1IRQBALANCE(1) irqbalance IRQBALANCE(1)
2
6 irqbalance - distribute hardware interrupts across processors on a mul‐
7 tiprocessor system
8
10 irqbalance
11
14 The purpose of irqbalance is to distribute hardware interrupts across
15 processors on a multiprocessor system in order to increase performance.
16
19 -o, --oneshot
20 Causes irqbalance to be run once, after which the daemon exits.
21 -d, --debug
24 Causes irqbalance to print extra debug information. Implies
25 --foreground.
26 -f, --foreground
29 Causes irqbalance to run in the foreground (without --debug).
30 -j, --journal
33 Enables log output optimized for systemd-journal.
34 -p, --powerthresh=<threshold>
37 Set the threshold at which we attempt to move a CPU into power‐
38 save mode If more than <threshold> CPUs are more than 1 standard
39 deviation below the average CPU softirq workload, and no CPUs
40 are more than 1 standard deviation above (and have more than 1
41 IRQ assigned to them), attempt to place 1 CPU in powersave mode.
42 In powersave mode, a CPU will not have any IRQs balanced to it,
43 in an effort to prevent that CPU from waking up without need.
44 -i, --banirq=<irqnum>
47 Add the specified IRQ to the set of banned IRQs. irqbalance will
48 not affect the affinity of any IRQs on the banned list, allowing
49 them to be specified manually. This option is additive and can
50 be specified multiple times. For example to ban IRQs 43 and 44
51 from balancing, use the following command line: irqbalance
52 --banirq=43 --banirq=44
53 -m, --banmod=<module_name>
56 Add the specified module to the set of banned modules, similar
57 to --banirq. irqbalance will not affect the affinity of any
58 IRQs of given modules, allowing them to be specified manually.
59 This option is additive and can be specified multiple times. For
60 example to ban all IRQs of module foo and module bar from bal‐
61 ancing, use the following command line: irqbalance --banmod=foo
62 --banmod=bar
63 -c, --deepestcache=<integer>
66 This allows a user to specify the cache level at which irqbal‐
67 ance partitions cache domains. Specifying a deeper cache may
68 allow a greater degree of flexibility for irqbalance to assign
69 IRQ affinity to achieve greater performance increases, but set‐
70 ting a cache depth too large on some systems (specifically where
71 all CPUs on a system share the deepest cache level), will cause
72 irqbalance to see balancing as unnecessary. irqbalance --deep‐
73 estcache=2
74 The default value for deepestcache is 2.
76 -l, --policyscript=<script>
79 When specified, the referenced script or directory will execute
80 once for each discovered IRQ, with the sysfs device path and IRQ
81 number passed as arguments. Note that the device path argument
82 will point to the parent directory from which the IRQ attributes
83 directory may be directly opened. Policy scripts specified need
84 to be owned and executable by the user of irqbalance process, if
85 a directory is specified, non-executable files will be skipped.
86 The script may specify zero or more key=value pairs that will
87 guide irqbalance in the management of that IRQ. Key=value pairs
88 are printed by the script on stdout and will be captured and in‐
89 terpreted by irqbalance. Irqbalance expects a zero exit code
90 from the provided utility. Recognized key=value pairs are:
91 ban=[true | false]
93 Directs irqbalance to exclude the passed in IRQ from balancing.
94 balance_level=[none | package | cache | core]
96 This allows a user to override the balance level of a given IRQ.
97 By default the balance level is determined automatically based
98 on the pci device class of the device that owns the IRQ.
99 numa_node=<integer>
101 This allows a user to override the NUMA node that sysfs indi‐
102 cates a given device IRQ is local to. Often, systems will not
103 specify this information in ACPI, and as a result devices are
104 considered equidistant from all NUMA nodes in a system. This
105 option allows for that hardware provided information to be over‐
106 ridden, so that irqbalance can bias IRQ affinity for these de‐
107 vices toward its most local node. Note that specifying a -1
108 here forces irqbalance to consider an interrupt from a device to
109 be equidistant from all nodes.
110 Note that, if a directory is specified rather than a regular file, all
112 files in
113 the directory will be considered policy scripts, and executed on
114 adding of an irq to a database. If such a directory is speci‐
115 fied, scripts in the directory must additionally exit with one
116 of the following exit codes:
117 0 This indicates the script has a policy for the referenced irq,
119 and that further script processing should stop
120 1 This indicates that the script has no policy for the referenced
122 irq, and that script processing should continue
123 2 This indicates that an error has occurred in the script, and it
125 should be skipped (further processing to continue)
126 --migrateval, -e <val>
129 Specify a minimum migration ratio to trigger a rebalancing Nor‐
130 mally any improvement in load distribution will trigger the mi‐
131 gration of an irq, as long as preforming the migration will not
132 simply move the load to a new cpu. By specifying a migration
133 value, the load balance improvement is subject to hysteresis de‐
134 fined by this value, which is inversely propotional to the
135 value. For example, a value of 2 in this option tells irqbal‐
136 ance that the improvement in load distribution must be at least
137 50%, a value of 4 indicates the load distribution improvement
138 must be at least 25%, etc
139 -s, --pid=<file>
142 Have irqbalance write its process id to the specified file. By
143 default no pidfile is written. The written pidfile is automati‐
144 cally unlinked when irqbalance exits. It is ignored when used
145 with --debug or --foreground.
146 -t, --interval=<time>
148 Set the measurement time for irqbalance. irqbalance will sleep
149 for <time> seconds between samples of the irq load on the system
150 cpus. Defaults to 10.
151
153 IRQBALANCE_ONESHOT
154 Same as --oneshot.
155 IRQBALANCE_DEBUG
158 Same as --debug.
159 IRQBALANCE_BANNED_CPUS
162 Provides a mask of CPUs which irqbalance should ignore and never
163 assign interrupts to. If not specified, irqbalance use mask of
164 isolated and adaptive-ticks CPUs on the system as the default
165 value. The "isolcpus=" boot parameter specifies the isolated
166 CPUs. The "nohz_full=" boot parameter specifies the adaptive-
167 ticks CPUs. By default, no CPU will be an isolated or adaptive-
168 ticks CPU. This is a hexmask without the leading ’0x’. On sys‐
169 tems with large numbers of processors, each group of eight hex
170 digits is separated by a comma ’,’. i.e. ‘export IRQBAL‐
171 ANCE_BANNED_CPUS=fc0‘ would prevent irqbalance from assigning
172 irqs to the 7th-12th cpus (cpu6-cpu11) or ‘export IRQBAL‐
173 ANCE_BANNED_CPUS=ff000000,00000001‘ would prevent irqbalance
174 from assigning irqs to the 1st (cpu0) and 57th-64th cpus
175 (cpu56-cpu63). Notes: This environment variable will be dis‐
176 carded, please use IRQBALANCE_BANNED_CPULIST instead. Before
177 deleting this environment variable, Introduce a deprecation pe‐
178 riod first for the consider of compatibility.
179 IRQBALANCE_BANNED_CPULIST
182 Provides a cpulist which irqbalance should ignore and never as‐
183 sign interrupts to. If not specified, irqbalance use mask of
184 isolated and adaptive-ticks CPUs on the system as the default
185 value.
186
189 SIGHUP Forces a rescan of the available IRQs and system topology.
190
193 irqbalance is able to communicate via socket and return it's current
194 assignment tree and setup, as well as set new settings based on sent
195 values. Socket is abstract, with a name in form of irqbalance<PID>.sock
196 , where <PID> is the process ID of irqbalance instance to communicate
197 with. Possible values to send:
198 stats Retrieve assignment tree of IRQs to CPUs, in recursive manner.
200 For each CPU node in tree, it's type, number, load and whether
201 the save mode is active are sent. For each assigned IRQ type,
202 it's number, load, number of IRQs since last rebalancing and
203 it's class are sent. Refer to types.h file for explanation of
204 defines.
205 setup Get the current value of sleep interval, mask of banned CPUs and
207 list of banned IRQs.
208 settings sleep <s>
210 Set new value of sleep interval, <s> >= 1.
211 settings cpus <cpu_number1> <cpu_number2> ...
213 Ban listed CPUs from IRQ handling, all old values of banned CPUs
214 are forgotten.
215 settings ban irqs <irq1> <irq2> ...
217 Ban listed IRQs from being balanced, all old values of banned
218 IRQs are forgotten.
219 irqbalance checks SCM_CREDENTIALS of sender (only root user is allowed
221 to interact). Based on chosen tools, ancillary message with creden‐
222 tials needs to be sent with request.
223
226 https://github.com/Irqbalance/irqbalance
227Linux Dec 2006 IRQBALANCE(1)