Commit | Line | Data |
---|---|---|
fd2ed4d2 MP |
1 | DM statistics |
2 | ============= | |
3 | ||
4 | Device Mapper supports the collection of I/O statistics on user-defined | |
5 | regions of a DM device. If no regions are defined no statistics are | |
6 | collected so there isn't any performance impact. Only bio-based DM | |
7 | devices are currently supported. | |
8 | ||
9 | Each user-defined region specifies a starting sector, length and step. | |
10 | Individual statistics will be collected for each step-sized area within | |
11 | the range specified. | |
12 | ||
13 | The I/O statistics counters for each step-sized area of a region are | |
14 | in the same format as /sys/block/*/stat or /proc/diskstats (see: | |
15 | Documentation/iostats.txt). But two extra counters (12 and 13) are | |
dfcfac3e MP |
16 | provided: total time spent reading and writing. When the histogram |
17 | argument is used, the 14th parameter is reported that represents the | |
18 | histogram of latencies. All these counters may be accessed by sending | |
19 | the @stats_print message to the appropriate DM device via dmsetup. | |
c96aec34 MP |
20 | |
21 | The reported times are in milliseconds and the granularity depends on | |
22 | the kernel ticks. When the option precise_timestamps is used, the | |
23 | reported times are in nanoseconds. | |
fd2ed4d2 MP |
24 | |
25 | Each region has a corresponding unique identifier, which we call a | |
26 | region_id, that is assigned when the region is created. The region_id | |
27 | must be supplied when querying statistics about the region, deleting the | |
28 | region, etc. Unique region_ids enable multiple userspace programs to | |
29 | request and process statistics for the same DM device without stepping | |
30 | on each other's data. | |
31 | ||
32 | The creation of DM statistics will allocate memory via kmalloc or | |
33 | fallback to using vmalloc space. At most, 1/4 of the overall system | |
34 | memory may be allocated by DM statistics. The admin can see how much | |
35 | memory is used by reading | |
36 | /sys/module/dm_mod/parameters/stats_current_allocated_bytes | |
37 | ||
38 | Messages | |
39 | ======== | |
40 | ||
c96aec34 MP |
41 | @stats_create <range> <step> |
42 | [<number_of_optional_arguments> <optional_arguments>...] | |
43 | [<program_id> [<aux_data>]] | |
fd2ed4d2 MP |
44 | |
45 | Create a new region and return the region_id. | |
46 | ||
47 | <range> | |
48 | "-" - whole device | |
49 | "<start_sector>+<length>" - a range of <length> 512-byte sectors | |
50 | starting with <start_sector>. | |
51 | ||
52 | <step> | |
53 | "<area_size>" - the range is subdivided into areas each containing | |
54 | <area_size> sectors. | |
55 | "/<number_of_areas>" - the range is subdivided into the specified | |
56 | number of areas. | |
57 | ||
c96aec34 MP |
58 | <number_of_optional_arguments> |
59 | The number of optional arguments | |
60 | ||
61 | <optional_arguments> | |
62 | The following optional arguments are supported | |
63 | precise_timestamps - use precise timer with nanosecond resolution | |
64 | instead of the "jiffies" variable. When this argument is | |
65 | used, the resulting times are in nanoseconds instead of | |
66 | milliseconds. Precise timestamps are a little bit slower | |
67 | to obtain than jiffies-based timestamps. | |
dfcfac3e MP |
68 | histogram:n1,n2,n3,n4,... - collect histogram of latencies. The |
69 | numbers n1, n2, etc are times that represent the boundaries | |
70 | of the histogram. If precise_timestamps is not used, the | |
71 | times are in milliseconds, otherwise they are in | |
72 | nanoseconds. For each range, the kernel will report the | |
73 | number of requests that completed within this range. For | |
74 | example, if we use "histogram:10,20,30", the kernel will | |
75 | report four numbers a:b:c:d. a is the number of requests | |
76 | that took 0-10 ms to complete, b is the number of requests | |
77 | that took 10-20 ms to complete, c is the number of requests | |
78 | that took 20-30 ms to complete and d is the number of | |
79 | requests that took more than 30 ms to complete. | |
c96aec34 | 80 | |
fd2ed4d2 MP |
81 | <program_id> |
82 | An optional parameter. A name that uniquely identifies | |
83 | the userspace owner of the range. This groups ranges together | |
84 | so that userspace programs can identify the ranges they | |
85 | created and ignore those created by others. | |
86 | The kernel returns this string back in the output of | |
87 | @stats_list message, but it doesn't use it for anything else. | |
c96aec34 MP |
88 | If we omit the number of optional arguments, program id must not |
89 | be a number, otherwise it would be interpreted as the number of | |
90 | optional arguments. | |
fd2ed4d2 MP |
91 | |
92 | <aux_data> | |
93 | An optional parameter. A word that provides auxiliary data | |
94 | that is useful to the client program that created the range. | |
95 | The kernel returns this string back in the output of | |
96 | @stats_list message, but it doesn't use this value for anything. | |
97 | ||
98 | @stats_delete <region_id> | |
99 | ||
100 | Delete the region with the specified id. | |
101 | ||
102 | <region_id> | |
103 | region_id returned from @stats_create | |
104 | ||
105 | @stats_clear <region_id> | |
106 | ||
107 | Clear all the counters except the in-flight i/o counters. | |
108 | ||
109 | <region_id> | |
110 | region_id returned from @stats_create | |
111 | ||
112 | @stats_list [<program_id>] | |
113 | ||
114 | List all regions registered with @stats_create. | |
115 | ||
116 | <program_id> | |
117 | An optional parameter. | |
118 | If this parameter is specified, only matching regions | |
119 | are returned. | |
120 | If it is not specified, all regions are returned. | |
121 | ||
122 | Output format: | |
123 | <region_id>: <start_sector>+<length> <step> <program_id> <aux_data> | |
bd49784f MP |
124 | precise_timestamps histogram:n1,n2,n3,... |
125 | ||
126 | The strings "precise_timestamps" and "histogram" are printed only | |
127 | if they were specified when creating the region. | |
fd2ed4d2 MP |
128 | |
129 | @stats_print <region_id> [<starting_line> <number_of_lines>] | |
130 | ||
131 | Print counters for each step-sized area of a region. | |
132 | ||
133 | <region_id> | |
134 | region_id returned from @stats_create | |
135 | ||
136 | <starting_line> | |
137 | The index of the starting line in the output. | |
138 | If omitted, all lines are returned. | |
139 | ||
140 | <number_of_lines> | |
141 | The number of lines to include in the output. | |
142 | If omitted, all lines are returned. | |
143 | ||
144 | Output format for each step-sized area of a region: | |
145 | ||
146 | <start_sector>+<length> counters | |
147 | ||
148 | The first 11 counters have the same meaning as | |
149 | /sys/block/*/stat or /proc/diskstats. | |
150 | ||
151 | Please refer to Documentation/iostats.txt for details. | |
152 | ||
153 | 1. the number of reads completed | |
154 | 2. the number of reads merged | |
155 | 3. the number of sectors read | |
156 | 4. the number of milliseconds spent reading | |
157 | 5. the number of writes completed | |
158 | 6. the number of writes merged | |
159 | 7. the number of sectors written | |
160 | 8. the number of milliseconds spent writing | |
161 | 9. the number of I/Os currently in progress | |
162 | 10. the number of milliseconds spent doing I/Os | |
163 | 11. the weighted number of milliseconds spent doing I/Os | |
164 | ||
165 | Additional counters: | |
166 | 12. the total time spent reading in milliseconds | |
167 | 13. the total time spent writing in milliseconds | |
168 | ||
169 | @stats_print_clear <region_id> [<starting_line> <number_of_lines>] | |
170 | ||
171 | Atomically print and then clear all the counters except the | |
172 | in-flight i/o counters. Useful when the client consuming the | |
173 | statistics does not want to lose any statistics (those updated | |
174 | between printing and clearing). | |
175 | ||
176 | <region_id> | |
177 | region_id returned from @stats_create | |
178 | ||
179 | <starting_line> | |
180 | The index of the starting line in the output. | |
181 | If omitted, all lines are printed and then cleared. | |
182 | ||
183 | <number_of_lines> | |
184 | The number of lines to process. | |
185 | If omitted, all lines are printed and then cleared. | |
186 | ||
187 | @stats_set_aux <region_id> <aux_data> | |
188 | ||
189 | Store auxiliary data aux_data for the specified region. | |
190 | ||
191 | <region_id> | |
192 | region_id returned from @stats_create | |
193 | ||
194 | <aux_data> | |
195 | The string that identifies data which is useful to the client | |
196 | program that created the range. The kernel returns this | |
197 | string back in the output of @stats_list message, but it | |
198 | doesn't use this value for anything. | |
199 | ||
200 | Examples | |
201 | ======== | |
202 | ||
203 | Subdivide the DM device 'vol' into 100 pieces and start collecting | |
204 | statistics on them: | |
205 | ||
206 | dmsetup message vol 0 @stats_create - /100 | |
207 | ||
208 | Set the auxillary data string to "foo bar baz" (the escape for each | |
209 | space must also be escaped, otherwise the shell will consume them): | |
210 | ||
211 | dmsetup message vol 0 @stats_set_aux 0 foo\\ bar\\ baz | |
212 | ||
213 | List the statistics: | |
214 | ||
215 | dmsetup message vol 0 @stats_list | |
216 | ||
217 | Print the statistics: | |
218 | ||
219 | dmsetup message vol 0 @stats_print 0 | |
220 | ||
221 | Delete the statistics: | |
222 | ||
223 | dmsetup message vol 0 @stats_delete 0 |