1 Anticipatory IO scheduler
2 -------------------------
3 Nick Piggin <piggin@cyberone.com.au> 13 Sep 2003
5 Attention! Database servers, especially those using "TCQ" disks should
6 investigate performance with the 'deadline' IO scheduler. Any system with high
7 disk performance requirements should do so, in fact.
9 If you see unusual performance characteristics of your disk systems, or you
10 see big performance regressions versus the deadline scheduler, please email
11 me. Database users don't bother unless you're willing to test a lot of patches
12 from me ;) its a known issue.
14 Also, users with hardware RAID controllers, doing striping, may find
15 highly variable performance results with using the as-iosched. The
16 as-iosched anticipatory implementation is based on the notion that a disk
17 device has only one physical seeking head. A striped RAID controller
18 actually has a head for each physical device in the logical RAID device.
20 However, setting the antic_expire (see tunable parameters below) produces
21 very similar behavior to the deadline IO scheduler.
24 Selecting IO schedulers
25 -----------------------
26 To choose IO schedulers at boot time, use the argument 'elevator=deadline'.
27 'noop' and 'as' (the default) are also available. IO schedulers are assigned
28 globally at boot time only presently.
31 Anticipatory IO scheduler Policies
32 ----------------------------------
33 The as-iosched implementation implements several layers of policies
34 to determine when an IO request is dispatched to the disk controller.
35 Here are the policies outlined, in order of application.
37 1. one-way Elevator algorithm.
39 The elevator algorithm is similar to that used in deadline scheduler, with
40 the addition that it allows limited backward movement of the elevator
41 (i.e. seeks backwards). A seek backwards can occur when choosing between
42 two IO requests where one is behind the elevator's current position, and
43 the other is in front of the elevator's position. If the seek distance to
44 the request in back of the elevator is less than half the seek distance to
45 the request in front of the elevator, then the request in back can be chosen.
46 Backward seeks are also limited to a maximum of MAXBACK (1024*1024) sectors.
47 This favors forward movement of the elevator, while allowing opportunistic
48 "short" backward seeks.
50 2. FIFO expiration times for reads and for writes.
52 This is again very similar to the deadline IO scheduler. The expiration
53 times for requests on these lists is tunable using the parameters read_expire
54 and write_expire discussed below. When a read or a write expires in this way,
55 the IO scheduler will interrupt its current elevator sweep or read anticipation
56 to service the expired request.
58 3. Read and write request batching
60 A batch is a collection of read requests or a collection of write
61 requests. The as scheduler alternates dispatching read and write batches
62 to the driver. In the case a read batch, the scheduler submits read
63 requests to the driver as long as there are read requests to submit, and
64 the read batch time limit has not been exceeded (read_batch_expire).
65 The read batch time limit begins counting down only when there are
66 competing write requests pending.
68 In the case of a write batch, the scheduler submits write requests to
69 the driver as long as there are write requests available, and the
70 write batch time limit has not been exceeded (write_batch_expire).
71 However, the length of write batches will be gradually shortened
72 when read batches frequently exceed their time limit.
74 When changing between batch types, the scheduler waits for all requests
75 from the previous batch to complete before scheduling requests for the
78 The read and write fifo expiration times described in policy 2 above
79 are checked only when in scheduling IO of a batch for the corresponding
80 (read/write) type. So for example, the read FIFO timeout values are
81 tested only during read batches. Likewise, the write FIFO timeout
82 values are tested only during write batches. For this reason,
83 it is generally not recommended for the read batch time
84 to be longer than the write expiration time, nor for the write batch
85 time to exceed the read expiration time (see tunable parameters below).
87 When the IO scheduler changes from a read to a write batch,
88 it begins the elevator from the request that is on the head of the
89 write expiration FIFO. Likewise, when changing from a write batch to
90 a read batch, scheduler begins the elevator from the first entry
91 on the read expiration FIFO.
95 Read anticipation occurs only when scheduling a read batch.
96 This implementation of read anticipation allows only one read request
97 to be dispatched to the disk controller at a time. In
98 contrast, many write requests may be dispatched to the disk controller
99 at a time during a write batch. It is this characteristic that can make
100 the anticipatory scheduler perform anomalously with controllers supporting
101 TCQ, or with hardware striped RAID devices. Setting the antic_expire
102 queue parameter (see below) to zero disables this behavior, and the
103 anticipatory scheduler behaves essentially like the deadline scheduler.
105 When read anticipation is enabled (antic_expire is not zero), reads
106 are dispatched to the disk controller one at a time.
107 At the end of each read request, the IO scheduler examines its next
108 candidate read request from its sorted read list. If that next request
109 is from the same process as the request that just completed,
110 or if the next request in the queue is "very close" to the
111 just completed request, it is dispatched immediately. Otherwise,
112 statistics (average think time, average seek distance) on the process
113 that submitted the just completed request are examined. If it seems
114 likely that that process will submit another request soon, and that
115 request is likely to be near the just completed request, then the IO
116 scheduler will stop dispatching more read requests for up time (antic_expire)
117 milliseconds, hoping that process will submit a new request near the one
118 that just completed. If such a request is made, then it is dispatched
119 immediately. If the antic_expire wait time expires, then the IO scheduler
120 will dispatch the next read request from the sorted read queue.
122 To decide whether an anticipatory wait is worthwhile, the scheduler
123 maintains statistics for each process that can be used to compute
124 mean "think time" (the time between read requests), and mean seek
125 distance for that process. One observation is that these statistics
126 are associated with each process, but those statistics are not associated
127 with a specific IO device. So for example, if a process is doing IO
128 on several file systems on separate devices, the statistics will be
129 a combination of IO behavior from all those devices.
132 Tuning the anticipatory IO scheduler
133 ------------------------------------
134 When using 'as', the anticipatory IO scheduler there are 5 parameters under
135 /sys/block/*/queue/iosched/. All are units of milliseconds.
139 Controls how long until a read request becomes "expired". It also controls the
140 interval between which expired requests are served, so set to 50, a request
141 might take anywhere < 100ms to be serviced _if_ it is the next on the
142 expired list. Obviously request expiration strategies won't make the disk
143 go faster. The result basically equates to the timeslice a single reader
144 gets in the presence of other IO. 100*((seek time / read_expire) + 1) is
145 very roughly the % streaming read efficiency your disk should get with
149 Controls how much time a batch of reads is given before pending writes are
150 served. A higher value is more efficient. This might be set below read_expire
151 if writes are to be given higher priority than reads, but reads are to be
152 as efficient as possible when there are no writes. Generally though, it
153 should be some multiple of read_expire.
156 * write_batch_expire are equivalent to the above, for writes.
159 Controls the maximum amount of time we can anticipate a good read (one
160 with a short seek distance from the most recently completed request) before
161 giving up. Many other factors may cause anticipation to be stopped early,
162 or some processes will not be "anticipated" at all. Should be a bit higher
163 for big seek time devices though not a linear correspondence - most
164 processes have only a few ms thinktime.