Set traffic capturing options

There are two main ways of deploying Packetbeat:

The first option has the big advantage that there is no overhead of any kind on your application servers. But it requires dedicated networking gear, which is generally not available on cloud setups.

In both cases, the sniffing performance (reading packets passively from the network) is very important. In the case of a dedicated server, better sniffing performance means that less hardware is required. When Packetbeat is installed on an existing application server, better sniffing performance means less overhead.

Currently Packetbeat has several options for traffic capturing:

The af_packet option, also known as "memory-mapped sniffing," makes use of a Linux-specific feature. This could be the optimal sniffing mode for both the dedicated server and when Packetbeat is deployed on an existing application server.

The way it works is that both the kernel and the user space program map the same memory zone, and a simple circular buffer is organized in this memory zone. The kernel writes packets into the circular buffer, and the user space program reads from it. The poll system call is used for getting a notification for the first packet available, but the remaining available packets can be simply read via memory access.

The af_packet sniffer can be further tuned to use more memory in exchange for better performance. The larger the size of the circular buffer, the fewer system calls are needed, which means that fewer CPU cycles are consumed. The default size of the buffer is 30 MB, but you can increase it like this:

packetbeat.interfaces.device: eth0
packetbeat.interfaces.type: af_packet
packetbeat.interfaces.buffer_size_mb: 100

You can specify the following options in the packetbeat.interfaces section of the packetbeat.yml config file. Here is an example configuration:

packetbeat.interfaces.device: any
packetbeat.interfaces.snaplen: 1514
packetbeat.interfaces.type: af_packet
packetbeat.interfaces.buffer_size_mb: 100

The network device to capture traffic from. The specified device is set automatically to promiscuous mode, meaning that Packetbeat can capture traffic from other hosts on the same LAN.

Example:

packetbeat.interfaces.device: eth0

On Linux, you can specify any for the device, and Packetbeat captures all messages sent or received by the server where Packetbeat is installed.

The device option also accepts specifying the device by its index in the list of devices available for sniffing. To obtain the list of available devices, run Packetbeat with the following command:

packetbeat devices

This command returns a list that looks something like the following:

0: en0 (No description available)
1: awdl0 (No description available)
2: bridge0 (No description available)
3: fw0 (No description available)
4: en1 (No description available)
5: en2 (No description available)
6: p2p0 (No description available)
7: en4 (No description available)
8: lo0 (No description available)

The following example sets up sniffing on the first interface in the list:

packetbeat.interfaces.device: 0

Specifying the index is especially useful on Windows where device names can be long.

The maximum size of the packets to capture. The default is 65535, which is large enough for almost all networks and interface types. If you sniff on a physical network interface, the optimal setting is the MTU size. On virtual interfaces, however, it’s safer to accept the default value.

Example:

packetbeat.interfaces.device: eth0
packetbeat.interfaces.snaplen: 1514

Packetbeat supports two sniffer types:

The default sniffer type is pcap.

Here is an example configuration that specifies the af_packet sniffing type:

packetbeat.interfaces.device: eth0
packetbeat.interfaces.type: af_packet

On Linux, if you are trying to optimize the CPU usage of Packetbeat, we recommend trying the af_packet option.

If you use the af_packet sniffer, you can tune its behaviour by specifying the following options:

The maximum size of the shared memory buffer to use between the kernel and user space. A bigger buffer usually results in lower CPU usage, but consumes more memory. This setting is only available for the af_packet sniffer type. The default is 30 MB.

Example:

packetbeat.interfaces.device: eth0
packetbeat.interfaces.type: af_packet
packetbeat.interfaces.buffer_size_mb: 100

Packetbeat automatically generates a BPF for capturing only the traffic on ports where it expects to find known protocols. For example, if you have configured port 80 for HTTP and port 3306 for MySQL, Packetbeat generates the following BPF filter: "port 80 or port 3306".

However, if the traffic contains VLAN tags, the filter that Packetbeat generates is ineffective because the offset is moved by four bytes. To fix this, you can enable the with_vlans option, which generates a BPF filter that looks like this: "port 80 or port 3306 or (vlan and (port 80 or port 3306))".

Packetbeat automatically generates a BPF for capturing only the traffic on ports where it expects to find known protocols. For example, if you have configured port 80 for HTTP and port 3306 for MySQL, Packetbeat generates the following BPF filter: "port 80 or port 3306".

You can use the bpf_filter setting to overwrite the generated BPF filter. For example:

packetbeat.interfaces.device: eth0
packetbeat.interfaces.bpf_filter: "net 192.168.238.0/0 and port 80 or port 3306"

If the ignore_outgoing option is enabled, Packetbeat ignores all the transactions initiated from the server running Packetbeat.

This is useful when two Packetbeat instances publish the same transactions. Because one Packetbeat sees the transaction in its outgoing queue and the other sees it in its incoming queue, you can end up with duplicate transactions. To remove the duplicates, you can enable the packetbeat.ignore_outgoing option on one of the servers.

For example, in the following scenario, you see a 3-server architecture where a Beat is installed on each server. t1 is the transaction exchanged between Server1 and Server2, and t2 is the transaction between Server2 and Server3.

By default, each transaction is indexed twice because Beat2 sees both transactions. So you would see the following published transactions (when ignore_outgoing is false):

To avoid duplicates, you can force your Beats to send only the incoming transactions and ignore the transactions created by the local server. So you would see the following published transactions (when ignore_outgoing is true):