blob: 4ed7b5ceeed212c1b7a734451456896f4759ee9d [file] [log] [blame]
Vivek Goyal72f924f2009-12-03 12:59:57 -05001 Block IO Controller
2 ===================
3Overview
4========
5cgroup subsys "blkio" implements the block io controller. There seems to be
6a need of various kinds of IO control policies (like proportional BW, max BW)
7both at leaf nodes as well as at intermediate nodes in a storage hierarchy.
8Plan is to use the same cgroup based management interface for blkio controller
9and based on user options switch IO policies in the background.
10
Vivek Goyal2786c4e2010-09-15 17:06:38 -040011Currently two IO control policies are implemented. First one is proportional
12weight time based division of disk policy. It is implemented in CFQ. Hence
13this policy takes effect only on leaf nodes when CFQ is being used. The second
14one is throttling policy which can be used to specify upper IO rate limits
15on devices. This policy is implemented in generic block layer and can be
16used on leaf nodes as well as higher level logical devices like device mapper.
Vivek Goyal72f924f2009-12-03 12:59:57 -050017
18HOWTO
19=====
Vivek Goyal2786c4e2010-09-15 17:06:38 -040020Proportional Weight division of bandwidth
21-----------------------------------------
Vivek Goyal72f924f2009-12-03 12:59:57 -050022You can do a very simple testing of running two dd threads in two different
23cgroups. Here is what you can do.
24
Vivek Goyalafc24d42010-04-26 19:27:56 +020025- Enable Block IO controller
26 CONFIG_BLK_CGROUP=y
27
Vivek Goyal72f924f2009-12-03 12:59:57 -050028- Enable group scheduling in CFQ
29 CONFIG_CFQ_GROUP_IOSCHED=y
30
31- Compile and boot into kernel and mount IO controller (blkio).
32
33 mount -t cgroup -o blkio none /cgroup
34
35- Create two cgroups
36 mkdir -p /cgroup/test1/ /cgroup/test2
37
38- Set weights of group test1 and test2
39 echo 1000 > /cgroup/test1/blkio.weight
40 echo 500 > /cgroup/test2/blkio.weight
41
42- Create two same size files (say 512MB each) on same disk (file1, file2) and
43 launch two dd threads in different cgroup to read those files.
44
45 sync
46 echo 3 > /proc/sys/vm/drop_caches
47
48 dd if=/mnt/sdb/zerofile1 of=/dev/null &
49 echo $! > /cgroup/test1/tasks
50 cat /cgroup/test1/tasks
51
52 dd if=/mnt/sdb/zerofile2 of=/dev/null &
53 echo $! > /cgroup/test2/tasks
54 cat /cgroup/test2/tasks
55
56- At macro level, first dd should finish first. To get more precise data, keep
57 on looking at (with the help of script), at blkio.disk_time and
58 blkio.disk_sectors files of both test1 and test2 groups. This will tell how
59 much disk time (in milli seconds), each group got and how many secotors each
60 group dispatched to the disk. We provide fairness in terms of disk time, so
61 ideally io.disk_time of cgroups should be in proportion to the weight.
62
Vivek Goyal2786c4e2010-09-15 17:06:38 -040063Throttling/Upper Limit policy
64-----------------------------
65- Enable Block IO controller
66 CONFIG_BLK_CGROUP=y
67
68- Enable throttling in block layer
69 CONFIG_BLK_DEV_THROTTLING=y
70
71- Mount blkio controller
72 mount -t cgroup -o blkio none /cgroup/blkio
73
74- Specify a bandwidth rate on particular device for root group. The format
75 for policy is "<major>:<minor> <byes_per_second>".
76
77 echo "8:16 1048576" > /cgroup/blkio/blkio.read_bps_device
78
79 Above will put a limit of 1MB/second on reads happening for root group
80 on device having major/minor number 8:16.
81
82- Run dd to read a file and see if rate is throttled to 1MB/s or not.
83
84 # dd if=/mnt/common/zerofile of=/dev/null bs=4K count=1024
85 # iflag=direct
86 1024+0 records in
87 1024+0 records out
88 4194304 bytes (4.2 MB) copied, 4.0001 s, 1.0 MB/s
89
90 Limits for writes can be put using blkio.write_bps_device file.
91
Vivek Goyalbdc85df2010-11-15 19:37:36 +010092Hierarchical Cgroups
93====================
94- Currently none of the IO control policy supports hierarhical groups. But
95 cgroup interface does allow creation of hierarhical cgroups and internally
96 IO policies treat them as flat hierarchy.
97
98 So this patch will allow creation of cgroup hierarhcy but at the backend
99 everything will be treated as flat. So if somebody created a hierarchy like
100 as follows.
101
102 root
103 / \
104 test1 test2
105 |
106 test3
107
108 CFQ and throttling will practically treat all groups at same level.
109
110 pivot
111 / | \ \
112 root test1 test2 test3
113
114 Down the line we can implement hierarchical accounting/control support
115 and also introduce a new cgroup file "use_hierarchy" which will control
116 whether cgroup hierarchy is viewed as flat or hierarchical by the policy..
117 This is how memory controller also has implemented the things.
118
Vivek Goyal72f924f2009-12-03 12:59:57 -0500119Various user visible config options
120===================================
Vivek Goyalafc24d42010-04-26 19:27:56 +0200121CONFIG_BLK_CGROUP
122 - Block IO controller.
123
124CONFIG_DEBUG_BLK_CGROUP
125 - Debug help. Right now some additional stats file show up in cgroup
126 if this option is enabled.
127
Vivek Goyal72f924f2009-12-03 12:59:57 -0500128CONFIG_CFQ_GROUP_IOSCHED
129 - Enables group scheduling in CFQ. Currently only 1 level of group
130 creation is allowed.
131
Vivek Goyal2786c4e2010-09-15 17:06:38 -0400132CONFIG_BLK_DEV_THROTTLING
133 - Enable block device throttling support in block layer.
134
Vivek Goyal72f924f2009-12-03 12:59:57 -0500135Details of cgroup files
136=======================
Vivek Goyal2786c4e2010-09-15 17:06:38 -0400137Proportional weight policy files
138--------------------------------
Vivek Goyal72f924f2009-12-03 12:59:57 -0500139- blkio.weight
Gui Jianfengda69da12010-04-13 16:07:50 +0800140 - Specifies per cgroup weight. This is default weight of the group
141 on all the devices until and unless overridden by per device rule.
142 (See blkio.weight_device).
Vivek Goyal72f924f2009-12-03 12:59:57 -0500143 Currently allowed range of weights is from 100 to 1000.
144
Gui Jianfengda69da12010-04-13 16:07:50 +0800145- blkio.weight_device
146 - One can specify per cgroup per device rules using this interface.
147 These rules override the default value of group weight as specified
148 by blkio.weight.
149
150 Following is the format.
151
152 #echo dev_maj:dev_minor weight > /path/to/cgroup/blkio.weight_device
153 Configure weight=300 on /dev/sdb (8:16) in this cgroup
154 # echo 8:16 300 > blkio.weight_device
155 # cat blkio.weight_device
156 dev weight
157 8:16 300
158
159 Configure weight=500 on /dev/sda (8:0) in this cgroup
160 # echo 8:0 500 > blkio.weight_device
161 # cat blkio.weight_device
162 dev weight
163 8:0 500
164 8:16 300
165
166 Remove specific weight for /dev/sda in this cgroup
167 # echo 8:0 0 > blkio.weight_device
168 # cat blkio.weight_device
169 dev weight
170 8:16 300
171
Vivek Goyal72f924f2009-12-03 12:59:57 -0500172- blkio.time
173 - disk time allocated to cgroup per device in milliseconds. First
174 two fields specify the major and minor number of the device and
175 third field specifies the disk time allocated to group in
176 milliseconds.
177
178- blkio.sectors
179 - number of sectors transferred to/from disk by the group. First
180 two fields specify the major and minor number of the device and
181 third field specifies the number of sectors transferred by the
182 group to/from the device.
183
Divyesh Shah84c124d2010-04-09 08:31:19 +0200184- blkio.io_service_bytes
185 - Number of bytes transferred to/from the disk by the group. These
186 are further divided by the type of operation - read or write, sync
187 or async. First two fields specify the major and minor number of the
188 device, third field specifies the operation type and the fourth field
189 specifies the number of bytes.
190
191- blkio.io_serviced
192 - Number of IOs completed to/from the disk by the group. These
193 are further divided by the type of operation - read or write, sync
194 or async. First two fields specify the major and minor number of the
195 device, third field specifies the operation type and the fourth field
196 specifies the number of IOs.
197
198- blkio.io_service_time
199 - Total amount of time between request dispatch and request completion
200 for the IOs done by this cgroup. This is in nanoseconds to make it
201 meaningful for flash devices too. For devices with queue depth of 1,
202 this time represents the actual service time. When queue_depth > 1,
203 that is no longer true as requests may be served out of order. This
204 may cause the service time for a given IO to include the service time
205 of multiple IOs when served out of order which may result in total
206 io_service_time > actual time elapsed. This time is further divided by
207 the type of operation - read or write, sync or async. First two fields
208 specify the major and minor number of the device, third field
209 specifies the operation type and the fourth field specifies the
210 io_service_time in ns.
211
212- blkio.io_wait_time
213 - Total amount of time the IOs for this cgroup spent waiting in the
214 scheduler queues for service. This can be greater than the total time
215 elapsed since it is cumulative io_wait_time for all IOs. It is not a
216 measure of total time the cgroup spent waiting but rather a measure of
217 the wait_time for its individual IOs. For devices with queue_depth > 1
218 this metric does not include the time spent waiting for service once
219 the IO is dispatched to the device but till it actually gets serviced
220 (there might be a time lag here due to re-ordering of requests by the
221 device). This is in nanoseconds to make it meaningful for flash
222 devices too. This time is further divided by the type of operation -
223 read or write, sync or async. First two fields specify the major and
224 minor number of the device, third field specifies the operation type
225 and the fourth field specifies the io_wait_time in ns.
226
Divyesh Shah812d4022010-04-08 21:14:23 -0700227- blkio.io_merged
228 - Total number of bios/requests merged into requests belonging to this
229 cgroup. This is further divided by the type of operation - read or
230 write, sync or async.
231
Divyesh Shahcdc11842010-04-08 21:15:10 -0700232- blkio.io_queued
233 - Total number of requests queued up at any given instant for this
234 cgroup. This is further divided by the type of operation - read or
235 write, sync or async.
236
237- blkio.avg_queue_size
Vivek Goyalafc24d42010-04-26 19:27:56 +0200238 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
Divyesh Shahcdc11842010-04-08 21:15:10 -0700239 The average queue size for this cgroup over the entire time of this
240 cgroup's existence. Queue size samples are taken each time one of the
241 queues of this cgroup gets a timeslice.
242
Divyesh Shah812df482010-04-08 21:15:35 -0700243- blkio.group_wait_time
Vivek Goyalafc24d42010-04-26 19:27:56 +0200244 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
Divyesh Shah812df482010-04-08 21:15:35 -0700245 This is the amount of time the cgroup had to wait since it became busy
246 (i.e., went from 0 to 1 request queued) to get a timeslice for one of
247 its queues. This is different from the io_wait_time which is the
248 cumulative total of the amount of time spent by each IO in that cgroup
249 waiting in the scheduler queue. This is in nanoseconds. If this is
250 read when the cgroup is in a waiting (for timeslice) state, the stat
251 will only report the group_wait_time accumulated till the last time it
252 got a timeslice and will not include the current delta.
253
254- blkio.empty_time
Vivek Goyalafc24d42010-04-26 19:27:56 +0200255 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
Divyesh Shah812df482010-04-08 21:15:35 -0700256 This is the amount of time a cgroup spends without any pending
257 requests when not being served, i.e., it does not include any time
258 spent idling for one of the queues of the cgroup. This is in
259 nanoseconds. If this is read when the cgroup is in an empty state,
260 the stat will only report the empty_time accumulated till the last
261 time it had a pending request and will not include the current delta.
262
263- blkio.idle_time
Vivek Goyalafc24d42010-04-26 19:27:56 +0200264 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
Divyesh Shah812df482010-04-08 21:15:35 -0700265 This is the amount of time spent by the IO scheduler idling for a
266 given cgroup in anticipation of a better request than the exising ones
267 from other queues/cgroups. This is in nanoseconds. If this is read
268 when the cgroup is in an idling state, the stat will only report the
269 idle_time accumulated till the last idle period and will not include
270 the current delta.
271
Vivek Goyal72f924f2009-12-03 12:59:57 -0500272- blkio.dequeue
Vivek Goyalafc24d42010-04-26 19:27:56 +0200273 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. This
Vivek Goyal72f924f2009-12-03 12:59:57 -0500274 gives the statistics about how many a times a group was dequeued
275 from service tree of the device. First two fields specify the major
276 and minor number of the device and third field specifies the number
277 of times a group was dequeued from a particular device.
278
Vivek Goyal2786c4e2010-09-15 17:06:38 -0400279Throttling/Upper limit policy files
280-----------------------------------
281- blkio.throttle.read_bps_device
282 - Specifies upper limit on READ rate from the device. IO rate is
283 specified in bytes per second. Rules are per deivce. Following is
284 the format.
285
286 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.read_bps_device
287
288- blkio.throttle.write_bps_device
289 - Specifies upper limit on WRITE rate to the device. IO rate is
290 specified in bytes per second. Rules are per deivce. Following is
291 the format.
292
293 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.write_bps_device
294
295- blkio.throttle.read_iops_device
296 - Specifies upper limit on READ rate from the device. IO rate is
297 specified in IO per second. Rules are per deivce. Following is
298 the format.
299
300 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.read_iops_device
301
302- blkio.throttle.write_iops_device
303 - Specifies upper limit on WRITE rate to the device. IO rate is
304 specified in io per second. Rules are per deivce. Following is
305 the format.
306
307 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.write_iops_device
308
309Note: If both BW and IOPS rules are specified for a device, then IO is
310 subjectd to both the constraints.
311
312- blkio.throttle.io_serviced
313 - Number of IOs (bio) completed to/from the disk by the group (as
314 seen by throttling policy). These are further divided by the type
315 of operation - read or write, sync or async. First two fields specify
316 the major and minor number of the device, third field specifies the
317 operation type and the fourth field specifies the number of IOs.
318
319 blkio.io_serviced does accounting as seen by CFQ and counts are in
320 number of requests (struct request). On the other hand,
321 blkio.throttle.io_serviced counts number of IO in terms of number
322 of bios as seen by throttling policy. These bios can later be
323 merged by elevator and total number of requests completed can be
324 lesser.
325
326- blkio.throttle.io_service_bytes
327 - Number of bytes transferred to/from the disk by the group. These
328 are further divided by the type of operation - read or write, sync
329 or async. First two fields specify the major and minor number of the
330 device, third field specifies the operation type and the fourth field
331 specifies the number of bytes.
332
333 These numbers should roughly be same as blkio.io_service_bytes as
334 updated by CFQ. The difference between two is that
335 blkio.io_service_bytes will not be updated if CFQ is not operating
336 on request queue.
337
338Common files among various policies
339-----------------------------------
Divyesh Shah84c124d2010-04-09 08:31:19 +0200340- blkio.reset_stats
341 - Writing an int to this file will result in resetting all the stats
342 for that cgroup.
343
Vivek Goyal72f924f2009-12-03 12:59:57 -0500344CFQ sysfs tunable
345=================
346/sys/block/<disk>/queue/iosched/group_isolation
Vivek Goyal6d6ac1c2010-08-23 12:25:29 +0200347-----------------------------------------------
Vivek Goyal72f924f2009-12-03 12:59:57 -0500348
349If group_isolation=1, it provides stronger isolation between groups at the
350expense of throughput. By default group_isolation is 0. In general that
351means that if group_isolation=0, expect fairness for sequential workload
352only. Set group_isolation=1 to see fairness for random IO workload also.
353
354Generally CFQ will put random seeky workload in sync-noidle category. CFQ
355will disable idling on these queues and it does a collective idling on group
356of such queues. Generally these are slow moving queues and if there is a
357sync-noidle service tree in each group, that group gets exclusive access to
358disk for certain period. That means it will bring the throughput down if
359group does not have enough IO to drive deeper queue depths and utilize disk
360capacity to the fullest in the slice allocated to it. But the flip side is
361that even a random reader should get better latencies and overall throughput
362if there are lots of sequential readers/sync-idle workload running in the
363system.
364
365If group_isolation=0, then CFQ automatically moves all the random seeky queues
366in the root group. That means there will be no service differentiation for
367that kind of workload. This leads to better throughput as we do collective
368idling on root sync-noidle tree.
369
370By default one should run with group_isolation=0. If that is not sufficient
371and one wants stronger isolation between groups, then set group_isolation=1
372but this will come at cost of reduced throughput.
373
Vivek Goyal6d6ac1c2010-08-23 12:25:29 +0200374/sys/block/<disk>/queue/iosched/slice_idle
375------------------------------------------
376On a faster hardware CFQ can be slow, especially with sequential workload.
377This happens because CFQ idles on a single queue and single queue might not
378drive deeper request queue depths to keep the storage busy. In such scenarios
379one can try setting slice_idle=0 and that would switch CFQ to IOPS
380(IO operations per second) mode on NCQ supporting hardware.
381
382That means CFQ will not idle between cfq queues of a cfq group and hence be
383able to driver higher queue depth and achieve better throughput. That also
384means that cfq provides fairness among groups in terms of IOPS and not in
385terms of disk time.
386
387/sys/block/<disk>/queue/iosched/group_idle
388------------------------------------------
389If one disables idling on individual cfq queues and cfq service trees by
390setting slice_idle=0, group_idle kicks in. That means CFQ will still idle
391on the group in an attempt to provide fairness among groups.
392
393By default group_idle is same as slice_idle and does not do anything if
394slice_idle is enabled.
395
396One can experience an overall throughput drop if you have created multiple
397groups and put applications in that group which are not driving enough
398IO to keep disk busy. In that case set group_idle=0, and CFQ will not idle
399on individual groups and throughput should improve.
400
Vivek Goyal72f924f2009-12-03 12:59:57 -0500401What works
402==========
403- Currently only sync IO queues are support. All the buffered writes are
404 still system wide and not per group. Hence we will not see service
405 differentiation between buffered writes between groups.