1 CFQ (Complete Fairness Queueing)
2 ===============================
4 The main aim of CFQ scheduler is to provide a fair allocation of the disk
5 I/O bandwidth for all the processes which requests an I/O operation.
7 CFQ maintains the per process queue for the processes which request I/O
8 operation(syncronous requests). In case of asynchronous requests, all the
9 requests from all the processes are batched together according to their
10 process's I/O priority.
12 CFQ ioscheduler tunables
13 ========================
17 This specifies how long CFQ should idle for next request on certain cfq queues
18 (for sequential workloads) and service trees (for random workloads) before
19 queue is expired and CFQ selects next queue to dispatch from.
21 By default slice_idle is a non-zero value. That means by default we idle on
22 queues/service trees. This can be very helpful on highly seeky media like
23 single spindle SATA/SAS disks where we can cut down on overall number of
24 seeks and see improved throughput.
26 Setting slice_idle to 0 will remove all the idling on queues/service tree
27 level and one should see an overall improved throughput on faster storage
28 devices like multiple SATA/SAS disks in hardware RAID configuration. The down
29 side is that isolation provided from WRITES also goes down and notion of
30 IO priority becomes weaker.
32 So depending on storage and workload, it might be useful to set slice_idle=0.
33 In general I think for SATA/SAS disks and software RAID of SATA/SAS disks
34 keeping slice_idle enabled should be useful. For any configurations where
35 there are multiple spindles behind single LUN (Host based hardware RAID
36 controller or for storage arrays), setting slice_idle=0 might end up in better
37 throughput and acceptable latencies.
41 This specifies, given in Kbytes, the maximum "distance" for backward seeking.
42 The distance is the amount of space from the current head location to the
43 sectors that are backward in terms of distance.
45 This parameter allows the scheduler to anticipate requests in the "backward"
46 direction and consider them as being the "next" if they are within this
47 distance from the current head location.
51 This parameter is used to compute the cost of backward seeking. If the
52 backward distance of request is just 1/back_seek_penalty from a "front"
53 request, then the seeking cost of two requests is considered equivalent.
55 So scheduler will not bias toward one or the other request (otherwise scheduler
56 will bias toward front request). Default value of back_seek_penalty is 2.
60 This parameter is used to set the timeout of asynchronous requests. Default
61 value of this is 248ms.
65 This parameter is used to set the timeout of synchronous requests. Default
66 value of this is 124ms. In case to favor synchronous requests over asynchronous
67 one, this value should be decreased relative to fifo_expire_async.
71 This parameter is same as of slice_sync but for asynchronous queue. The
72 default value is 40ms.
76 This parameter is used to limit the dispatching of asynchronous request to
77 device request queue in queue's slice time. The maximum number of request that
78 are allowed to be dispatched also depends upon the io priority. Default value
83 When a queue is selected for execution, the queues IO requests are only
84 executed for a certain amount of time(time_slice) before switching to another
85 queue. This parameter is used to calculate the time slice of synchronous
88 time_slice is computed using the below equation:-
89 time_slice = slice_sync + (slice_sync/5 * (4 - prio)). To increase the
90 time_slice of synchronous queue, increase the value of slice_sync. Default
95 This specifies the number of request dispatched to the device queue. In a
96 queue's time slice, a request will not be dispatched if the number of request
97 in the device exceeds this parameter. This parameter is used for synchronous
100 In case of storage with several disk, this setting can limit the parallel
101 processing of request. Therefore, increasing the value can imporve the
102 performace although this can cause the latency of some I/O to increase due
103 to more number of requests.
105 CFQ IOPS Mode for group scheduling
106 ===================================
107 Basic CFQ design is to provide priority based time slices. Higher priority
108 process gets bigger time slice and lower priority process gets smaller time
109 slice. Measuring time becomes harder if storage is fast and supports NCQ and
110 it would be better to dispatch multiple requests from multiple cfq queues in
111 request queue at a time. In such scenario, it is not possible to measure time
112 consumed by single queue accurately.
114 What is possible though is to measure number of requests dispatched from a
115 single queue and also allow dispatch from multiple cfq queue at the same time.
116 This effectively becomes the fairness in terms of IOPS (IO operations per
119 If one sets slice_idle=0 and if storage supports NCQ, CFQ internally switches
120 to IOPS mode and starts providing fairness in terms of number of requests
121 dispatched. Note that this mode switching takes effect only for group
122 scheduling. For non-cgroup users nothing should change.
124 CFQ IO scheduler Idling Theory
125 ===============================
126 Idling on a queue is primarily about waiting for the next request to come
127 on same queue after completion of a request. In this process CFQ will not
128 dispatch requests from other cfq queues even if requests are pending there.
130 The rationale behind idling is that it can cut down on number of seeks
131 on rotational media. For example, if a process is doing dependent
132 sequential reads (next read will come on only after completion of previous
133 one), then not dispatching request from other queue should help as we
134 did not move the disk head and kept on dispatching sequential IO from
137 CFQ has following service trees and various queues are put on these trees.
139 sync-idle sync-noidle async
141 All cfq queues doing synchronous sequential IO go on to sync-idle tree.
142 On this tree we idle on each queue individually.
144 All synchronous non-sequential queues go on sync-noidle tree. Also any
145 request which are marked with REQ_NOIDLE go on this service tree. On this
146 tree we do not idle on individual queues instead idle on the whole group
147 of queues or the tree. So if there are 4 queues waiting for IO to dispatch
148 we will idle only once last queue has dispatched the IO and there is
149 no more IO on this service tree.
151 All async writes go on async service tree. There is no idling on async
154 CFQ has some optimizations for SSDs and if it detects a non-rotational
155 media which can support higher queue depth (multiple requests at in
156 flight at a time), then it cuts down on idling of individual queues and
157 all the queues move to sync-noidle tree and only tree idle remains. This
158 tree idling provides isolation with buffered write queues on async tree.
162 Q1. Why to idle at all on queues marked with REQ_NOIDLE.
164 A1. We only do tree idle (all queues on sync-noidle tree) on queues marked
165 with REQ_NOIDLE. This helps in providing isolation with all the sync-idle
166 queues. Otherwise in presence of many sequential readers, other
167 synchronous IO might not get fair share of disk.
169 For example, if there are 10 sequential readers doing IO and they get
170 100ms each. If a REQ_NOIDLE request comes in, it will be scheduled
171 roughly after 1 second. If after completion of REQ_NOIDLE request we
172 do not idle, and after a couple of milli seconds a another REQ_NOIDLE
173 request comes in, again it will be scheduled after 1second. Repeat it
174 and notice how a workload can lose its disk share and suffer due to
175 multiple sequential readers.
177 fsync can generate dependent IO where bunch of data is written in the
178 context of fsync, and later some journaling data is written. Journaling
179 data comes in only after fsync has finished its IO (atleast for ext4
180 that seemed to be the case). Now if one decides not to idle on fsync
181 thread due to REQ_NOIDLE, then next journaling write will not get
182 scheduled for another second. A process doing small fsync, will suffer
183 badly in presence of multiple sequential readers.
185 Hence doing tree idling on threads using REQ_NOIDLE flag on requests
186 provides isolation from multiple sequential readers and at the same
187 time we do not idle on individual threads.
189 Q2. When to specify REQ_NOIDLE
190 A2. I would think whenever one is doing synchronous write and not expecting
191 more writes to be dispatched from same context soon, should be able
192 to specify REQ_NOIDLE on writes and that probably should work well for