus 1,2,3,4826 827To add a CPU to a cpuset, write the new list of CPUs including the828CPU to be added. To add 6 to the above cpuset::829 830 # /bin/echo 1-4,6 > cpuset.cpus -> set cpus list to cpus 1,2,3,4,6831 832Similarly to remove a CPU from a cpuset, write the new list of CPUs833without the CPU to be removed.834 835To remove all the CPUs::836 837 # /bin/echo "" > cpuset.cpus -> clear cpus list838 8392.3 Setting flags840-----------------841 842The syntax is very simple::843 844 # /bin/echo 1 > cpuset.cpu_exclusive -> set flag 'cpuset.cpu_exclusive'845 # /bin/echo 0 > cpuset.cpu_exclusive -> unset flag 'cpuset.cpu_exclusive'846 8472.4 Attaching processes848-----------------------849 850::851 852 # /bin/echo PID > tasks853 854Note that it is PID, not PIDs. You can only attach ONE task at a time.855If you have several tasks to attach, you have to do it one after another::856 857 # /bin/echo PID1 > tasks858 # /bin/echo PID2 > tasks859 ...860 # /bin/echo PIDn > tasks861 862 8633. Questions864============865 866Q:867 what's up with this '/bin/echo' ?868 869A:870 bash's builtin 'echo' command does not check calls to write() against871 errors. If you use it in the cpuset file system, you won't be872 able to tell whether a command succeeded or failed.873 874Q:875 When I attach processes, only the first of the line gets really attached !876 877A:878 We can only return one error code per call to write(). So you should also879 put only ONE pid.880 8814. Contact882==========883 884Web: http://www.bullopensource.org/cpuset885