Extended Main Memory

帮忙翻译如下内容/* * Configure a filter for packet traces. * * This supplements the packet trace feature so that only packets matching * the filter are included in the trace. Currently the only filter is to * keep packets that include a certain node in the trace or exclude a certain * node in the trace. * * The count of traced packets in the "trace add" command is still used to * create a certain number of traces. The "trace filter" command specifies * how many of those packets should be retained in the trace. * * For example, 1Mpps of traffic is arriving and one of those packets is being * dropped. To capture the trace for only that dropped packet, you can do: * trace filter include error-drop 1 * trace add dpdk-input 1000000 * <wait one second> * show trace * * Note that the filter could be implemented by capturing all traces and just * reducing traces displayed by the "show trace" function. But that would * require a lot of memory for storing the traces, making that infeasible. * * To remove traces from the trace pool that do not include a certain node * requires that the trace be "complete" before applying the filter. To * accomplish this, the trace pool is filtered upon each iteraction of the * main vlib loop. Doing so keeps the number of allocated traces down to a * reasonably low number. This requires that tracing for a buffer is not * performed after the vlib main loop interation completes. i.e. you can't * save away a buffer temporarily then inject it back into the graph and * expect that the trace_index is still valid (such as a traffic manager might * do). A new trace buffer should be allocated for those types of packets. * * The filter can be extended to support multiple nodes and other match * criteria (e.g. input sw_if_index, mac address) but for now just checks if * a specified node is in the trace or not in the trace. */