TEXT   129

blkio-controller.txt

Guest on 6th September 2021 09:44:14 AM

  1.                                 Block IO Controller
  2.                                 ===================
  3. Overview
  4. ========
  5. cgroup subsys "blkio" implements the block io controller. There seems to be
  6. a need of various kinds of IO control policies (like proportional BW, max BW)
  7. both at leaf nodes as well as at intermediate nodes in a storage hierarchy.
  8. Plan is to use the same cgroup based management interface for blkio controller
  9. and based on user options switch IO policies in the background.
  10.  
  11. Currently two IO control policies are implemented. First one is proportional
  12. weight time based division of disk policy. It is implemented in CFQ. Hence
  13. this policy takes effect only on leaf nodes when CFQ is being used. The second
  14. one is throttling policy which can be used to specify upper IO rate limits
  15. on devices. This policy is implemented in generic block layer and can be
  16. used on leaf nodes as well as higher level logical devices like device mapper.
  17.  
  18. HOWTO
  19. =====
  20. Proportional Weight division of bandwidth
  21. -----------------------------------------
  22. You can do a very simple testing of running two dd threads in two different
  23. cgroups. 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.  
  66. Throttling/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>  <bytes_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.  
  95. Hierarchical Cgroups
  96. ====================
  97.  
  98. Both CFQ and throttling implement hierarchy support; however,
  99. throttling's hierarchy support is enabled iff "sane_behavior" is
  100. enabled from cgroup side, which currently is a development option and
  101. not publicly available.
  102.  
  103. If somebody created a hierarchy like as follows.
  104.  
  105.                         root
  106.                         /  \
  107.                      test1 test2
  108.                         |
  109.                      test3
  110.  
  111. CFQ by default and throttling with "sane_behavior" will handle the
  112. hierarchy correctly.  For details on CFQ hierarchy support, refer to
  113. Documentation/block/cfq-iosched.txt.  For throttling, all limits apply
  114. to the whole subtree while all statistics are local to the IOs
  115. directly generated by tasks in that cgroup.
  116.  
  117. Throttling without "sane_behavior" enabled from cgroup side will
  118. practically treat all groups at same level as if it looks like the
  119. following.
  120.  
  121.                                 pivot
  122.                              /  /   \  \
  123.                         root  test1 test2  test3
  124.  
  125. Various user visible config options
  126. ===================================
  127. CONFIG_BLK_CGROUP
  128.         - Block IO controller.
  129.  
  130. CONFIG_DEBUG_BLK_CGROUP
  131.         - Debug help. Right now some additional stats file show up in cgroup
  132.           if this option is enabled.
  133.  
  134. CONFIG_CFQ_GROUP_IOSCHED
  135.         - Enables group scheduling in CFQ. Currently only 1 level of group
  136.           creation is allowed.
  137.  
  138. CONFIG_BLK_DEV_THROTTLING
  139.         - Enable block device throttling support in block layer.
  140.  
  141. Details of cgroup files
  142. =======================
  143. Proportional weight policy files
  144. --------------------------------
  145. - blkio.weight
  146.         - Specifies per cgroup weight. This is default weight of the group
  147.           on all the devices until and unless overridden by per device rule.
  148.           (See blkio.weight_device).
  149.           Currently allowed range of weights is from 10 to 1000.
  150.  
  151. - blkio.weight_device
  152.         - One can specify per cgroup per device rules using this interface.
  153.           These rules override the default value of group weight as specified
  154.           by blkio.weight.
  155.  
  156.           Following is the format.
  157.  
  158.           # echo dev_maj:dev_minor weight > blkio.weight_device
  159.           Configure weight=300 on /dev/sdb (8:16) in this cgroup
  160.           # echo 8:16 300 > blkio.weight_device
  161.           # cat blkio.weight_device
  162.           dev     weight
  163.           8:16    300
  164.  
  165.           Configure weight=500 on /dev/sda (8:0) in this cgroup
  166.           # echo 8:0 500 > blkio.weight_device
  167.           # cat blkio.weight_device
  168.           dev     weight
  169.           8:0     500
  170.           8:16    300
  171.  
  172.           Remove specific weight for /dev/sda in this cgroup
  173.           # echo 8:0 0 > blkio.weight_device
  174.           # cat blkio.weight_device
  175.           dev     weight
  176.           8:16    300
  177.  
  178. - blkio.leaf_weight[_device]
  179.         - Equivalents of blkio.weight[_device] for the purpose of
  180.           deciding how much weight tasks in the given cgroup has while
  181.           competing with the cgroup's child cgroups. For details,
  182.           please refer to Documentation/block/cfq-iosched.txt.
  183.  
  184. - blkio.time
  185.         - disk time allocated to cgroup per device in milliseconds. First
  186.           two fields specify the major and minor number of the device and
  187.           third field specifies the disk time allocated to group in
  188.           milliseconds.
  189.  
  190. - blkio.sectors
  191.         - number of sectors transferred to/from disk by the group. First
  192.           two fields specify the major and minor number of the device and
  193.           third field specifies the number of sectors transferred by the
  194.           group to/from the device.
  195.  
  196. - blkio.io_service_bytes
  197.         - Number of bytes transferred to/from the disk by the group. These
  198.           are further divided by the type of operation - read or write, sync
  199.           or async. First two fields specify the major and minor number of the
  200.           device, third field specifies the operation type and the fourth field
  201.           specifies the number of bytes.
  202.  
  203. - blkio.io_serviced
  204.         - Number of IOs completed to/from the disk by the group. These
  205.           are further divided by the type of operation - read or write, sync
  206.           or async. First two fields specify the major and minor number of the
  207.           device, third field specifies the operation type and the fourth field
  208.           specifies the number of IOs.
  209.  
  210. - blkio.io_service_time
  211.         - Total amount of time between request dispatch and request completion
  212.           for the IOs done by this cgroup. This is in nanoseconds to make it
  213.           meaningful for flash devices too. For devices with queue depth of 1,
  214.           this time represents the actual service time. When queue_depth > 1,
  215.           that is no longer true as requests may be served out of order. This
  216.           may cause the service time for a given IO to include the service time
  217.           of multiple IOs when served out of order which may result in total
  218.           io_service_time > actual time elapsed. This time is further divided by
  219.           the type of operation - read or write, sync or async. First two fields
  220.           specify the major and minor number of the device, third field
  221.           specifies the operation type and the fourth field specifies the
  222.           io_service_time in ns.
  223.  
  224. - blkio.io_wait_time
  225.         - Total amount of time the IOs for this cgroup spent waiting in the
  226.           scheduler queues for service. This can be greater than the total time
  227.           elapsed since it is cumulative io_wait_time for all IOs. It is not a
  228.           measure of total time the cgroup spent waiting but rather a measure of
  229.           the wait_time for its individual IOs. For devices with queue_depth > 1
  230.           this metric does not include the time spent waiting for service once
  231.           the IO is dispatched to the device but till it actually gets serviced
  232.           (there might be a time lag here due to re-ordering of requests by the
  233.           device). This is in nanoseconds to make it meaningful for flash
  234.           devices too. This time is further divided by the type of operation -
  235.           read or write, sync or async. First two fields specify the major and
  236.           minor number of the device, third field specifies the operation type
  237.           and the fourth field specifies the io_wait_time in ns.
  238.  
  239. - blkio.io_merged
  240.         - Total number of bios/requests merged into requests belonging to this
  241.           cgroup. This is further divided by the type of operation - read or
  242.           write, sync or async.
  243.  
  244. - blkio.io_queued
  245.         - Total number of requests queued up at any given instant for this
  246.           cgroup. This is further divided by the type of operation - read or
  247.           write, sync or async.
  248.  
  249. - blkio.avg_queue_size
  250.         - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
  251.           The average queue size for this cgroup over the entire time of this
  252.           cgroup's existence. Queue size samples are taken each time one of the
  253.           queues of this cgroup gets a timeslice.
  254.  
  255. - blkio.group_wait_time
  256.         - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
  257.           This is the amount of time the cgroup had to wait since it became busy
  258.           (i.e., went from 0 to 1 request queued) to get a timeslice for one of
  259.           its queues. This is different from the io_wait_time which is the
  260.           cumulative total of the amount of time spent by each IO in that cgroup
  261.           waiting in the scheduler queue. This is in nanoseconds. If this is
  262.           read when the cgroup is in a waiting (for timeslice) state, the stat
  263.           will only report the group_wait_time accumulated till the last time it
  264.           got a timeslice and will not include the current delta.
  265.  
  266. - blkio.empty_time
  267.         - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
  268.           This is the amount of time a cgroup spends without any pending
  269.           requests when not being served, i.e., it does not include any time
  270.           spent idling for one of the queues of the cgroup. This is in
  271.           nanoseconds. If this is read when the cgroup is in an empty state,
  272.           the stat will only report the empty_time accumulated till the last
  273.           time it had a pending request and will not include the current delta.
  274.  
  275. - blkio.idle_time
  276.         - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y.
  277.           This is the amount of time spent by the IO scheduler idling for a
  278.           given cgroup in anticipation of a better request than the existing ones
  279.           from other queues/cgroups. This is in nanoseconds. If this is read
  280.           when the cgroup is in an idling state, the stat will only report the
  281.           idle_time accumulated till the last idle period and will not include
  282.           the current delta.
  283.  
  284. - blkio.dequeue
  285.         - Debugging aid only enabled if CONFIG_DEBUG_BLK_CGROUP=y. This
  286.           gives the statistics about how many a times a group was dequeued
  287.           from service tree of the device. First two fields specify the major
  288.           and minor number of the device and third field specifies the number
  289.           of times a group was dequeued from a particular device.
  290.  
  291. - blkio.*_recursive
  292.         - Recursive version of various stats. These files show the
  293.           same information as their non-recursive counterparts but
  294.           include stats from all the descendant cgroups.
  295.  
  296. Throttling/Upper limit policy files
  297. -----------------------------------
  298. - blkio.throttle.read_bps_device
  299.         - Specifies upper limit on READ rate from the device. IO rate is
  300.           specified in bytes per second. Rules are per device. Following is
  301.           the format.
  302.  
  303.   echo "<major>:<minor>  <rate_bytes_per_second>" > /cgrp/blkio.throttle.read_bps_device
  304.  
  305. - blkio.throttle.write_bps_device
  306.         - Specifies upper limit on WRITE rate to the device. IO rate is
  307.           specified in bytes per second. Rules are per device. Following is
  308.           the format.
  309.  
  310.   echo "<major>:<minor>  <rate_bytes_per_second>" > /cgrp/blkio.throttle.write_bps_device
  311.  
  312. - blkio.throttle.read_iops_device
  313.         - Specifies upper limit on READ rate from the device. IO rate is
  314.           specified in IO per second. Rules are per device. Following is
  315.           the format.
  316.  
  317.   echo "<major>:<minor>  <rate_io_per_second>" > /cgrp/blkio.throttle.read_iops_device
  318.  
  319. - blkio.throttle.write_iops_device
  320.         - Specifies upper limit on WRITE rate to the device. IO rate is
  321.           specified in io per second. Rules are per device. Following is
  322.           the format.
  323.  
  324.   echo "<major>:<minor>  <rate_io_per_second>" > /cgrp/blkio.throttle.write_iops_device
  325.  
  326. Note: If both BW and IOPS rules are specified for a device, then IO is
  327.       subjected to both the constraints.
  328.  
  329. - blkio.throttle.io_serviced
  330.         - Number of IOs (bio) completed to/from the disk by the group (as
  331.           seen by throttling policy). These are further divided by the type
  332.           of operation - read or write, sync or async. First two fields specify
  333.           the major and minor number of the device, third field specifies the
  334.           operation type and the fourth field specifies the number of IOs.
  335.  
  336.           blkio.io_serviced does accounting as seen by CFQ and counts are in
  337.           number of requests (struct request). On the other hand,
  338.           blkio.throttle.io_serviced counts number of IO in terms of number
  339.           of bios as seen by throttling policy.  These bios can later be
  340.           merged by elevator and total number of requests completed can be
  341.           lesser.
  342.  
  343. - blkio.throttle.io_service_bytes
  344.         - Number of bytes transferred to/from the disk by the group. These
  345.           are further divided by the type of operation - read or write, sync
  346.           or async. First two fields specify the major and minor number of the
  347.           device, third field specifies the operation type and the fourth field
  348.           specifies the number of bytes.
  349.  
  350.           These numbers should roughly be same as blkio.io_service_bytes as
  351.           updated by CFQ. The difference between two is that
  352.           blkio.io_service_bytes will not be updated if CFQ is not operating
  353.           on request queue.
  354.  
  355. Common files among various policies
  356. -----------------------------------
  357. - blkio.reset_stats
  358.         - Writing an int to this file will result in resetting all the stats
  359.           for that cgroup.
  360.  
  361. CFQ sysfs tunable
  362. =================
  363. /sys/block/<disk>/queue/iosched/slice_idle
  364. ------------------------------------------
  365. On a faster hardware CFQ can be slow, especially with sequential workload.
  366. This happens because CFQ idles on a single queue and single queue might not
  367. drive deeper request queue depths to keep the storage busy. In such scenarios
  368. one can try setting slice_idle=0 and that would switch CFQ to IOPS
  369. (IO operations per second) mode on NCQ supporting hardware.
  370.  
  371. That means CFQ will not idle between cfq queues of a cfq group and hence be
  372. able to driver higher queue depth and achieve better throughput. That also
  373. means that cfq provides fairness among groups in terms of IOPS and not in
  374. terms of disk time.
  375.  
  376. /sys/block/<disk>/queue/iosched/group_idle
  377. ------------------------------------------
  378. If one disables idling on individual cfq queues and cfq service trees by
  379. setting slice_idle=0, group_idle kicks in. That means CFQ will still idle
  380. on the group in an attempt to provide fairness among groups.
  381.  
  382. By default group_idle is same as slice_idle and does not do anything if
  383. slice_idle is enabled.
  384.  
  385. One can experience an overall throughput drop if you have created multiple
  386. groups and put applications in that group which are not driving enough
  387. IO to keep disk busy. In that case set group_idle=0, and CFQ will not idle
  388. on individual groups and throughput should improve.
  389.  
  390. What works
  391. ==========
  392. - Currently only sync IO queues are support. All the buffered writes are
  393.   still system wide and not per group. Hence we will not see service
  394.   differentiation between buffered writes between groups.

Raw Paste


Login or Register to edit or fork this paste. It's free.