1 REDUCING OS JITTER DUE TO PER-CPU KTHREADS
3 This document lists per-CPU kthreads in the Linux kernel and presents
4 options to control their OS jitter. Note that non-per-CPU kthreads are
5 not listed here. To reduce OS jitter from non-per-CPU kthreads, bind
6 them to a "housekeeping" CPU dedicated to such work.
11 o Documentation/IRQ-affinity.txt: Binding interrupts to sets of CPUs.
13 o Documentation/cgroups: Using cgroups to bind tasks to sets of CPUs.
15 o man taskset: Using the taskset command to bind tasks to sets
18 o man sched_setaffinity: Using the sched_setaffinity() system
19 call to bind tasks to sets of CPUs.
21 o /sys/devices/system/cpu/cpuN/online: Control CPU N's hotplug state,
22 writing "0" to offline and "1" to online.
24 o In order to locate kernel-generated OS jitter on CPU N:
26 cd /sys/kernel/debug/tracing
27 echo 1 > max_graph_depth # Increase the "1" for more detail
28 echo function_graph > current_tracer
30 cat per_cpu/cpuN/trace
36 Purpose: Periodically process Infiniband-related work.
37 To reduce its OS jitter, do any of the following:
38 1. Don't use eHCA Infiniband hardware, instead choosing hardware
39 that does not require per-CPU kthreads. This will prevent these
40 kthreads from being created in the first place. (This will
41 work for most people, as this hardware, though important, is
42 relatively old and is produced in relatively low unit volumes.)
43 2. Do all eHCA-Infiniband-related work on other CPUs, including
45 3. Rework the eHCA driver so that its per-CPU kthreads are
46 provisioned only on selected CPUs.
50 Purpose: Handle threaded interrupts.
51 To reduce its OS jitter, do the following:
52 1. Use irq affinity to force the irq threads to execute on
56 Purpose: Handle Bluetooth work.
57 To reduce its OS jitter, do one of the following:
58 1. Don't use Bluetooth, in which case these kthreads won't be
59 created in the first place.
60 2. Use irq affinity to force Bluetooth-related interrupts to
61 occur on some other CPU and furthermore initiate all
62 Bluetooth activity on some other CPU.
65 Purpose: Execute softirq handlers when threaded or when under heavy load.
66 To reduce its OS jitter, each softirq vector must be handled
67 separately as follows:
68 TIMER_SOFTIRQ: Do all of the following:
69 1. To the extent possible, keep the CPU out of the kernel when it
70 is non-idle, for example, by avoiding system calls and by forcing
71 both kernel threads and interrupts to execute elsewhere.
72 2. Build with CONFIG_HOTPLUG_CPU=y. After boot completes, force
73 the CPU offline, then bring it back online. This forces
74 recurring timers to migrate elsewhere. If you are concerned
75 with multiple CPUs, force them all offline before bringing the
76 first one back online. Once you have onlined the CPUs in question,
77 do not offline any other CPUs, because doing so could force the
78 timer back onto one of the CPUs in question.
79 NET_TX_SOFTIRQ and NET_RX_SOFTIRQ: Do all of the following:
80 1. Force networking interrupts onto other CPUs.
81 2. Initiate any network I/O on other CPUs.
82 3. Once your application has started, prevent CPU-hotplug operations
83 from being initiated from tasks that might run on the CPU to
84 be de-jittered. (It is OK to force this CPU offline and then
85 bring it back online before you start your application.)
86 BLOCK_SOFTIRQ: Do all of the following:
87 1. Force block-device interrupts onto some other CPU.
88 2. Initiate any block I/O on other CPUs.
89 3. Once your application has started, prevent CPU-hotplug operations
90 from being initiated from tasks that might run on the CPU to
91 be de-jittered. (It is OK to force this CPU offline and then
92 bring it back online before you start your application.)
93 BLOCK_IOPOLL_SOFTIRQ: Do all of the following:
94 1. Force block-device interrupts onto some other CPU.
95 2. Initiate any block I/O and block-I/O polling on other CPUs.
96 3. Once your application has started, prevent CPU-hotplug operations
97 from being initiated from tasks that might run on the CPU to
98 be de-jittered. (It is OK to force this CPU offline and then
99 bring it back online before you start your application.)
100 TASKLET_SOFTIRQ: Do one or more of the following:
101 1. Avoid use of drivers that use tasklets. (Such drivers will contain
102 calls to things like tasklet_schedule().)
103 2. Convert all drivers that you must use from tasklets to workqueues.
104 3. Force interrupts for drivers using tasklets onto other CPUs,
105 and also do I/O involving these drivers on other CPUs.
106 SCHED_SOFTIRQ: Do all of the following:
107 1. Avoid sending scheduler IPIs to the CPU to be de-jittered,
108 for example, ensure that at most one runnable kthread is present
109 on that CPU. If a thread that expects to run on the de-jittered
110 CPU awakens, the scheduler will send an IPI that can result in
111 a subsequent SCHED_SOFTIRQ.
112 2. Build with CONFIG_RCU_NOCB_CPU=y, CONFIG_RCU_NOCB_CPU_ALL=y,
113 CONFIG_NO_HZ_FULL=y, and, in addition, ensure that the CPU
114 to be de-jittered is marked as an adaptive-ticks CPU using the
115 "nohz_full=" boot parameter. This reduces the number of
116 scheduler-clock interrupts that the de-jittered CPU receives,
117 minimizing its chances of being selected to do the load balancing
118 work that runs in SCHED_SOFTIRQ context.
119 3. To the extent possible, keep the CPU out of the kernel when it
120 is non-idle, for example, by avoiding system calls and by
121 forcing both kernel threads and interrupts to execute elsewhere.
122 This further reduces the number of scheduler-clock interrupts
123 received by the de-jittered CPU.
124 HRTIMER_SOFTIRQ: Do all of the following:
125 1. To the extent possible, keep the CPU out of the kernel when it
126 is non-idle. For example, avoid system calls and force both
127 kernel threads and interrupts to execute elsewhere.
128 2. Build with CONFIG_HOTPLUG_CPU=y. Once boot completes, force the
129 CPU offline, then bring it back online. This forces recurring
130 timers to migrate elsewhere. If you are concerned with multiple
131 CPUs, force them all offline before bringing the first one
132 back online. Once you have onlined the CPUs in question, do not
133 offline any other CPUs, because doing so could force the timer
134 back onto one of the CPUs in question.
135 RCU_SOFTIRQ: Do at least one of the following:
136 1. Offload callbacks and keep the CPU in either dyntick-idle or
137 adaptive-ticks state by doing all of the following:
138 a. Build with CONFIG_RCU_NOCB_CPU=y, CONFIG_RCU_NOCB_CPU_ALL=y,
139 CONFIG_NO_HZ_FULL=y, and, in addition ensure that the CPU
140 to be de-jittered is marked as an adaptive-ticks CPU using
141 the "nohz_full=" boot parameter. Bind the rcuo kthreads
142 to housekeeping CPUs, which can tolerate OS jitter.
143 b. To the extent possible, keep the CPU out of the kernel
144 when it is non-idle, for example, by avoiding system
145 calls and by forcing both kernel threads and interrupts
146 to execute elsewhere.
147 2. Enable RCU to do its processing remotely via dyntick-idle by
148 doing all of the following:
149 a. Build with CONFIG_NO_HZ=y and CONFIG_RCU_FAST_NO_HZ=y.
150 b. Ensure that the CPU goes idle frequently, allowing other
151 CPUs to detect that it has passed through an RCU quiescent
152 state. If the kernel is built with CONFIG_NO_HZ_FULL=y,
153 userspace execution also allows other CPUs to detect that
154 the CPU in question has passed through a quiescent state.
155 c. To the extent possible, keep the CPU out of the kernel
156 when it is non-idle, for example, by avoiding system
157 calls and by forcing both kernel threads and interrupts
158 to execute elsewhere.
160 Name: kworker/%u:%d%s (cpu, id, priority)
161 Purpose: Execute workqueue requests
162 To reduce its OS jitter, do any of the following:
163 1. Run your workload at a real-time priority, which will allow
164 preempting the kworker daemons.
165 2. A given workqueue can be made visible in the sysfs filesystem
166 by passing the WQ_SYSFS to that workqueue's alloc_workqueue().
167 Such a workqueue can be confined to a given subset of the
168 CPUs using the /sys/devices/virtual/workqueue/*/cpumask sysfs
169 files. The set of WQ_SYSFS workqueues can be displayed using
170 "ls sys/devices/virtual/workqueue". That said, the workqueues
171 maintainer would like to caution people against indiscriminately
172 sprinkling WQ_SYSFS across all the workqueues. The reason for
173 caution is that it is easy to add WQ_SYSFS, but because sysfs is
174 part of the formal user/kernel API, it can be nearly impossible
175 to remove it, even if its addition was a mistake.
176 3. Do any of the following needed to avoid jitter that your
177 application cannot tolerate:
178 a. Build your kernel with CONFIG_SLUB=y rather than
179 CONFIG_SLAB=y, thus avoiding the slab allocator's periodic
180 use of each CPU's workqueues to run its cache_reap()
182 b. Avoid using oprofile, thus avoiding OS jitter from
184 c. Limit your CPU frequency so that a CPU-frequency
185 governor is not required, possibly enlisting the aid of
186 special heatsinks or other cooling technologies. If done
187 correctly, and if you CPU architecture permits, you should
188 be able to build your kernel with CONFIG_CPU_FREQ=n to
189 avoid the CPU-frequency governor periodically running
190 on each CPU, including cs_dbs_timer() and od_dbs_timer().
191 WARNING: Please check your CPU specifications to
192 make sure that this is safe on your particular system.
193 d. It is not possible to entirely get rid of OS jitter
194 from vmstat_update() on CONFIG_SMP=y systems, but you
195 can decrease its frequency by writing a large value
196 to /proc/sys/vm/stat_interval. The default value is
197 HZ, for an interval of one second. Of course, larger
198 values will make your virtual-memory statistics update
199 more slowly. Of course, you can also run your workload
200 at a real-time priority, thus preempting vmstat_update(),
201 but if your workload is CPU-bound, this is a bad idea.
202 However, there is an RFC patch from Christoph Lameter
203 (based on an earlier one from Gilad Ben-Yossef) that
204 reduces or even eliminates vmstat overhead for some
205 workloads at https://lkml.org/lkml/2013/9/4/379.
206 e. If running on high-end powerpc servers, build with
207 CONFIG_PPC_RTAS_DAEMON=n. This prevents the RTAS
208 daemon from running on each CPU every second or so.
209 (This will require editing Kconfig files and will defeat
210 this platform's RAS functionality.) This avoids jitter
211 due to the rtas_event_scan() function.
212 WARNING: Please check your CPU specifications to
213 make sure that this is safe on your particular system.
214 f. If running on Cell Processor, build your kernel with
215 CBE_CPUFREQ_SPU_GOVERNOR=n to avoid OS jitter from
217 WARNING: Please check your CPU specifications to
218 make sure that this is safe on your particular system.
219 g. If running on PowerMAC, build your kernel with
220 CONFIG_PMAC_RACKMETER=n to disable the CPU-meter,
221 avoiding OS jitter from rackmeter_do_timer().
224 Purpose: Execute RCU callbacks in CONFIG_RCU_BOOST=y kernels.
225 To reduce its OS jitter, do at least one of the following:
226 1. Build the kernel with CONFIG_PREEMPT=n. This prevents these
227 kthreads from being created in the first place, and also obviates
228 the need for RCU priority boosting. This approach is feasible
229 for workloads that do not require high degrees of responsiveness.
230 2. Build the kernel with CONFIG_RCU_BOOST=n. This prevents these
231 kthreads from being created in the first place. This approach
232 is feasible only if your workload never requires RCU priority
233 boosting, for example, if you ensure frequent idle time on all
234 CPUs that might execute within the kernel.
235 3. Build with CONFIG_RCU_NOCB_CPU=y and CONFIG_RCU_NOCB_CPU_ALL=y,
236 which offloads all RCU callbacks to kthreads that can be moved
237 off of CPUs susceptible to OS jitter. This approach prevents the
238 rcuc/%u kthreads from having any work to do, so that they are
240 4. Ensure that the CPU never enters the kernel, and, in particular,
241 avoid initiating any CPU hotplug operations on this CPU. This is
242 another way of preventing any callbacks from being queued on the
243 CPU, again preventing the rcuc/%u kthreads from having any work
246 Name: rcuob/%d, rcuop/%d, and rcuos/%d
247 Purpose: Offload RCU callbacks from the corresponding CPU.
248 To reduce its OS jitter, do at least one of the following:
249 1. Use affinity, cgroups, or other mechanism to force these kthreads
250 to execute on some other CPU.
251 2. Build with CONFIG_RCU_NOCB_CPU=n, which will prevent these
252 kthreads from being created in the first place. However, please
253 note that this will not eliminate OS jitter, but will instead
254 shift it to RCU_SOFTIRQ.
257 Purpose: Detect software lockups on each CPU.
258 To reduce its OS jitter, do at least one of the following:
259 1. Build with CONFIG_LOCKUP_DETECTOR=n, which will prevent these
260 kthreads from being created in the first place.
261 2. Echo a zero to /proc/sys/kernel/watchdog to disable the
263 3. Echo a large number of /proc/sys/kernel/watchdog_thresh in
264 order to reduce the frequency of OS jitter due to the watchdog
265 timer down to a level that is acceptable for your workload.