aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-12-30 15:43:21 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2009-12-30 15:43:21 -0500
commite48b7b66a6531f02f1264c7196f7069a9ce9251a (patch)
treed45ce978262e3c32ce8fe460516bb9aae0cc2fb4 /Documentation
parent5ccf73bb4dc7cc9e1f761202a34de5714164724f (diff)
parent9bd3f98821a83041e77ee25158b80b535d02d7b4 (diff)
Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
* 'for-linus' of git://git.kernel.dk/linux-2.6-block: block: blk_rq_err_sectors cleanup block: Honor the gfp_mask for alloc_page() in blkdev_issue_discard() block: Fix incorrect alignment offset reporting and update documentation cfq-iosched: don't regard requests with long distance as close aoe: switch to the new bio_flush_dcache_pages() interface drivers/block/mg_disk.c: use resource_size() drivers/block/DAC960.c: use DAC960_V2_Controller block: Fix topology stacking for data and discard alignment drbd: remove unused #include <linux/version.h> drbd: remove duplicated #include drbd: Fix test of unsigned in _drbd_fault_random() drbd: Constify struct file_operations cfq-iosched: Remove prio_change logic for workload selection cfq-iosched: Get rid of nr_groups cfq-iosched: Remove the check for same cfq group from allow_merge drbd: fix test of unsigned in _drbd_fault_random() block: remove Documentation/block/as-iosched.txt
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/block/00-INDEX2
-rw-r--r--Documentation/block/as-iosched.txt172
2 files changed, 0 insertions, 174 deletions
diff --git a/Documentation/block/00-INDEX b/Documentation/block/00-INDEX
index 961a0513f8c3..a406286f6f3e 100644
--- a/Documentation/block/00-INDEX
+++ b/Documentation/block/00-INDEX
@@ -1,7 +1,5 @@
100-INDEX 100-INDEX
2 - This file 2 - This file
3as-iosched.txt
4 - Anticipatory IO scheduler
5barrier.txt 3barrier.txt
6 - I/O Barriers 4 - I/O Barriers
7biodoc.txt 5biodoc.txt
diff --git a/Documentation/block/as-iosched.txt b/Documentation/block/as-iosched.txt
deleted file mode 100644
index 738b72be128e..000000000000
--- a/Documentation/block/as-iosched.txt
+++ /dev/null
@@ -1,172 +0,0 @@
1Anticipatory IO scheduler
2-------------------------
3Nick Piggin <piggin@cyberone.com.au> 13 Sep 2003
4
5Attention! Database servers, especially those using "TCQ" disks should
6investigate performance with the 'deadline' IO scheduler. Any system with high
7disk performance requirements should do so, in fact.
8
9If you see unusual performance characteristics of your disk systems, or you
10see big performance regressions versus the deadline scheduler, please email
11me. Database users don't bother unless you're willing to test a lot of patches
12from me ;) its a known issue.
13
14Also, users with hardware RAID controllers, doing striping, may find
15highly variable performance results with using the as-iosched. The
16as-iosched anticipatory implementation is based on the notion that a disk
17device has only one physical seeking head. A striped RAID controller
18actually has a head for each physical device in the logical RAID device.
19
20However, setting the antic_expire (see tunable parameters below) produces
21very similar behavior to the deadline IO scheduler.
22
23Selecting IO schedulers
24-----------------------
25Refer to Documentation/block/switching-sched.txt for information on
26selecting an io scheduler on a per-device basis.
27
28Anticipatory IO scheduler Policies
29----------------------------------
30The as-iosched implementation implements several layers of policies
31to determine when an IO request is dispatched to the disk controller.
32Here are the policies outlined, in order of application.
33
341. one-way Elevator algorithm.
35
36The elevator algorithm is similar to that used in deadline scheduler, with
37the addition that it allows limited backward movement of the elevator
38(i.e. seeks backwards). A seek backwards can occur when choosing between
39two IO requests where one is behind the elevator's current position, and
40the other is in front of the elevator's position. If the seek distance to
41the request in back of the elevator is less than half the seek distance to
42the request in front of the elevator, then the request in back can be chosen.
43Backward seeks are also limited to a maximum of MAXBACK (1024*1024) sectors.
44This favors forward movement of the elevator, while allowing opportunistic
45"short" backward seeks.
46
472. FIFO expiration times for reads and for writes.
48
49This is again very similar to the deadline IO scheduler. The expiration
50times for requests on these lists is tunable using the parameters read_expire
51and write_expire discussed below. When a read or a write expires in this way,
52the IO scheduler will interrupt its current elevator sweep or read anticipation
53to service the expired request.
54
553. Read and write request batching
56
57A batch is a collection of read requests or a collection of write
58requests. The as scheduler alternates dispatching read and write batches
59to the driver. In the case a read batch, the scheduler submits read
60requests to the driver as long as there are read requests to submit, and
61the read batch time limit has not been exceeded (read_batch_expire).
62The read batch time limit begins counting down only when there are
63competing write requests pending.
64
65In the case of a write batch, the scheduler submits write requests to
66the driver as long as there are write requests available, and the
67write batch time limit has not been exceeded (write_batch_expire).
68However, the length of write batches will be gradually shortened
69when read batches frequently exceed their time limit.
70
71When changing between batch types, the scheduler waits for all requests
72from the previous batch to complete before scheduling requests for the
73next batch.
74
75The read and write fifo expiration times described in policy 2 above
76are checked only when in scheduling IO of a batch for the corresponding
77(read/write) type. So for example, the read FIFO timeout values are
78tested only during read batches. Likewise, the write FIFO timeout
79values are tested only during write batches. For this reason,
80it is generally not recommended for the read batch time
81to be longer than the write expiration time, nor for the write batch
82time to exceed the read expiration time (see tunable parameters below).
83
84When the IO scheduler changes from a read to a write batch,
85it begins the elevator from the request that is on the head of the
86write expiration FIFO. Likewise, when changing from a write batch to
87a read batch, scheduler begins the elevator from the first entry
88on the read expiration FIFO.
89
904. Read anticipation.
91
92Read anticipation occurs only when scheduling a read batch.
93This implementation of read anticipation allows only one read request
94to be dispatched to the disk controller at a time. In
95contrast, many write requests may be dispatched to the disk controller
96at a time during a write batch. It is this characteristic that can make
97the anticipatory scheduler perform anomalously with controllers supporting
98TCQ, or with hardware striped RAID devices. Setting the antic_expire
99queue parameter (see below) to zero disables this behavior, and the
100anticipatory scheduler behaves essentially like the deadline scheduler.
101
102When read anticipation is enabled (antic_expire is not zero), reads
103are dispatched to the disk controller one at a time.
104At the end of each read request, the IO scheduler examines its next
105candidate read request from its sorted read list. If that next request
106is from the same process as the request that just completed,
107or if the next request in the queue is "very close" to the
108just completed request, it is dispatched immediately. Otherwise,
109statistics (average think time, average seek distance) on the process
110that submitted the just completed request are examined. If it seems
111likely that that process will submit another request soon, and that
112request is likely to be near the just completed request, then the IO
113scheduler will stop dispatching more read requests for up to (antic_expire)
114milliseconds, hoping that process will submit a new request near the one
115that just completed. If such a request is made, then it is dispatched
116immediately. If the antic_expire wait time expires, then the IO scheduler
117will dispatch the next read request from the sorted read queue.
118
119To decide whether an anticipatory wait is worthwhile, the scheduler
120maintains statistics for each process that can be used to compute
121mean "think time" (the time between read requests), and mean seek
122distance for that process. One observation is that these statistics
123are associated with each process, but those statistics are not associated
124with a specific IO device. So for example, if a process is doing IO
125on several file systems on separate devices, the statistics will be
126a combination of IO behavior from all those devices.
127
128
129Tuning the anticipatory IO scheduler
130------------------------------------
131When using 'as', the anticipatory IO scheduler there are 5 parameters under
132/sys/block/*/queue/iosched/. All are units of milliseconds.
133
134The parameters are:
135* read_expire
136 Controls how long until a read request becomes "expired". It also controls the
137 interval between which expired requests are served, so set to 50, a request
138 might take anywhere < 100ms to be serviced _if_ it is the next on the
139 expired list. Obviously request expiration strategies won't make the disk
140 go faster. The result basically equates to the timeslice a single reader
141 gets in the presence of other IO. 100*((seek time / read_expire) + 1) is
142 very roughly the % streaming read efficiency your disk should get with
143 multiple readers.
144
145* read_batch_expire
146 Controls how much time a batch of reads is given before pending writes are
147 served. A higher value is more efficient. This might be set below read_expire
148 if writes are to be given higher priority than reads, but reads are to be
149 as efficient as possible when there are no writes. Generally though, it
150 should be some multiple of read_expire.
151
152* write_expire, and
153* write_batch_expire are equivalent to the above, for writes.
154
155* antic_expire
156 Controls the maximum amount of time we can anticipate a good read (one
157 with a short seek distance from the most recently completed request) before
158 giving up. Many other factors may cause anticipation to be stopped early,
159 or some processes will not be "anticipated" at all. Should be a bit higher
160 for big seek time devices though not a linear correspondence - most
161 processes have only a few ms thinktime.
162
163In addition to the tunables above there is a read-only file named est_time
164which, when read, will show:
165
166 - The probability of a task exiting without a cooperating task
167 submitting an anticipated IO.
168
169 - The current mean think time.
170
171 - The seek distance used to determine if an incoming IO is better.
172