|
| 1 | + |
| 2 | +The Exact Capture application supports a number of configuration options in both short and long form. |
| 3 | +For example: |
| 4 | +``` |
| 5 | +$ exact-capture -i exanic0:0 --log-report-int 10 .... |
| 6 | +``` |
| 7 | + |
| 8 | +A [quick start](quick.md) guide is available for getting started. |
| 9 | +The following table lists all commands available: |
| 10 | + |
| 11 | +<table> |
| 12 | + <tr> |
| 13 | + <th>Short</th> |
| 14 | + <th>Long</th> |
| 15 | + <th>Default</th> |
| 16 | + <th>Description</th> |
| 17 | + </tr> |
| 18 | + <tr> |
| 19 | + <td>i</td> |
| 20 | + <td>input</td> |
| 21 | + <td><em>(required)</em></td> |
| 22 | + <td> |
| 23 | + The ExaNIC interface(s) to capture on |
| 24 | + </td> |
| 25 | + </tr> |
| 26 | + <tr> |
| 27 | + <td>o</td> |
| 28 | + <td>output</td> |
| 29 | + <td><em>(required)</em></td> |
| 30 | + <td> |
| 31 | + The destination directory and filename stub to output to. |
| 32 | + Filenames will be output in the following format /output/dir/base_xx.expcap. |
| 33 | + Where xx is a unique file index. |
| 34 | + For details on the expcap format please see the Exact Capture Output Format (expcap) section later in this document. |
| 35 | + </td> |
| 36 | + </tr> |
| 37 | + <tr> |
| 38 | + <td>c</td> |
| 39 | + <td><a name="cpus">cpus</a></td> |
| 40 | + <td><em>(required)</em></td> |
| 41 | + <td> |
| 42 | + The list of CPUs to assign threads to for management, listening and writing threads. |
| 43 | + This is specified in the the following format, <code>m:ls,ls,ls:ws,ws,ws</code>. |
| 44 | + Where m is the core number for management, and ls/ws are comma separated lists of listener and writer CPU core numbers. |
| 45 | + For example <code>--cpus=5:2,3:7,6,1</code> would configure Exact Capture to run the management thread on CPU 5, with two NIC listener threads on on CPUs 2 and 3 respectively, and three (or more) disk writer threads on cores 1,6 & 7 respectively. |
| 46 | + </br></br> |
| 47 | + <strong>Note:</strong> the number of listener CPUs must be exactly equal to the number of ExaNIC <code>--interfaces</code> in use. |
| 48 | + Furthermore listener threads cannot share CPUs with management or writer threads. |
| 49 | + If there are fewer writer threads than <code>--outputs</code>, writer threads will be reused. |
| 50 | + </td> |
| 51 | + <tr> |
| 52 | + <td>s</td> |
| 53 | + <td>snaplen</td> |
| 54 | + <td>2048B</td> |
| 55 | + <td> |
| 56 | + In some cases it is not necessary / useful to capture the entire packet. |
| 57 | + Set the snap length to determine the maximum size of packet that can be captured. |
| 58 | + This value cannot be 0 or less. |
| 59 | + </td> |
| 60 | + </tr> |
| 61 | + <tr> |
| 62 | + <td>m</td> |
| 63 | + <td>maxfile</td> |
| 64 | + <td>0 <em>(unlimited)</em></td> |
| 65 | + <td> |
| 66 | + High rate capture can produce very large file sizes. |
| 67 | + To reduce the file sizes, Exact Capture can cap the file size to a maximum, and will start a new file each time it is reached. |
| 68 | + A value of 0 or less puts no limit on the output file size. |
| 69 | + </td> |
| 70 | + </tr> |
| 71 | + <tr> |
| 72 | + <td>l</td> |
| 73 | + <td>logfile</td> |
| 74 | + <td><em>(none)</em></td> |
| 75 | + <td> |
| 76 | + Exact capture can optionally write log messages to a log file specified. |
| 77 | + </td> |
| 78 | + </tr> |
| 79 | + <tr> |
| 80 | + <td>t</td> |
| 81 | + <td>log-report-int</td> |
| 82 | + <td>1.0</td> |
| 83 | + <td> |
| 84 | + This sets the statistics calculation and logging interval in seconds. |
| 85 | + </td> |
| 86 | + </tr> |
| 87 | + <tr> |
| 88 | + <td>v</td> |
| 89 | + <td>verbose</td> |
| 90 | + <td><em>(flag)</em></td> |
| 91 | + <td> |
| 92 | + Enabling verbose mode will produce 2 output log lines every log interval (see above). |
| 93 | + These log lines will include summary statistics of the performance of all listener threads and all writer threads. |
| 94 | + </td> |
| 95 | + </tr> |
| 96 | + <tr> |
| 97 | + <td>V</td> |
| 98 | + <td>more-verbose</td> |
| 99 | + <td><em>(flag)</em></td> |
| 100 | + <td> |
| 101 | + Enabling more verbose mode will produce 1 output log line for every listener and writer thread. |
| 102 | + Each log line will include per-thread statistics counters/statistics. |
| 103 | + This can be combined with <code>--verbose</code> mode above. |
| 104 | + </td> |
| 105 | + </tr> |
| 106 | + <tr> |
| 107 | + <td>d</td> |
| 108 | + <td>debug-logging</td> |
| 109 | + <td><em>(flag)</em></td> |
| 110 | + <td> |
| 111 | + Debug logging mode enables display of the full file path, process ID and thread ID in each output log line. |
| 112 | + This is useful to track where a given log message originated. |
| 113 | + </td> |
| 114 | + </tr> |
| 115 | + <tr> |
| 116 | + <td>T</td> |
| 117 | + <td>no-log-ts</td> |
| 118 | + <td><em>(flag)</em></td> |
| 119 | + <td> |
| 120 | + By default, logs include a timestamp. |
| 121 | + This can make the output overly verbose. |
| 122 | + Use this flag to disable timestamps. |
| 123 | + </td> |
| 124 | + </tr> |
| 125 | + <tr> |
| 126 | + <td>w</td> |
| 127 | + <td>no-warn-overflow</td> |
| 128 | + <td><em>(flag)</em></td> |
| 129 | + <td> |
| 130 | + Software overflows will produce a warning. |
| 131 | + This may be problematic if the system is underperforming and these happen often. |
| 132 | + The flag disables these warnings. |
| 133 | + </td> |
| 134 | + </tr> |
| 135 | + <tr> |
| 136 | + <td>n</td> |
| 137 | + <td>no-spin</td> |
| 138 | + <td><em>(flag)</em></td> |
| 139 | + <td> |
| 140 | + By default Exact Capture outputs a progress “spinner” to the console. |
| 141 | + This flag disables it. |
| 142 | + </td> |
| 143 | + </tr> |
| 144 | + <tr> |
| 145 | + <td>n</td> |
| 146 | + <td>no-spin</td> |
| 147 | + <td><em>(flag)</em></td> |
| 148 | + <td> |
| 149 | + Exact Capture supports several performance testing modes. |
| 150 | + These can be used to give a sense of the best possible performance that you can expect from your system configuration. |
| 151 | + The modes are as follows: |
| 152 | + <ol> |
| 153 | + <li> No performance testing </li> |
| 154 | + <li> |
| 155 | + Replace all ExaNIC interfaces with a dummy interface. |
| 156 | + ExaNICs are no longer a performance limitation. |
| 157 | + This tests the maximum possible receive rate that your system can achieve for 64B frames. |
| 158 | + <strong>Note that 10GbE line-rate with 64B frames is about 7Gb/s (due to Ethernet interfame gap overheads).</strong> |
| 159 | + </li> |
| 160 | + <li> |
| 161 | + Replace the internal memory queue with a dummy interface on both sides. |
| 162 | + This tests the maximum performance possible when when system memory is not the bottleneck. |
| 163 | + This is also a good test of disk writing speed (for minimum sized packets). |
| 164 | + </li> |
| 165 | + <li> |
| 166 | + Replace the ExaDisk interface with a dummy. |
| 167 | + This tests the performance through the system when disk writing speed is not a limitation. |
| 168 | + This may be helpful to debug cases where your disks are not configured/performing correctly. |
| 169 | + </li> |
| 170 | + <li> |
| 171 | + Replace both the ExaNIC and the internal memory ring with dummies. |
| 172 | + Can be used to measure the absolute best performance possible when NICs and memory are not the limitations. |
| 173 | + Can also help to find interference bugs between ExaNICs and ExaDisks sharing limited PCIe bandwidth. |
| 174 | + </li> |
| 175 | + <li> |
| 176 | + Replace the ExaNIC and ExaDisk with dummies. |
| 177 | + This is useful for testing the maximum achievable application throughput, including through system memory, but excluding reading from and writing to real hardware. |
| 178 | + </li> |
| 179 | + <li> |
| 180 | + Replace the memory queues and ExaDisk with dummies. |
| 181 | + Can be help to find interference bugs between ExaNICs and ExaDisks sharing limited PCIe bandwidth. |
| 182 | + </li> |
| 183 | + <li> |
| 184 | + Replace the ExaNIC, memory queue and ExaDisk interfaces with dummies. |
| 185 | + Useful for determining the overheads within the application (i.e. CPU speed) issues. |
| 186 | + </li> |
| 187 | + </td> |
| 188 | + </tr> |
| 189 | + |
| 190 | +</table> |
0 commit comments