| Brendan Gregg | ddce4db | 2016-02-15 22:24:02 -0800 | [diff] [blame] | 1 | Demonstrations of btrfsdist, the Linux eBPF/bcc version. |
| 2 | |
| 3 | |
| 4 | btrfsdist traces btrfs reads, writes, opens, and fsyncs, and summarizes their |
| 5 | latency as a power-of-2 histogram. For example: |
| 6 | |
| 7 | # ./btrfsdist |
| 8 | Tracing btrfs operation latency... Hit Ctrl-C to end. |
| 9 | ^C |
| 10 | |
| 11 | operation = 'read' |
| 12 | usecs : count distribution |
| 13 | 0 -> 1 : 15 | | |
| 14 | 2 -> 3 : 1308 |******* | |
| 15 | 4 -> 7 : 198 |* | |
| 16 | 8 -> 15 : 0 | | |
| 17 | 16 -> 31 : 11 | | |
| 18 | 32 -> 63 : 361 |* | |
| 19 | 64 -> 127 : 55 | | |
| 20 | 128 -> 255 : 104 | | |
| 21 | 256 -> 511 : 7312 |****************************************| |
| 22 | 512 -> 1023 : 387 |** | |
| 23 | 1024 -> 2047 : 10 | | |
| 24 | 2048 -> 4095 : 4 | | |
| 25 | |
| 26 | operation = 'write' |
| 27 | usecs : count distribution |
| 28 | 0 -> 1 : 0 | | |
| 29 | 2 -> 3 : 0 | | |
| 30 | 4 -> 7 : 0 | | |
| 31 | 8 -> 15 : 4 |****************************************| |
| 32 | |
| 33 | operation = 'open' |
| 34 | usecs : count distribution |
| 35 | 0 -> 1 : 1 |********** | |
| 36 | 2 -> 3 : 4 |****************************************| |
| 37 | |
| 38 | This output shows a bi-modal distribution for read latency, with a faster |
| 39 | mode of 1,308 reads that took between 2 and 3 microseconds, and a slower |
| 40 | mode of over 7,312 reads that took between 256 and 511 microseconds. It's |
| 41 | likely that the faster mode was a hit from the in-memory file system cache, |
| 42 | and the slower mode is a read from a storage device (disk). |
| 43 | |
| 44 | This "latency" is measured from when the operation was issued from the VFS |
| 45 | interface to the file system, to when it completed. This spans everything: |
| 46 | block device I/O (disk I/O), file system CPU cycles, file system locks, run |
| 47 | queue latency, etc. This is a better measure of the latency suffered by |
| 48 | applications reading from the file system than measuring this down at the |
| 49 | block device interface. |
| 50 | |
| 51 | Note that this only traces the common file system operations previously |
| 52 | listed: other file system operations (eg, inode operations including |
| 53 | getattr()) are not traced. |
| 54 | |
| 55 | |
| 56 | An optional interval and a count can be provided, as well as -m to show the |
| 57 | distributions in milliseconds. For example, two second summaries, five times: |
| 58 | |
| 59 | # ./btrfsdist 2 5 |
| 60 | Tracing btrfs operation latency... Hit Ctrl-C to end. |
| 61 | |
| 62 | 03:40:49: |
| 63 | |
| 64 | operation = 'read' |
| 65 | usecs : count distribution |
| 66 | 0 -> 1 : 15 | | |
| 67 | 2 -> 3 : 833 |******** | |
| 68 | 4 -> 7 : 127 |* | |
| 69 | 8 -> 15 : 0 | | |
| 70 | 16 -> 31 : 8 | | |
| 71 | 32 -> 63 : 907 |******** | |
| 72 | 64 -> 127 : 91 | | |
| 73 | 128 -> 255 : 246 |** | |
| 74 | 256 -> 511 : 4164 |****************************************| |
| 75 | 512 -> 1023 : 193 |* | |
| 76 | 1024 -> 2047 : 4 | | |
| 77 | 2048 -> 4095 : 6 | | |
| 78 | 4096 -> 8191 : 2 | | |
| 79 | |
| 80 | 03:40:51: |
| 81 | |
| 82 | operation = 'read' |
| 83 | usecs : count distribution |
| 84 | 0 -> 1 : 25 | | |
| 85 | 2 -> 3 : 1491 |*************** | |
| 86 | 4 -> 7 : 218 |** | |
| 87 | 8 -> 15 : 0 | | |
| 88 | 16 -> 31 : 16 | | |
| 89 | 32 -> 63 : 1527 |*************** | |
| 90 | 64 -> 127 : 319 |*** | |
| 91 | 128 -> 255 : 429 |**** | |
| 92 | 256 -> 511 : 3841 |****************************************| |
| 93 | 512 -> 1023 : 232 |** | |
| 94 | 1024 -> 2047 : 3 | | |
| 95 | 2048 -> 4095 : 6 | | |
| 96 | 4096 -> 8191 : 1 | | |
| 97 | 8192 -> 16383 : 1 | | |
| 98 | |
| 99 | 03:40:53: |
| 100 | |
| 101 | operation = 'read' |
| 102 | usecs : count distribution |
| 103 | 0 -> 1 : 27 | | |
| 104 | 2 -> 3 : 2999 |********************************* | |
| 105 | 4 -> 7 : 407 |**** | |
| 106 | 8 -> 15 : 0 | | |
| 107 | 16 -> 31 : 46 | | |
| 108 | 32 -> 63 : 3538 |****************************************| |
| 109 | 64 -> 127 : 595 |****** | |
| 110 | 128 -> 255 : 621 |******* | |
| 111 | 256 -> 511 : 3532 |*************************************** | |
| 112 | 512 -> 1023 : 212 |** | |
| 113 | 1024 -> 2047 : 1 | | |
| 114 | 2048 -> 4095 : 0 | | |
| 115 | 4096 -> 8191 : 0 | | |
| 116 | 8192 -> 16383 : 0 | | |
| 117 | 16384 -> 32767 : 1 | | |
| 118 | |
| 119 | 03:40:55: |
| 120 | |
| 121 | operation = 'read' |
| 122 | usecs : count distribution |
| 123 | 0 -> 1 : 221 | | |
| 124 | 2 -> 3 : 12580 |****************************************| |
| 125 | 4 -> 7 : 1366 |**** | |
| 126 | 8 -> 15 : 0 | | |
| 127 | 16 -> 31 : 289 | | |
| 128 | 32 -> 63 : 10782 |********************************** | |
| 129 | 64 -> 127 : 1232 |*** | |
| 130 | 128 -> 255 : 807 |** | |
| 131 | 256 -> 511 : 2299 |******* | |
| 132 | 512 -> 1023 : 135 | | |
| 133 | 1024 -> 2047 : 5 | | |
| 134 | 2048 -> 4095 : 2 | | |
| 135 | |
| 136 | 03:40:57: |
| 137 | |
| 138 | operation = 'read' |
| 139 | usecs : count distribution |
| 140 | 0 -> 1 : 73951 |************************* | |
| 141 | 2 -> 3 : 117639 |****************************************| |
| 142 | 4 -> 7 : 7943 |** | |
| 143 | 8 -> 15 : 1841 | | |
| 144 | 16 -> 31 : 1143 | | |
| 145 | 32 -> 63 : 5006 |* | |
| 146 | 64 -> 127 : 483 | | |
| 147 | 128 -> 255 : 242 | | |
| 148 | 256 -> 511 : 253 | | |
| 149 | 512 -> 1023 : 84 | | |
| 150 | 1024 -> 2047 : 23 | | |
| 151 | |
| 152 | This shows a read workload that begins bimodal, and eventually the second |
| 153 | mode disappears. The reason for this is that the workload cached during |
| 154 | tracing. Note that the rate also increased, with over 200k reads for the |
| 155 | final two second sample. |
| 156 | |
| 157 | |
| 158 | USAGE message: |
| 159 | |
| 160 | # ./btrfsdist -h |
| 161 | usage: btrfsdist [-h] [-T] [-m] [-p PID] [interval] [count] |
| 162 | |
| 163 | Summarize btrfs operation latency |
| 164 | |
| 165 | positional arguments: |
| 166 | interval output interval, in seconds |
| 167 | count number of outputs |
| 168 | |
| 169 | optional arguments: |
| 170 | -h, --help show this help message and exit |
| 171 | -T, --notimestamp don't include timestamp on interval output |
| 172 | -m, --milliseconds output in milliseconds |
| 173 | -p PID, --pid PID trace this PID only |
| 174 | |
| 175 | examples: |
| 176 | ./btrfsdist # show operation latency as a histogram |
| 177 | ./btrfsdist -p 181 # trace PID 181 only |
| 178 | ./btrfsdist 1 10 # print 1 second summaries, 10 times |
| 179 | ./btrfsdist -m 5 # 5s summaries, milliseconds |