| Brendan Gregg | 23c96fe | 2016-02-12 02:25:32 -0800 | [diff] [blame] | 1 | Demonstrations of xfsdist, the Linux eBPF/bcc version. |
| 2 | |
| 3 | |
| 4 | xfsdist traces XFS reads, writes, opens, and fsyncs, and summarizes their |
| 5 | latency as a power-of-2 histogram. For example: |
| 6 | |
| 7 | # ./xfsdist |
| 8 | Tracing XFS operation latency... Hit Ctrl-C to end. |
| 9 | ^C |
| 10 | |
| 11 | operation = 'read' |
| 12 | usecs : count distribution |
| 13 | 0 -> 1 : 0 | | |
| 14 | 2 -> 3 : 362 | | |
| 15 | 4 -> 7 : 807 |* | |
| 16 | 8 -> 15 : 20686 |****************************************| |
| 17 | 16 -> 31 : 512 | | |
| 18 | 32 -> 63 : 4 | | |
| 19 | 64 -> 127 : 2744 |***** | |
| 20 | 128 -> 255 : 7127 |************* | |
| 21 | 256 -> 511 : 2483 |**** | |
| 22 | 512 -> 1023 : 1281 |** | |
| 23 | 1024 -> 2047 : 39 | | |
| 24 | 2048 -> 4095 : 5 | | |
| 25 | 4096 -> 8191 : 1 | | |
| 26 | |
| 27 | operation = 'open' |
| 28 | usecs : count distribution |
| 29 | 0 -> 1 : 0 | | |
| 30 | 2 -> 3 : 3 |****************************************| |
| 31 | |
| 32 | This output shows a bi-modal distribution for read latency, with a faster |
| 33 | mode of 20,686 reads that took between 8 and 15 microseconds, and a slower |
| 34 | mode of over 10,000 reads that took between 64 and 1023 microseconds. It's |
| 35 | likely that the faster mode was a hit from the in-memory file system cache, |
| 36 | and the slower mode is a read from a storage device (disk). |
| 37 | |
| 38 | This "latency" is measured from when the operation was issued from the VFS |
| 39 | interface to the file system, to when it completed. This spans everything: |
| 40 | block device I/O (disk I/O), file system CPU cycles, file system locks, run |
| 41 | queue latency, etc. This is a better measure of the latency suffered by |
| 42 | applications reading from the file system than measuring this down at the |
| 43 | block device interface. |
| 44 | |
| 45 | Note that this only traces the common file system operations previously |
| 46 | listed: other file system operations (eg, inode operations including |
| 47 | getattr()) are not traced. |
| 48 | |
| 49 | |
| 50 | An optional interval and a count can be provided, as well as -m to show the |
| 51 | distributions in milliseconds. For example: |
| 52 | |
| 53 | # ./xfsdist -m 1 5 |
| 54 | Tracing XFS operation latency... Hit Ctrl-C to end. |
| 55 | |
| 56 | 10:14:15: |
| 57 | |
| 58 | operation = 'read' |
| 59 | msecs : count distribution |
| 60 | 0 -> 1 : 1366 |****************************************| |
| 61 | 2 -> 3 : 86 |** | |
| 62 | 4 -> 7 : 95 |** | |
| 63 | 8 -> 15 : 132 |*** | |
| 64 | 16 -> 31 : 72 |** | |
| 65 | |
| 66 | operation = 'write' |
| 67 | msecs : count distribution |
| 68 | 0 -> 1 : 685 |****************************************| |
| 69 | |
| 70 | 10:14:16: |
| 71 | |
| 72 | operation = 'read' |
| 73 | msecs : count distribution |
| 74 | 0 -> 1 : 984 |****************************************| |
| 75 | 2 -> 3 : 66 |** | |
| 76 | 4 -> 7 : 67 |** | |
| 77 | 8 -> 15 : 104 |**** | |
| 78 | 16 -> 31 : 70 |** | |
| 79 | 32 -> 63 : 12 | | |
| 80 | |
| 81 | operation = 'write' |
| 82 | msecs : count distribution |
| 83 | 0 -> 1 : 536 |****************************************| |
| 84 | |
| 85 | 10:14:17: |
| 86 | |
| 87 | operation = 'read' |
| 88 | msecs : count distribution |
| 89 | 0 -> 1 : 1262 |****************************************| |
| 90 | 2 -> 3 : 75 |** | |
| 91 | 4 -> 7 : 80 |** | |
| 92 | 8 -> 15 : 119 |*** | |
| 93 | 16 -> 31 : 75 |** | |
| 94 | 32 -> 63 : 3 | | |
| 95 | |
| 96 | operation = 'write' |
| 97 | msecs : count distribution |
| 98 | 0 -> 1 : 639 |****************************************| |
| 99 | |
| 100 | 10:14:18: |
| 101 | |
| 102 | operation = 'read' |
| 103 | msecs : count distribution |
| 104 | 0 -> 1 : 1070 |****************************************| |
| 105 | 2 -> 3 : 58 |** | |
| 106 | 4 -> 7 : 74 |** | |
| 107 | 8 -> 15 : 140 |***** | |
| 108 | 16 -> 31 : 60 |** | |
| 109 | 32 -> 63 : 5 | | |
| 110 | |
| 111 | operation = 'write' |
| 112 | msecs : count distribution |
| 113 | 0 -> 1 : 556 |****************************************| |
| 114 | |
| 115 | 10:14:19: |
| 116 | |
| 117 | operation = 'read' |
| 118 | msecs : count distribution |
| 119 | 0 -> 1 : 1176 |****************************************| |
| 120 | 2 -> 3 : 53 |* | |
| 121 | 4 -> 7 : 94 |*** | |
| 122 | 8 -> 15 : 112 |*** | |
| 123 | 16 -> 31 : 77 |** | |
| 124 | 32 -> 63 : 3 | | |
| 125 | |
| 126 | operation = 'write' |
| 127 | msecs : count distribution |
| 128 | 0 -> 1 : 613 |****************************************| |
| 129 | |
| 130 | This shows a mixed read/write workload, where the slower read mode was around |
| 131 | 10 ms. |
| 132 | |
| 133 | |
| 134 | USAGE message: |
| 135 | |
| 136 | # ./xfsdist -h |
| 137 | usage: xfsdist [-h] [-T] [-m] [-p PID] [interval] [count] |
| 138 | |
| 139 | Summarize XFS operation latency |
| 140 | |
| 141 | positional arguments: |
| 142 | interval output interval, in seconds |
| 143 | count number of outputs |
| 144 | |
| 145 | optional arguments: |
| 146 | -h, --help show this help message and exit |
| 147 | -T, --notimestamp don't include timestamp on interval output |
| 148 | -m, --milliseconds output in milliseconds |
| 149 | -p PID, --pid PID trace this PID only |
| 150 | |
| 151 | examples: |
| 152 | ./xfsdist # show operation latency as a histogram |
| 153 | ./xfsdist -p 181 # trace PID 181 only |
| 154 | ./xfsdist 1 10 # print 1 second summaries, 10 times |
| 155 | ./xfsdist -m 5 # 5s summaries, milliseconds |