1=================== 2Block IO Controller 3=================== 4 5Overview 6======== 7cgroup subsys "blkio" implements the block io controller. There seems to be 8a need of various kinds of IO control policies (like proportional BW, max BW) 9both at leaf nodes as well as at intermediate nodes in a storage hierarchy. 10Plan is to use the same cgroup based management interface for blkio controller 11and based on user options switch IO policies in the background. 12 13One IO control policy is throttling policy which can be used to 14specify upper IO rate limits on devices. This policy is implemented in 15generic block layer and can be used on leaf nodes as well as higher 16level logical devices like device mapper. 17 18HOWTO 19===== 20Throttling/Upper Limit policy 21----------------------------- 22- Enable Block IO controller:: 23 24 CONFIG_BLK_CGROUP=y 25 26- Enable throttling in block layer:: 27 28 CONFIG_BLK_DEV_THROTTLING=y 29 30- Mount blkio controller (see cgroups.txt, Why are cgroups needed?):: 31 32 mount -t cgroup -o blkio none /sys/fs/cgroup/blkio 33 34- Specify a bandwidth rate on particular device for root group. The format 35 for policy is "<major>:<minor> <bytes_per_second>":: 36 37 echo "8:16 1048576" > /sys/fs/cgroup/blkio/blkio.throttle.read_bps_device 38 39 Above will put a limit of 1MB/second on reads happening for root group 40 on device having major/minor number 8:16. 41 42- Run dd to read a file and see if rate is throttled to 1MB/s or not:: 43 44 # dd iflag=direct if=/mnt/common/zerofile of=/dev/null bs=4K count=1024 45 1024+0 records in 46 1024+0 records out 47 4194304 bytes (4.2 MB) copied, 4.0001 s, 1.0 MB/s 48 49 Limits for writes can be put using blkio.throttle.write_bps_device file. 50 51Hierarchical Cgroups 52==================== 53 54Throttling implements hierarchy support; however, 55throttling's hierarchy support is enabled iff "sane_behavior" is 56enabled from cgroup side, which currently is a development option and 57not publicly available. 58 59If somebody created a hierarchy like as follows:: 60 61 root 62 / \ 63 test1 test2 64 | 65 test3 66 67Throttling with "sane_behavior" will handle the 68hierarchy correctly. For throttling, all limits apply 69to the whole subtree while all statistics are local to the IOs 70directly generated by tasks in that cgroup. 71 72Throttling without "sane_behavior" enabled from cgroup side will 73practically treat all groups at same level as if it looks like the 74following:: 75 76 pivot 77 / / \ \ 78 root test1 test2 test3 79 80Various user visible config options 81=================================== 82CONFIG_BLK_CGROUP 83 - Block IO controller. 84 85CONFIG_BFQ_CGROUP_DEBUG 86 - Debug help. Right now some additional stats file show up in cgroup 87 if this option is enabled. 88 89CONFIG_BLK_DEV_THROTTLING 90 - Enable block device throttling support in block layer. 91 92Details of cgroup files 93======================= 94Proportional weight policy files 95-------------------------------- 96- blkio.weight 97 - Specifies per cgroup weight. This is default weight of the group 98 on all the devices until and unless overridden by per device rule. 99 (See blkio.weight_device). 100 Currently allowed range of weights is from 10 to 1000. 101 102- blkio.weight_device 103 - One can specify per cgroup per device rules using this interface. 104 These rules override the default value of group weight as specified 105 by blkio.weight. 106 107 Following is the format:: 108 109 # echo dev_maj:dev_minor weight > blkio.weight_device 110 111 Configure weight=300 on /dev/sdb (8:16) in this cgroup:: 112 113 # echo 8:16 300 > blkio.weight_device 114 # cat blkio.weight_device 115 dev weight 116 8:16 300 117 118 Configure weight=500 on /dev/sda (8:0) in this cgroup:: 119 120 # echo 8:0 500 > blkio.weight_device 121 # cat blkio.weight_device 122 dev weight 123 8:0 500 124 8:16 300 125 126 Remove specific weight for /dev/sda in this cgroup:: 127 128 # echo 8:0 0 > blkio.weight_device 129 # cat blkio.weight_device 130 dev weight 131 8:16 300 132 133- blkio.leaf_weight[_device] 134 - Equivalents of blkio.weight[_device] for the purpose of 135 deciding how much weight tasks in the given cgroup has while 136 competing with the cgroup's child cgroups. For details, 137 please refer to Documentation/block/cfq-iosched.txt. 138 139- blkio.time 140 - disk time allocated to cgroup per device in milliseconds. First 141 two fields specify the major and minor number of the device and 142 third field specifies the disk time allocated to group in 143 milliseconds. 144 145- blkio.sectors 146 - number of sectors transferred to/from disk by the group. First 147 two fields specify the major and minor number of the device and 148 third field specifies the number of sectors transferred by the 149 group to/from the device. 150 151- blkio.io_service_bytes 152 - Number of bytes transferred to/from the disk by the group. These 153 are further divided by the type of operation - read or write, sync 154 or async. First two fields specify the major and minor number of the 155 device, third field specifies the operation type and the fourth field 156 specifies the number of bytes. 157 158- blkio.io_serviced 159 - Number of IOs (bio) issued to the disk by the group. These 160 are further divided by the type of operation - read or write, sync 161 or async. First two fields specify the major and minor number of the 162 device, third field specifies the operation type and the fourth field 163 specifies the number of IOs. 164 165- blkio.io_service_time 166 - Total amount of time between request dispatch and request completion 167 for the IOs done by this cgroup. This is in nanoseconds to make it 168 meaningful for flash devices too. For devices with queue depth of 1, 169 this time represents the actual service time. When queue_depth > 1, 170 that is no longer true as requests may be served out of order. This 171 may cause the service time for a given IO to include the service time 172 of multiple IOs when served out of order which may result in total 173 io_service_time > actual time elapsed. This time is further divided by 174 the type of operation - read or write, sync or async. First two fields 175 specify the major and minor number of the device, third field 176 specifies the operation type and the fourth field specifies the 177 io_service_time in ns. 178 179- blkio.io_wait_time 180 - Total amount of time the IOs for this cgroup spent waiting in the 181 scheduler queues for service. This can be greater than the total time 182 elapsed since it is cumulative io_wait_time for all IOs. It is not a 183 measure of total time the cgroup spent waiting but rather a measure of 184 the wait_time for its individual IOs. For devices with queue_depth > 1 185 this metric does not include the time spent waiting for service once 186 the IO is dispatched to the device but till it actually gets serviced 187 (there might be a time lag here due to re-ordering of requests by the 188 device). This is in nanoseconds to make it meaningful for flash 189 devices too. This time is further divided by the type of operation - 190 read or write, sync or async. First two fields specify the major and 191 minor number of the device, third field specifies the operation type 192 and the fourth field specifies the io_wait_time in ns. 193 194- blkio.io_merged 195 - Total number of bios/requests merged into requests belonging to this 196 cgroup. This is further divided by the type of operation - read or 197 write, sync or async. 198 199- blkio.io_queued 200 - Total number of requests queued up at any given instant for this 201 cgroup. This is further divided by the type of operation - read or 202 write, sync or async. 203 204- blkio.avg_queue_size 205 - Debugging aid only enabled if CONFIG_BFQ_CGROUP_DEBUG=y. 206 The average queue size for this cgroup over the entire time of this 207 cgroup's existence. Queue size samples are taken each time one of the 208 queues of this cgroup gets a timeslice. 209 210- blkio.group_wait_time 211 - Debugging aid only enabled if CONFIG_BFQ_CGROUP_DEBUG=y. 212 This is the amount of time the cgroup had to wait since it became busy 213 (i.e., went from 0 to 1 request queued) to get a timeslice for one of 214 its queues. This is different from the io_wait_time which is the 215 cumulative total of the amount of time spent by each IO in that cgroup 216 waiting in the scheduler queue. This is in nanoseconds. If this is 217 read when the cgroup is in a waiting (for timeslice) state, the stat 218 will only report the group_wait_time accumulated till the last time it 219 got a timeslice and will not include the current delta. 220 221- blkio.empty_time 222 - Debugging aid only enabled if CONFIG_BFQ_CGROUP_DEBUG=y. 223 This is the amount of time a cgroup spends without any pending 224 requests when not being served, i.e., it does not include any time 225 spent idling for one of the queues of the cgroup. This is in 226 nanoseconds. If this is read when the cgroup is in an empty state, 227 the stat will only report the empty_time accumulated till the last 228 time it had a pending request and will not include the current delta. 229 230- blkio.idle_time 231 - Debugging aid only enabled if CONFIG_BFQ_CGROUP_DEBUG=y. 232 This is the amount of time spent by the IO scheduler idling for a 233 given cgroup in anticipation of a better request than the existing ones 234 from other queues/cgroups. This is in nanoseconds. If this is read 235 when the cgroup is in an idling state, the stat will only report the 236 idle_time accumulated till the last idle period and will not include 237 the current delta. 238 239- blkio.dequeue 240 - Debugging aid only enabled if CONFIG_BFQ_CGROUP_DEBUG=y. This 241 gives the statistics about how many a times a group was dequeued 242 from service tree of the device. First two fields specify the major 243 and minor number of the device and third field specifies the number 244 of times a group was dequeued from a particular device. 245 246- blkio.*_recursive 247 - Recursive version of various stats. These files show the 248 same information as their non-recursive counterparts but 249 include stats from all the descendant cgroups. 250 251Throttling/Upper limit policy files 252----------------------------------- 253- blkio.throttle.read_bps_device 254 - Specifies upper limit on READ rate from the device. IO rate is 255 specified in bytes per second. Rules are per device. Following is 256 the format:: 257 258 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.throttle.read_bps_device 259 260- blkio.throttle.write_bps_device 261 - Specifies upper limit on WRITE rate to the device. IO rate is 262 specified in bytes per second. Rules are per device. Following is 263 the format:: 264 265 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.throttle.write_bps_device 266 267- blkio.throttle.read_iops_device 268 - Specifies upper limit on READ rate from the device. IO rate is 269 specified in IO per second. Rules are per device. Following is 270 the format:: 271 272 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.throttle.read_iops_device 273 274- blkio.throttle.write_iops_device 275 - Specifies upper limit on WRITE rate to the device. IO rate is 276 specified in io per second. Rules are per device. Following is 277 the format:: 278 279 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.throttle.write_iops_device 280 281Note: If both BW and IOPS rules are specified for a device, then IO is 282 subjected to both the constraints. 283 284- blkio.throttle.io_serviced 285 - Number of IOs (bio) issued to the disk by the group. These 286 are further divided by the type of operation - read or write, sync 287 or async. First two fields specify the major and minor number of the 288 device, third field specifies the operation type and the fourth field 289 specifies the number of IOs. 290 291- blkio.throttle.io_service_bytes 292 - Number of bytes transferred to/from the disk by the group. These 293 are further divided by the type of operation - read or write, sync 294 or async. First two fields specify the major and minor number of the 295 device, third field specifies the operation type and the fourth field 296 specifies the number of bytes. 297 298Common files among various policies 299----------------------------------- 300- blkio.reset_stats 301 - Writing an int to this file will result in resetting all the stats 302 for that cgroup. 303