€•u“      Œsphinx.addnodes”Œdocument”“”)”}”(Œ	rawsource”Œ ”Œchildren”]”(Œtranslations”ŒLanguagesNode”“”)”}”(hhh]”(h Œpending_xref”“”)”}”(hhh]”Œdocutils.nodes”ŒText”“”ŒChinese (Simplified)”…””}”Œparent”hsbaŒ
attributes”}”(Œids”]”Œclasses”]”Œnames”]”Œdupnames”]”Œbackrefs”]”Œ	refdomain”Œstd”Œreftype”Œdoc”Œ	reftarget”Œ'/translations/zh_CN/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuŒtagname”hhhubh)”}”(hhh]”hŒChinese (Traditional)”…””}”hh2sbah}”(h]”h ]”h"]”h$]”h&]”Œ	refdomain”h)Œreftype”h+Œ	reftarget”Œ'/translations/zh_TW/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuh1hhhubh)”}”(hhh]”hŒItalian”…””}”hhFsbah}”(h]”h ]”h"]”h$]”h&]”Œ	refdomain”h)Œreftype”h+Œ	reftarget”Œ'/translations/it_IT/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuh1hhhubh)”}”(hhh]”hŒJapanese”…””}”hhZsbah}”(h]”h ]”h"]”h$]”h&]”Œ	refdomain”h)Œreftype”h+Œ	reftarget”Œ'/translations/ja_JP/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuh1hhhubh)”}”(hhh]”hŒKorean”…””}”hhnsbah}”(h]”h ]”h"]”h$]”h&]”Œ	refdomain”h)Œreftype”h+Œ	reftarget”Œ'/translations/ko_KR/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuh1hhhubh)”}”(hhh]”hŒSpanish”…””}”hh‚sbah}”(h]”h ]”h"]”h$]”h&]”Œ	refdomain”h)Œreftype”h+Œ	reftarget”Œ'/translations/sp_SP/scheduler/sched-bwc”Œmodname”NŒ	classname”NŒrefexplicit”ˆuh1hhhubeh}”(h]”h ]”h"]”h$]”h&]”Œcurrent_language”ŒEnglish”uh1h
hhŒ	_document”hŒsource”NŒline”NubhŒsection”“”)”}”(hhh]”(hŒtitle”“”)”}”(hŒCFS Bandwidth Control”h]”hŒCFS Bandwidth Control”…””}”(hh¨hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hh£hžhhŸŒA/var/lib/git/docbuild/linux/Documentation/scheduler/sched-bwc.rst”h KubhŒnote”“”)”}”(hŒThis document only discusses CPU bandwidth control for SCHED_NORMAL.
The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst”h]”hŒ	paragraph”“”)”}”(hŒThis document only discusses CPU bandwidth control for SCHED_NORMAL.
The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst”h]”hŒThis document only discusses CPU bandwidth control for SCHED_NORMAL.
The SCHED_RT case is covered in Documentation/scheduler/sched-rt-group.rst”…””}”(hh¿hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khh¹ubah}”(h]”h ]”h"]”h$]”h&]”uh1h·hh£hžhhŸh¶h Nubh¾)”}”(hŒ›CFS bandwidth control is a CONFIG_FAIR_GROUP_SCHED extension which allows the
specification of the maximum CPU bandwidth available to a group or hierarchy.”h]”hŒ›CFS bandwidth control is a CONFIG_FAIR_GROUP_SCHED extension which allows the
specification of the maximum CPU bandwidth available to a group or hierarchy.”…””}”(hhÓhžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K	hh£hžhubh¾)”}”(hXì  The bandwidth allowed for a group is specified using a quota and period. Within
each given "period" (microseconds), a task group is allocated up to "quota"
microseconds of CPU time. That quota is assigned to per-cpu run queues in
slices as threads in the cgroup become runnable. Once all quota has been
assigned any additional requests for quota will result in those threads being
throttled. Throttled threads will not be able to run again until the next
period when the quota is replenished.”h]”hXô  The bandwidth allowed for a group is specified using a quota and period. Within
each given â€œperiodâ€ (microseconds), a task group is allocated up to â€œquotaâ€
microseconds of CPU time. That quota is assigned to per-cpu run queues in
slices as threads in the cgroup become runnable. Once all quota has been
assigned any additional requests for quota will result in those threads being
throttled. Throttled threads will not be able to run again until the next
period when the quota is replenished.”…””}”(hháhžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khh£hžhubh¾)”}”(hX)  A group's unassigned quota is globally tracked, being refreshed back to
cfs_quota units at each period boundary. As threads consume this bandwidth it
is transferred to cpu-local "silos" on a demand basis. The amount transferred
within each of these updates is tunable and described as the "slice".”h]”hX3  A groupâ€™s unassigned quota is globally tracked, being refreshed back to
cfs_quota units at each period boundary. As threads consume this bandwidth it
is transferred to cpu-local â€œsilosâ€ on a demand basis. The amount transferred
within each of these updates is tunable and described as the â€œsliceâ€.”…””}”(hhïhžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khh£hžhubh¢)”}”(hhh]”(h§)”}”(hŒBurst feature”h]”hŒBurst feature”…””}”(hj   hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hhýhžhhŸh¶h Kubh¾)”}”(hŒ”This feature borrows time now against our future underrun, at the cost of
increased interference against the other system users. All nicely bounded.”h]”hŒ”This feature borrows time now against our future underrun, at the cost of
increased interference against the other system users. All nicely bounded.”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khhýhžhubh¾)”}”(hŒ9Traditional (UP-EDF) bandwidth control is something like:”h]”hŒ9Traditional (UP-EDF) bandwidth control is something like:”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KhhýhžhubhŒblock_quote”“”)”}”(hŒ(U = \Sum u_i) <= 1
”h]”h¾)”}”(hŒ(U = \Sum u_i) <= 1”h]”hŒ(U =  Sum u_i) <= 1”…””}”(hj0  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K hj,  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j*  hŸh¶h K hhýhžhubh¾)”}”(hXD  This guaranteeds both that every deadline is met and that the system is
stable. After all, if U were > 1, then for every second of walltime,
we'd have to run more than a second of program time, and obviously miss
our deadline, but the next deadline will be further out still, there is
never time to catch up, unbounded fail.”h]”hXF  This guaranteeds both that every deadline is met and that the system is
stable. After all, if U were > 1, then for every second of walltime,
weâ€™d have to run more than a second of program time, and obviously miss
our deadline, but the next deadline will be further out still, there is
never time to catch up, unbounded fail.”…””}”(hjD  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K"hhýhžhubh¾)”}”(hŒ’The burst feature observes that a workload doesn't always executes the full
quota; this enables one to describe u_i as a statistical distribution.”h]”hŒ”The burst feature observes that a workload doesnâ€™t always executes the full
quota; this enables one to describe u_i as a statistical distribution.”…””}”(hjR  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K(hhýhžhubh¾)”}”(hX‡  For example, have u_i = {x,e}_i, where x is the p(95) and x+e p(100)
(the traditional WCET). This effectively allows u to be smaller,
increasing the efficiency (we can pack more tasks in the system), but at
the cost of missing deadlines when all the odds line up. However, it
does maintain stability, since every overrun must be paired with an
underrun as long as our x is above the average.”h]”hX‡  For example, have u_i = {x,e}_i, where x is the p(95) and x+e p(100)
(the traditional WCET). This effectively allows u to be smaller,
increasing the efficiency (we can pack more tasks in the system), but at
the cost of missing deadlines when all the odds line up. However, it
does maintain stability, since every overrun must be paired with an
underrun as long as our x is above the average.”…””}”(hj`  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K+hhýhžhubh¾)”}”(hX¸  That is, suppose we have 2 tasks, both specify a p(95) value, then we
have a p(95)*p(95) = 90.25% chance both tasks are within their quota and
everything is good. At the same time we have a p(5)p(5) = 0.25% chance
both tasks will exceed their quota at the same time (guaranteed deadline
fail). Somewhere in between there's a threshold where one exceeds and
the other doesn't underrun enough to compensate; this depends on the
specific CDFs.”h]”hX¼  That is, suppose we have 2 tasks, both specify a p(95) value, then we
have a p(95)*p(95) = 90.25% chance both tasks are within their quota and
everything is good. At the same time we have a p(5)p(5) = 0.25% chance
both tasks will exceed their quota at the same time (guaranteed deadline
fail). Somewhere in between thereâ€™s a threshold where one exceeds and
the other doesnâ€™t underrun enough to compensate; this depends on the
specific CDFs.”…””}”(hjn  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K2hhýhžhubh¾)”}”(hŒ§At the same time, we can say that the worst case deadline miss, will be
\Sum e_i; that is, there is a bounded tardiness (under the assumption
that x+e is indeed WCET).”h]”hŒ§At the same time, we can say that the worst case deadline miss, will be
 Sum e_i; that is, there is a bounded tardiness (under the assumption
that x+e is indeed WCET).”…””}”(hj|  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K:hhýhžhubh¾)”}”(hXG  The interference when using burst is valued by the possibilities for
missing the deadline and the average WCET. Test results showed that when
there many cgroups or CPU is under utilized, the interference is
limited. More details are shown in:
https://lore.kernel.org/lkml/5371BD36-55AE-4F71-B9D7-B86DC32E3D2B@linux.alibaba.com/”h]”(hŒóThe interference when using burst is valued by the possibilities for
missing the deadline and the average WCET. Test results showed that when
there many cgroups or CPU is under utilized, the interference is
limited. More details are shown in:
”…””}”(hjŠ  hžhhŸNh NubhŒ	reference”“”)”}”(hŒThttps://lore.kernel.org/lkml/5371BD36-55AE-4F71-B9D7-B86DC32E3D2B@linux.alibaba.com/”h]”hŒThttps://lore.kernel.org/lkml/5371BD36-55AE-4F71-B9D7-B86DC32E3D2B@linux.alibaba.com/”…””}”(hj”  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”Œrefuri”j–  uh1j’  hjŠ  ubeh}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K>hhýhžhubeh}”(h]”Œburst-feature”ah ]”h"]”Œburst feature”ah$]”h&]”uh1h¡hh£hžhhŸh¶h Kubh¢)”}”(hhh]”(h§)”}”(hŒ
Management”h]”hŒ
Management”…””}”(hj´  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hj±  hžhhŸh¶h KEubh¾)”}”(hŒJQuota, period and burst are managed within the cpu subsystem via cgroupfs.”h]”hŒJQuota, period and burst are managed within the cpu subsystem via cgroupfs.”…””}”(hjÂ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KFhj±  hžhubh¸)”}”(hŒ¡The cgroupfs files described in this section are only applicable
to cgroup v1. For cgroup v2, see
:ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`.”h]”h¾)”}”(hŒ¡The cgroupfs files described in this section are only applicable
to cgroup v1. For cgroup v2, see
:ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`.”h]”(hŒbThe cgroupfs files described in this section are only applicable
to cgroup v1. For cgroup v2, see
”…””}”(hjÔ  hžhhŸNh Nubh)”}”(hŒ>:ref:`Documentation/admin-guide/cgroup-v2.rst <cgroup-v2-cpu>`”h]”hŒinline”“”)”}”(hjÞ  h]”hŒ'Documentation/admin-guide/cgroup-v2.rst”…””}”(hjâ  hžhhŸNh Nubah}”(h]”h ]”(Œxref”Œstd”Œstd-ref”eh"]”h$]”h&]”uh1jà  hjÜ  ubah}”(h]”h ]”h"]”h$]”h&]”Œrefdoc”Œscheduler/sched-bwc”Œ	refdomain”jí  Œreftype”Œref”Œrefexplicit”ˆŒrefwarn”ˆŒ	reftarget”Œcgroup-v2-cpu”uh1hhŸh¶h KIhjÔ  ubhŒ.”…””}”(hjÔ  hžhhŸNh Nubeh}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KIhjÐ  ubah}”(h]”h ]”h"]”h$]”h&]”uh1h·hj±  hžhhŸh¶h NubhŒbullet_list”“”)”}”(hhh]”(hŒ	list_item”“”)”}”(hŒHcpu.cfs_quota_us: run-time replenished within a period (in microseconds)”h]”h¾)”}”(hj  h]”hŒHcpu.cfs_quota_us: run-time replenished within a period (in microseconds)”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KMhj  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hŒ;cpu.cfs_period_us: the length of a period (in microseconds)”h]”h¾)”}”(hj1  h]”hŒ;cpu.cfs_period_us: the length of a period (in microseconds)”…””}”(hj3  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KNhj/  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hŒAcpu.stat: exports throttling statistics [explained further below]”h]”h¾)”}”(hjH  h]”hŒAcpu.stat: exports throttling statistics [explained further below]”…””}”(hjJ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KOhjF  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hŒEcpu.cfs_burst_us: the maximum accumulated run-time (in microseconds)
”h]”h¾)”}”(hŒDcpu.cfs_burst_us: the maximum accumulated run-time (in microseconds)”h]”hŒDcpu.cfs_burst_us: the maximum accumulated run-time (in microseconds)”…””}”(hja  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KPhj]  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubeh}”(h]”h ]”h"]”h$]”h&]”Œbullet”Œ-”uh1j  hŸh¶h KMhj±  hžhubh¾)”}”(hŒThe default values are::”h]”hŒThe default values are:”…””}”(hj}  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KRhj±  hžhubhŒliteral_block”“”)”}”(hŒ>cpu.cfs_period_us=100ms
cpu.cfs_quota_us=-1
cpu.cfs_burst_us=0”h]”hŒ>cpu.cfs_period_us=100ms
cpu.cfs_quota_us=-1
cpu.cfs_burst_us=0”…””}”hj  sbah}”(h]”h ]”h"]”h$]”h&]”Œ	xml:space”Œpreserve”uh1j‹  hŸh¶h KThj±  hžhubh¾)”}”(hŒîA value of -1 for cpu.cfs_quota_us indicates that the group does not have any
bandwidth restriction in place, such a group is described as an unconstrained
bandwidth group. This represents the traditional work-conserving behavior for
CFS.”h]”hŒîA value of -1 for cpu.cfs_quota_us indicates that the group does not have any
bandwidth restriction in place, such a group is described as an unconstrained
bandwidth group. This represents the traditional work-conserving behavior for
CFS.”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KXhj±  hžhubh¾)”}”(hXe  Writing any (valid) positive value(s) no smaller than cpu.cfs_burst_us will
enact the specified bandwidth limit. The minimum quota allowed for the quota or
period is 1ms. There is also an upper bound on the period length of 1s.
Additional restrictions exist when bandwidth limits are used in a hierarchical
fashion, these are explained in more detail below.”h]”hXe  Writing any (valid) positive value(s) no smaller than cpu.cfs_burst_us will
enact the specified bandwidth limit. The minimum quota allowed for the quota or
period is 1ms. There is also an upper bound on the period length of 1s.
Additional restrictions exist when bandwidth limits are used in a hierarchical
fashion, these are explained in more detail below.”…””}”(hj«  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K]hj±  hžhubh¾)”}”(hŒˆWriting any negative value to cpu.cfs_quota_us will remove the bandwidth limit
and return the group to an unconstrained state once more.”h]”hŒˆWriting any negative value to cpu.cfs_quota_us will remove the bandwidth limit
and return the group to an unconstrained state once more.”…””}”(hj¹  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kchj±  hžhubh¾)”}”(hX;  A value of 0 for cpu.cfs_burst_us indicates that the group can not accumulate
any unused bandwidth. It makes the traditional bandwidth control behavior for
CFS unchanged. Writing any (valid) positive value(s) no larger than
cpu.cfs_quota_us into cpu.cfs_burst_us will enact the cap on unused bandwidth
accumulation.”h]”hX;  A value of 0 for cpu.cfs_burst_us indicates that the group can not accumulate
any unused bandwidth. It makes the traditional bandwidth control behavior for
CFS unchanged. Writing any (valid) positive value(s) no larger than
cpu.cfs_quota_us into cpu.cfs_burst_us will enact the cap on unused bandwidth
accumulation.”…””}”(hjÇ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kfhj±  hžhubh¾)”}”(hŒxAny updates to a group's bandwidth specification will result in it becoming
unthrottled if it is in a constrained state.”h]”hŒzAny updates to a groupâ€™s bandwidth specification will result in it becoming
unthrottled if it is in a constrained state.”…””}”(hjÕ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Klhj±  hžhubeh}”(h]”Œ
management”ah ]”h"]”Œ
management”ah$]”h&]”uh1h¡hh£hžhhŸh¶h KEubh¢)”}”(hhh]”(h§)”}”(hŒSystem wide settings”h]”hŒSystem wide settings”…””}”(hjî  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hjë  hžhhŸh¶h Kpubh¾)”}”(hX  For efficiency run-time is transferred between the global pool and CPU local
"silos" in a batch fashion. This greatly reduces global accounting pressure
on large systems. The amount transferred each time such an update is required
is described as the "slice".”h]”hX  For efficiency run-time is transferred between the global pool and CPU local
â€œsilosâ€ in a batch fashion. This greatly reduces global accounting pressure
on large systems. The amount transferred each time such an update is required
is described as the â€œsliceâ€.”…””}”(hjü  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kqhjë  hžhubh¾)”}”(hŒThis is tunable via procfs::”h]”hŒThis is tunable via procfs:”…””}”(hj
  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kvhjë  hžhubjŒ  )”}”(hŒ;/proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)”h]”hŒ;/proc/sys/kernel/sched_cfs_bandwidth_slice_us (default=5ms)”…””}”hj  sbah}”(h]”h ]”h"]”h$]”h&]”j›  jœ  uh1j‹  hŸh¶h Kxhjë  hžhubh¾)”}”(hŒqLarger slice values will reduce transfer overheads, while smaller values allow
for more fine-grained consumption.”h]”hŒqLarger slice values will reduce transfer overheads, while smaller values allow
for more fine-grained consumption.”…””}”(hj&  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kzhjë  hžhubeh}”(h]”Œsystem-wide-settings”ah ]”h"]”Œsystem wide settings”ah$]”h&]”uh1h¡hh£hžhhŸh¶h Kpubh¢)”}”(hhh]”(h§)”}”(hŒ
Statistics”h]”hŒ
Statistics”…””}”(hj?  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hj<  hžhhŸh¶h K~ubh¾)”}”(hŒEA group's bandwidth statistics are exported via 5 fields in cpu.stat.”h]”hŒGA groupâ€™s bandwidth statistics are exported via 5 fields in cpu.stat.”…””}”(hjM  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khj<  hžhubh¾)”}”(hŒ	cpu.stat:”h]”hŒ	cpu.stat:”…””}”(hj[  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khj<  hžhubj  )”}”(hhh]”(j  )”}”(hŒ>nr_periods: Number of enforcement intervals that have elapsed.”h]”h¾)”}”(hjn  h]”hŒ>nr_periods: Number of enforcement intervals that have elapsed.”…””}”(hjp  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kƒhjl  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hji  hžhhŸh¶h Nubj  )”}”(hŒCnr_throttled: Number of times the group has been throttled/limited.”h]”h¾)”}”(hj…  h]”hŒCnr_throttled: Number of times the group has been throttled/limited.”…””}”(hj‡  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K„hjƒ  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hji  hžhhŸh¶h Nubj  )”}”(hŒmthrottled_time: The total time duration (in nanoseconds) for which entities
of the group have been throttled.”h]”h¾)”}”(hŒmthrottled_time: The total time duration (in nanoseconds) for which entities
of the group have been throttled.”h]”hŒmthrottled_time: The total time duration (in nanoseconds) for which entities
of the group have been throttled.”…””}”(hjž  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K…hjš  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hji  hžhhŸh¶h Nubj  )”}”(hŒ*nr_bursts: Number of periods burst occurs.”h]”h¾)”}”(hj´  h]”hŒ*nr_bursts: Number of periods burst occurs.”…””}”(hj¶  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K‡hj²  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hji  hžhhŸh¶h Nubj  )”}”(hŒlburst_time: Cumulative wall-time (in nanoseconds) that any CPUs has used
above quota in respective periods.
”h]”h¾)”}”(hŒkburst_time: Cumulative wall-time (in nanoseconds) that any CPUs has used
above quota in respective periods.”h]”hŒkburst_time: Cumulative wall-time (in nanoseconds) that any CPUs has used
above quota in respective periods.”…””}”(hjÍ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KˆhjÉ  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hji  hžhhŸh¶h Nubeh}”(h]”h ]”h"]”h$]”h&]”j{  j|  uh1j  hŸh¶h Kƒhj<  hžhubh¾)”}”(hŒThis interface is read-only.”h]”hŒThis interface is read-only.”…””}”(hjç  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K‹hj<  hžhubeh}”(h]”Œ
statistics”ah ]”h"]”Œ
statistics”ah$]”h&]”uh1h¡hh£hžhhŸh¶h K~ubh¢)”}”(hhh]”(h§)”}”(hŒHierarchical considerations”h]”hŒHierarchical considerations”…””}”(hj   hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hjý  hžhhŸh¶h KŽubh¾)”}”(hŒéThe interface enforces that an individual entity's bandwidth is always
attainable, that is: max(c_i) <= C. However, over-subscription in the
aggregate case is explicitly allowed to enable work-conserving semantics
within a hierarchy:”h]”hŒëThe interface enforces that an individual entityâ€™s bandwidth is always
attainable, that is: max(c_i) <= C. However, over-subscription in the
aggregate case is explicitly allowed to enable work-conserving semantics
within a hierarchy:”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Khjý  hžhubj+  )”}”(hŒe.g. \Sum (c_i) may exceed C
”h]”h¾)”}”(hŒe.g. \Sum (c_i) may exceed C”h]”hŒe.g.  Sum (c_i) may exceed C”…””}”(hj   hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K”hj  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j*  hŸh¶h K”hjý  hžhubh¾)”}”(hŒ;[ Where C is the parent's bandwidth, and c_i its children ]”h]”hŒ=[ Where C is the parentâ€™s bandwidth, and c_i its children ]”…””}”(hj4  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K–hjý  hžhubh¾)”}”(hŒ9There are two ways in which a group may become throttled:”h]”hŒ9There are two ways in which a group may become throttled:”…””}”(hjB  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K™hjý  hžhubj+  )”}”(hŒka. it fully consumes its own quota within a period
b. a parent's quota is fully consumed within its period
”h]”hŒenumerated_list”“”)”}”(hhh]”(j  )”}”(hŒ/it fully consumes its own quota within a period”h]”h¾)”}”(hj[  h]”hŒ/it fully consumes its own quota within a period”…””}”(hj]  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K›hjY  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hjV  ubj  )”}”(hŒ5a parent's quota is fully consumed within its period
”h]”h¾)”}”(hŒ4a parent's quota is fully consumed within its period”h]”hŒ6a parentâ€™s quota is fully consumed within its period”…””}”(hjt  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kœhjp  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j  hjV  ubeh}”(h]”h ]”h"]”h$]”h&]”Œenumtype”Œ
loweralpha”Œprefix”hŒsuffix”Œ.”uh1jT  hjP  ubah}”(h]”h ]”h"]”h$]”h&]”uh1j*  hŸh¶h K›hjý  hžhubh¾)”}”(hŒ…In case b) above, even though the child may have runtime remaining it will not
be allowed to until the parent's runtime is refreshed.”h]”hŒ‡In case b) above, even though the child may have runtime remaining it will not
be allowed to until the parentâ€™s runtime is refreshed.”…””}”(hj™  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kžhjý  hžhubeh}”(h]”Œhierarchical-considerations”ah ]”h"]”Œhierarchical considerations”ah$]”h&]”uh1h¡hh£hžhhŸh¶h KŽubh¢)”}”(hhh]”(h§)”}”(hŒCFS Bandwidth Quota Caveats”h]”hŒCFS Bandwidth Quota Caveats”…””}”(hj²  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hj¯  hžhhŸh¶h K¢ubh¾)”}”(hXC  Once a slice is assigned to a cpu it does not expire.  However all but 1ms of
the slice may be returned to the global pool if all threads on that cpu become
unrunnable. This is configured at compile time by the min_cfs_rq_runtime
variable. This is a performance tweak that helps prevent added contention on
the global lock.”h]”hXC  Once a slice is assigned to a cpu it does not expire.  However all but 1ms of
the slice may be returned to the global pool if all threads on that cpu become
unrunnable. This is configured at compile time by the min_cfs_rq_runtime
variable. This is a performance tweak that helps prevent added contention on
the global lock.”…””}”(hjÀ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K£hj¯  hžhubh¾)”}”(hŒpThe fact that cpu-local slices do not expire results in some interesting corner
cases that should be understood.”h]”hŒpThe fact that cpu-local slices do not expire results in some interesting corner
cases that should be understood.”…””}”(hjÎ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K©hj¯  hžhubh¾)”}”(hXv  For cgroup cpu constrained applications that are cpu limited this is a
relatively moot point because they will naturally consume the entirety of their
quota as well as the entirety of each cpu-local slice in each period. As a
result it is expected that nr_periods roughly equal nr_throttled, and that
cpuacct.usage will increase roughly equal to cfs_quota_us in each period.”h]”hXv  For cgroup cpu constrained applications that are cpu limited this is a
relatively moot point because they will naturally consume the entirety of their
quota as well as the entirety of each cpu-local slice in each period. As a
result it is expected that nr_periods roughly equal nr_throttled, and that
cpuacct.usage will increase roughly equal to cfs_quota_us in each period.”…””}”(hjÜ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K¬hj¯  hžhubh¾)”}”(hX—  For highly-threaded, non-cpu bound applications this non-expiration nuance
allows applications to briefly burst past their quota limits by the amount of
unused slice on each cpu that the task group is running on (typically at most
1ms per cpu or as defined by min_cfs_rq_runtime).  This slight burst only
applies if quota had been assigned to a cpu and then not fully used or returned
in previous periods. This burst amount will not be transferred between cores.
As a result, this mechanism still strictly limits the task group to quota
average usage, albeit over a longer time window than a single period.  This
also limits the burst ability to no more than 1ms per cpu.  This provides
better more predictable user experience for highly threaded applications with
small quota limits on high core count machines. It also eliminates the
propensity to throttle these applications while simultaneously using less than
quota amounts of cpu. Another way to say this, is that by allowing the unused
portion of a slice to remain valid across periods we have decreased the
possibility of wastefully expiring quota on cpu-local silos that don't need a
full slice's amount of cpu time.”h]”hX›  For highly-threaded, non-cpu bound applications this non-expiration nuance
allows applications to briefly burst past their quota limits by the amount of
unused slice on each cpu that the task group is running on (typically at most
1ms per cpu or as defined by min_cfs_rq_runtime).  This slight burst only
applies if quota had been assigned to a cpu and then not fully used or returned
in previous periods. This burst amount will not be transferred between cores.
As a result, this mechanism still strictly limits the task group to quota
average usage, albeit over a longer time window than a single period.  This
also limits the burst ability to no more than 1ms per cpu.  This provides
better more predictable user experience for highly threaded applications with
small quota limits on high core count machines. It also eliminates the
propensity to throttle these applications while simultaneously using less than
quota amounts of cpu. Another way to say this, is that by allowing the unused
portion of a slice to remain valid across periods we have decreased the
possibility of wastefully expiring quota on cpu-local silos that donâ€™t need a
full sliceâ€™s amount of cpu time.”…””}”(hjê  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h K²hj¯  hžhubh¾)”}”(hXí  The interaction between cpu-bound and non-cpu-bound-interactive applications
should also be considered, especially when single core usage hits 100%. If you
gave each of these applications half of a cpu-core and they both got scheduled
on the same CPU it is theoretically possible that the non-cpu bound application
will use up to 1ms additional quota in some periods, thereby preventing the
cpu-bound application from fully using its quota by that same amount. In these
instances it will be up to the CFS algorithm (see sched-design-CFS.rst) to
decide which application is chosen to run, as they will both be runnable and
have remaining quota. This runtime discrepancy will be made up in the following
periods when the interactive application idles.”h]”hXí  The interaction between cpu-bound and non-cpu-bound-interactive applications
should also be considered, especially when single core usage hits 100%. If you
gave each of these applications half of a cpu-core and they both got scheduled
on the same CPU it is theoretically possible that the non-cpu bound application
will use up to 1ms additional quota in some periods, thereby preventing the
cpu-bound application from fully using its quota by that same amount. In these
instances it will be up to the CFS algorithm (see sched-design-CFS.rst) to
decide which application is chosen to run, as they will both be runnable and
have remaining quota. This runtime discrepancy will be made up in the following
periods when the interactive application idles.”…””}”(hjø  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KÃhj¯  hžhubeh}”(h]”Œcfs-bandwidth-quota-caveats”ah ]”h"]”Œcfs bandwidth quota caveats”ah$]”h&]”uh1h¡hh£hžhhŸh¶h K¢ubh¢)”}”(hhh]”(h§)”}”(hŒExamples”h]”hŒExamples”…””}”(hj  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h¦hj  hžhhŸh¶h KÏubjU  )”}”(hhh]”(j  )”}”(hX  Limit a group to 1 CPU worth of runtime::

     If period is 250ms and quota is also 250ms, the group will get
     1 CPU worth of runtime every 250ms.

     # echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
     # echo 250000 > cpu.cfs_period_us /* period = 250ms */
”h]”(h¾)”}”(hŒ)Limit a group to 1 CPU worth of runtime::”h]”hŒ(Limit a group to 1 CPU worth of runtime:”…””}”(hj&  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KÐhj"  ubjŒ  )”}”(hŒÏIf period is 250ms and quota is also 250ms, the group will get
1 CPU worth of runtime every 250ms.

# echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
# echo 250000 > cpu.cfs_period_us /* period = 250ms */”h]”hŒÏIf period is 250ms and quota is also 250ms, the group will get
1 CPU worth of runtime every 250ms.

# echo 250000 > cpu.cfs_quota_us /* quota = 250ms */
# echo 250000 > cpu.cfs_period_us /* period = 250ms */”…””}”hj4  sbah}”(h]”h ]”h"]”h$]”h&]”j›  jœ  uh1j‹  hŸh¶h KÒhj"  ubeh}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hXX  Limit a group to 2 CPUs worth of runtime on a multi-CPU machine

With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
runtime every 500ms::

     # echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
     # echo 500000 > cpu.cfs_period_us /* period = 500ms */

     The larger period here allows for increased burst capacity.
”h]”(h¾)”}”(hŒ?Limit a group to 2 CPUs worth of runtime on a multi-CPU machine”h]”hŒ?Limit a group to 2 CPUs worth of runtime on a multi-CPU machine”…””}”(hjL  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KØhjH  ubh¾)”}”(hŒ[With 500ms period and 1000ms quota, the group can get 2 CPUs worth of
runtime every 500ms::”h]”hŒZWith 500ms period and 1000ms quota, the group can get 2 CPUs worth of
runtime every 500ms:”…””}”(hjZ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h KÚhjH  ubjŒ  )”}”(hŒª# echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
# echo 500000 > cpu.cfs_period_us /* period = 500ms */

The larger period here allows for increased burst capacity.”h]”hŒª# echo 1000000 > cpu.cfs_quota_us /* quota = 1000ms */
# echo 500000 > cpu.cfs_period_us /* period = 500ms */

The larger period here allows for increased burst capacity.”…””}”hjh  sbah}”(h]”h ]”h"]”h$]”h&]”j›  jœ  uh1j‹  hŸh¶h KÝhjH  ubeh}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hXC  Limit a group to 20% of 1 CPU.

With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::

     # echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
     # echo 50000 > cpu.cfs_period_us /* period = 50ms */

By using a small period here we are ensuring a consistent latency
response at the expense of burst capacity.
”h]”(h¾)”}”(hŒLimit a group to 20% of 1 CPU.”h]”hŒLimit a group to 20% of 1 CPU.”…””}”(hj€  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kâhj|  ubh¾)”}”(hŒAWith 50ms period, 10ms quota will be equivalent to 20% of 1 CPU::”h]”hŒ@With 50ms period, 10ms quota will be equivalent to 20% of 1 CPU:”…””}”(hjŽ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kähj|  ubjŒ  )”}”(hŒg# echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
# echo 50000 > cpu.cfs_period_us /* period = 50ms */”h]”hŒg# echo 10000 > cpu.cfs_quota_us /* quota = 10ms */
# echo 50000 > cpu.cfs_period_us /* period = 50ms */”…””}”hjœ  sbah}”(h]”h ]”h"]”h$]”h&]”j›  jœ  uh1j‹  hŸh¶h Kæhj|  ubh¾)”}”(hŒlBy using a small period here we are ensuring a consistent latency
response at the expense of burst capacity.”h]”hŒlBy using a small period here we are ensuring a consistent latency
response at the expense of burst capacity.”…””}”(hjª  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kéhj|  ubeh}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubj  )”}”(hXæ  Limit a group to 40% of 1 CPU, and allow accumulate up to 20% of 1 CPU
additionally, in case accumulation has been done.

With 50ms period, 20ms quota will be equivalent to 40% of 1 CPU.
And 10ms burst will be equivalent to 20% of 1 CPU::

     # echo 20000 > cpu.cfs_quota_us /* quota = 20ms */
     # echo 50000 > cpu.cfs_period_us /* period = 50ms */
     # echo 10000 > cpu.cfs_burst_us /* burst = 10ms */

Larger buffer setting (no larger than quota) allows greater burst capacity.”h]”(h¾)”}”(hŒxLimit a group to 40% of 1 CPU, and allow accumulate up to 20% of 1 CPU
additionally, in case accumulation has been done.”h]”hŒxLimit a group to 40% of 1 CPU, and allow accumulate up to 20% of 1 CPU
additionally, in case accumulation has been done.”…””}”(hjÂ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kìhj¾  ubh¾)”}”(hŒtWith 50ms period, 20ms quota will be equivalent to 40% of 1 CPU.
And 10ms burst will be equivalent to 20% of 1 CPU::”h]”hŒsWith 50ms period, 20ms quota will be equivalent to 40% of 1 CPU.
And 10ms burst will be equivalent to 20% of 1 CPU:”…””}”(hjÐ  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Kïhj¾  ubjŒ  )”}”(hŒš# echo 20000 > cpu.cfs_quota_us /* quota = 20ms */
# echo 50000 > cpu.cfs_period_us /* period = 50ms */
# echo 10000 > cpu.cfs_burst_us /* burst = 10ms */”h]”hŒš# echo 20000 > cpu.cfs_quota_us /* quota = 20ms */
# echo 50000 > cpu.cfs_period_us /* period = 50ms */
# echo 10000 > cpu.cfs_burst_us /* burst = 10ms */”…””}”hjÞ  sbah}”(h]”h ]”h"]”h$]”h&]”j›  jœ  uh1j‹  hŸh¶h Kòhj¾  ubh¾)”}”(hŒKLarger buffer setting (no larger than quota) allows greater burst capacity.”h]”hŒKLarger buffer setting (no larger than quota) allows greater burst capacity.”…””}”(hjì  hžhhŸNh Nubah}”(h]”h ]”h"]”h$]”h&]”uh1h½hŸh¶h Köhj¾  ubeh}”(h]”h ]”h"]”h$]”h&]”uh1j  hj  hžhhŸh¶h Nubeh}”(h]”h ]”h"]”h$]”h&]”jŽ  Œarabic”j  hj‘  j’  uh1jT  hj  hžhhŸh¶h KÐubeh}”(h]”Œexamples”ah ]”h"]”Œexamples”ah$]”h&]”uh1h¡hh£hžhhŸh¶h KÏubeh}”(h]”Œcfs-bandwidth-control”ah ]”h"]”Œcfs bandwidth control”ah$]”h&]”uh1h¡hhhžhhŸh¶h Kubeh}”(h]”h ]”h"]”h$]”h&]”Œsource”h¶uh1hŒcurrent_source”NŒcurrent_line”NŒsettings”Œdocutils.frontend”ŒValues”“”)”}”(h¦NŒ	generator”NŒ	datestamp”NŒsource_link”NŒ
source_url”NŒtoc_backlinks”Œentry”Œfootnote_backlinks”KŒsectnum_xform”KŒstrip_comments”NŒstrip_elements_with_classes”NŒstrip_classes”NŒreport_level”KŒ
halt_level”KŒexit_status_level”KŒdebug”NŒwarning_stream”NŒ	traceback”ˆŒinput_encoding”Œ	utf-8-sig”Œinput_encoding_error_handler”Œstrict”Œoutput_encoding”Œutf-8”Œoutput_encoding_error_handler”j:  Œerror_encoding”Œutf-8”Œerror_encoding_error_handler”Œbackslashreplace”Œlanguage_code”Œen”Œrecord_dependencies”NŒconfig”NŒ	id_prefix”hŒauto_id_prefix”Œid”Œdump_settings”NŒdump_internals”NŒdump_transforms”NŒdump_pseudo_xml”NŒexpose_internals”NŒstrict_visitor”NŒ_disable_config”NŒ_source”h¶Œ_destination”NŒ_config_files”]”Œ7/var/lib/git/docbuild/linux/Documentation/docutils.conf”aŒfile_insertion_enabled”ˆŒraw_enabled”KŒline_length_limit”M'Œpep_references”NŒpep_base_url”Œhttps://peps.python.org/”Œpep_file_url_template”Œpep-%04d”Œrfc_references”NŒrfc_base_url”Œ&https://datatracker.ietf.org/doc/html/”Œ	tab_width”KŒtrim_footnote_reference_space”‰Œsyntax_highlight”Œlong”Œsmart_quotes”ˆŒsmartquotes_locales”]”Œcharacter_level_inline_markup”‰Œdoctitle_xform”‰Œdocinfo_xform”KŒsectsubtitle_xform”‰Œimage_loading”Œlink”Œembed_stylesheet”‰Œcloak_email_addresses”ˆŒsection_self_link”‰Œenv”NubŒreporter”NŒindirect_targets”]”Œsubstitution_defs”}”Œsubstitution_names”}”Œrefnames”}”Œrefids”}”Œnameids”}”(j  j  j®  j«  jè  jå  j9  j6  jú  j÷  j¬  j©  j  j  j  j	  uŒ	nametypes”}”(j  ‰j®  ‰jè  ‰j9  ‰jú  ‰j¬  ‰j  ‰j  ‰uh}”(j  h£j«  hýjå  j±  j6  jë  j÷  j<  j©  jý  j  j¯  j	  j  uŒfootnote_refs”}”Œcitation_refs”}”Œautofootnotes”]”Œautofootnote_refs”]”Œsymbol_footnotes”]”Œsymbol_footnote_refs”]”Œ	footnotes”]”Œ	citations”]”Œautofootnote_start”KŒsymbol_footnote_start”K Œ
id_counter”Œcollections”ŒCounter”“”}”…”R”Œparse_messages”]”Œtransform_messages”]”Œtransformer”NŒinclude_log”]”Œ
decoration”Nhžhub.