1 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 11Currently 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. 17 18HOWTO 19===== 20Proportional Weight division of bandwidth 21----------------------------------------- 22You can do a very simple testing of running two dd threads in two different 23cgroups. Here is what you can do. 24 25- Enable Block IO controller 26 CONFIG_BLK_CGROUP=y 27 28- Enable group scheduling in CFQ 29 CONFIG_CFQ_GROUP_IOSCHED=y 30 31- Compile and boot into kernel and mount IO controller (blkio); see 32 cgroups.txt, Why are cgroups needed?. 33 34 mount -t tmpfs cgroup_root /sys/fs/cgroup 35 mkdir /sys/fs/cgroup/blkio 36 mount -t cgroup -o blkio none /sys/fs/cgroup/blkio 37 38- Create two cgroups 39 mkdir -p /sys/fs/cgroup/blkio/test1/ /sys/fs/cgroup/blkio/test2 40 41- Set weights of group test1 and test2 42 echo 1000 > /sys/fs/cgroup/blkio/test1/blkio.weight 43 echo 500 > /sys/fs/cgroup/blkio/test2/blkio.weight 44 45- Create two same size files (say 512MB each) on same disk (file1, file2) and 46 launch two dd threads in different cgroup to read those files. 47 48 sync 49 echo 3 > /proc/sys/vm/drop_caches 50 51 dd if=/mnt/sdb/zerofile1 of=/dev/null & 52 echo $! > /sys/fs/cgroup/blkio/test1/tasks 53 cat /sys/fs/cgroup/blkio/test1/tasks 54 55 dd if=/mnt/sdb/zerofile2 of=/dev/null & 56 echo $! > /sys/fs/cgroup/blkio/test2/tasks 57 cat /sys/fs/cgroup/blkio/test2/tasks 58 59- At macro level, first dd should finish first. To get more precise data, keep 60 on looking at (with the help of script), at blkio.disk_time and 61 blkio.disk_sectors files of both test1 and test2 groups. This will tell how 62 much disk time (in milli seconds), each group got and how many secotors each 63 group dispatched to the disk. We provide fairness in terms of disk time, so 64 ideally io.disk_time of cgroups should be in proportion to the weight. 65 66Throttling/Upper Limit policy 67----------------------------- 68- Enable Block IO controller 69 CONFIG_BLK_CGROUP=y 70 71- Enable throttling in block layer 72 CONFIG_BLK_DEV_THROTTLING=y 73 74- Mount blkio controller (see cgroups.txt, Why are cgroups needed?) 75 mount -t cgroup -o blkio none /sys/fs/cgroup/blkio 76 77- Specify a bandwidth rate on particular device for root group. The format 78 for policy is "<major>:<minor> <byes_per_second>". 79 80 echo "8:16 1048576" > /sys/fs/cgroup/blkio/blkio.throttle.read_bps_device 81 82 Above will put a limit of 1MB/second on reads happening for root group 83 on device having major/minor number 8:16. 84 85- Run dd to read a file and see if rate is throttled to 1MB/s or not. 86 87 # dd if=/mnt/common/zerofile of=/dev/null bs=4K count=1024 88 # iflag=direct 89 1024+0 records in 90 1024+0 records out 91 4194304 bytes (4.2 MB) copied, 4.0001 s, 1.0 MB/s 92 93 Limits for writes can be put using blkio.throttle.write_bps_device file. 94 95Hierarchical Cgroups 96==================== 97- Currently none of the IO control policy supports hierarchical groups. But 98 cgroup interface does allow creation of hierarchical cgroups and internally 99 IO policies treat them as flat hierarchy. 100 101 So this patch will allow creation of cgroup hierarchcy but at the backend 102 everything will be treated as flat. So if somebody created a hierarchy like 103 as follows. 104 105 root 106 / \ 107 test1 test2 108 | 109 test3 110 111 CFQ and throttling will practically treat all groups at same level. 112 113 pivot 114 / / \ \ 115 root test1 test2 test3 116 117 Down the line we can implement hierarchical accounting/control support 118 and also introduce a new cgroup file "use_hierarchy" which will control 119 whether cgroup hierarchy is viewed as flat or hierarchical by the policy.. 120 This is how memory controller also has implemented the things. 121 122Various user visible config options 123=================================== 124CONFIG_BLK_CGROUP 125 - Block IO controller. 126 127CONFIG_DEBUG_BLK_CGROUP 128 - Debug help. Right now some additional stats file show up in cgroup 129 if this option is enabled. 130 131CONFIG_CFQ_GROUP_IOSCHED 132 - Enables group scheduling in CFQ. Currently only 1 level of group 133 creation is allowed. 134 135CONFIG_BLK_DEV_THROTTLING 136 - Enable block device throttling support in block layer. 137 138Details of cgroup files 139======================= 140Proportional weight policy files 141-------------------------------- 142- blkio.weight 143 - Specifies per cgroup weight. This is default weight of the group 144 on all the devices until and unless overridden by per device rule. 145 (See blkio.weight_device). 146 Currently allowed range of weights is from 10 to 1000. 147 148- blkio.weight_device 149 - One can specify per cgroup per device rules using this interface. 150 These rules override the default value of group weight as specified 151 by blkio.weight. 152 153 Following is the format. 154 155 # echo dev_maj:dev_minor weight > blkio.weight_device 156 Configure weight=300 on /dev/sdb (8:16) in this cgroup 157 # echo 8:16 300 > blkio.weight_device 158 # cat blkio.weight_device 159 dev weight 160 8:16 300 161 162 Configure weight=500 on /dev/sda (8:0) in this cgroup 163 # echo 8:0 500 > blkio.weight_device 164 # cat blkio.weight_device 165 dev weight 166 8:0 500 167 8:16 300 168 169 Remove specific weight for /dev/sda in this cgroup 170 # echo 8:0 0 > blkio.weight_device 171 # cat blkio.weight_device 172 dev weight 173 8:16 300 174 175- blkio.time 176 - disk time allocated to cgroup per device in milliseconds. First 177 two fields specify the major and minor number of the device and 178 third field specifies the disk time allocated to group in 179 milliseconds. 180 181- blkio.sectors 182 - number of sectors transferred to/from disk by the group. First 183 two fields specify the major and minor number of the device and 184 third field specifies the number of sectors transferred by the 185 group to/from the device. 186 187- blkio.io_service_bytes 188 - Number of bytes transferred to/from the disk by the group. These 189 are further divided by the type of operation - read or write, sync 190 or async. First two fields specify the major and minor number of the 191 device, third field specifies the operation type and the fourth field 192 specifies the number of bytes. 193 194- blkio.io_serviced 195 - Number of IOs completed to/from the disk by the group. These 196 are further divided by the type of operation - read or write, sync 197 or async. First two fields specify the major and minor number of the 198 device, third field specifies the operation type and the fourth field 199 specifies the number of IOs. 200 201- blkio.io_service_time 202 - Total amount of time between request dispatch and request completion 203 for the IOs done by this cgroup. This is in nanoseconds to make it 204 meaningful for flash devices too. For devices with queue depth of 1, 205 this time represents the actual service time. When queue_depth > 1, 206 that is no longer true as requests may be served out of order. This 207 may cause the service time for a given IO to include the service time 208 of multiple IOs when served out of order which may result in total 209 io_service_time > actual time elapsed. This time is further divided by 210 the type of operation - read or write, sync or async. First two fields 211 specify the major and minor number of the device, third field 212 specifies the operation type and the fourth field specifies the 213 io_service_time in ns. 214 215- blkio.io_wait_time 216 - Total amount of time the IOs for this cgroup spent waiting in the 217 scheduler queues for service. This can be greater than the total time 218 elapsed since it is cumulative io_wait_time for all IOs. It is not a 219 measure of total time the cgroup spent waiting but rather a measure of 220 the wait_time for its individual IOs. For devices with queue_depth > 1 221 this metric does not include the time spent waiting for service once 222 the IO is dispatched to the device but till it actually gets serviced 223 (there might be a time lag here due to re-ordering of requests by the 224 device). This is in nanoseconds to make it meaningful for flash 225 devices too. This time is further divided by the type of operation - 226 read or write, sync or async. First two fields specify the major and 227 minor number of the device, third field specifies the operation type 228 and the fourth field specifies the io_wait_time in ns. 229 230- blkio.io_merged 231 - Total number of bios/requests merged into requests belonging to this 232 cgroup. This is further divided by the type of operation - read or 233 write, sync or async. 234 235- blkio.io_queued 236 - Total number of requests queued up at any given instant for this 237 cgroup. This is further divided by the type of operation - read or 238 write, sync or async. 239 240- blkio.avg_queue_size 241 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. 242 The average queue size for this cgroup over the entire time of this 243 cgroup's existence. Queue size samples are taken each time one of the 244 queues of this cgroup gets a timeslice. 245 246- blkio.group_wait_time 247 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. 248 This is the amount of time the cgroup had to wait since it became busy 249 (i.e., went from 0 to 1 request queued) to get a timeslice for one of 250 its queues. This is different from the io_wait_time which is the 251 cumulative total of the amount of time spent by each IO in that cgroup 252 waiting in the scheduler queue. This is in nanoseconds. If this is 253 read when the cgroup is in a waiting (for timeslice) state, the stat 254 will only report the group_wait_time accumulated till the last time it 255 got a timeslice and will not include the current delta. 256 257- blkio.empty_time 258 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. 259 This is the amount of time a cgroup spends without any pending 260 requests when not being served, i.e., it does not include any time 261 spent idling for one of the queues of the cgroup. This is in 262 nanoseconds. If this is read when the cgroup is in an empty state, 263 the stat will only report the empty_time accumulated till the last 264 time it had a pending request and will not include the current delta. 265 266- blkio.idle_time 267 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. 268 This is the amount of time spent by the IO scheduler idling for a 269 given cgroup in anticipation of a better request than the existing ones 270 from other queues/cgroups. This is in nanoseconds. If this is read 271 when the cgroup is in an idling state, the stat will only report the 272 idle_time accumulated till the last idle period and will not include 273 the current delta. 274 275- blkio.dequeue 276 - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. This 277 gives the statistics about how many a times a group was dequeued 278 from service tree of the device. First two fields specify the major 279 and minor number of the device and third field specifies the number 280 of times a group was dequeued from a particular device. 281 282Throttling/Upper limit policy files 283----------------------------------- 284- blkio.throttle.read_bps_device 285 - Specifies upper limit on READ rate from the device. IO rate is 286 specified in bytes per second. Rules are per device. Following is 287 the format. 288 289 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.throttle.read_bps_device 290 291- blkio.throttle.write_bps_device 292 - Specifies upper limit on WRITE rate to the device. IO rate is 293 specified in bytes per second. Rules are per device. Following is 294 the format. 295 296 echo "<major>:<minor> <rate_bytes_per_second>" > /cgrp/blkio.throttle.write_bps_device 297 298- blkio.throttle.read_iops_device 299 - Specifies upper limit on READ rate from the device. IO rate is 300 specified in IO per second. Rules are per device. Following is 301 the format. 302 303 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.throttle.read_iops_device 304 305- blkio.throttle.write_iops_device 306 - Specifies upper limit on WRITE rate to the device. IO rate is 307 specified in io per second. Rules are per device. Following is 308 the format. 309 310 echo "<major>:<minor> <rate_io_per_second>" > /cgrp/blkio.throttle.write_iops_device 311 312Note: If both BW and IOPS rules are specified for a device, then IO is 313 subjected to both the constraints. 314 315- blkio.throttle.io_serviced 316 - Number of IOs (bio) completed to/from the disk by the group (as 317 seen by throttling policy). These are further divided by the type 318 of operation - read or write, sync or async. First two fields specify 319 the major and minor number of the device, third field specifies the 320 operation type and the fourth field specifies the number of IOs. 321 322 blkio.io_serviced does accounting as seen by CFQ and counts are in 323 number of requests (struct request). On the other hand, 324 blkio.throttle.io_serviced counts number of IO in terms of number 325 of bios as seen by throttling policy. These bios can later be 326 merged by elevator and total number of requests completed can be 327 lesser. 328 329- blkio.throttle.io_service_bytes 330 - Number of bytes transferred to/from the disk by the group. These 331 are further divided by the type of operation - read or write, sync 332 or async. First two fields specify the major and minor number of the 333 device, third field specifies the operation type and the fourth field 334 specifies the number of bytes. 335 336 These numbers should roughly be same as blkio.io_service_bytes as 337 updated by CFQ. The difference between two is that 338 blkio.io_service_bytes will not be updated if CFQ is not operating 339 on request queue. 340 341Common files among various policies 342----------------------------------- 343- blkio.reset_stats 344 - Writing an int to this file will result in resetting all the stats 345 for that cgroup. 346 347CFQ sysfs tunable 348================= 349/sys/block/<disk>/queue/iosched/slice_idle 350------------------------------------------ 351On a faster hardware CFQ can be slow, especially with sequential workload. 352This happens because CFQ idles on a single queue and single queue might not 353drive deeper request queue depths to keep the storage busy. In such scenarios 354one can try setting slice_idle=0 and that would switch CFQ to IOPS 355(IO operations per second) mode on NCQ supporting hardware. 356 357That means CFQ will not idle between cfq queues of a cfq group and hence be 358able to driver higher queue depth and achieve better throughput. That also 359means that cfq provides fairness among groups in terms of IOPS and not in 360terms of disk time. 361 362/sys/block/<disk>/queue/iosched/group_idle 363------------------------------------------ 364If one disables idling on individual cfq queues and cfq service trees by 365setting slice_idle=0, group_idle kicks in. That means CFQ will still idle 366on the group in an attempt to provide fairness among groups. 367 368By default group_idle is same as slice_idle and does not do anything if 369slice_idle is enabled. 370 371One can experience an overall throughput drop if you have created multiple 372groups and put applications in that group which are not driving enough 373IO to keep disk busy. In that case set group_idle=0, and CFQ will not idle 374on individual groups and throughput should improve. 375 376What works 377========== 378- Currently only sync IO queues are support. All the buffered writes are 379 still system wide and not per group. Hence we will not see service 380 differentiation between buffered writes between groups. 381