sFlow Datagram Flow

As shown in the diagram above, the sFlow datagram generated on LC1 can be sent to sFlow collector(s) via:

• LC's own data (in-band) ports - This has the least CPU overhead in terms of forwarding the sFlow datagram to the collectors.
• Another LC's data (in-band) ports - This has some amount of overhead in forwarding the sFlow datagram to the collectors since it has to forward from one LC to another LC before exiting through the other LC's data (in-band) port.
• MM management port - This has the maximum CPU overhead since the MM CPU has to process the messages (sflow datagrams) forwarded by the LC and then route them out through its management port.

Note

Wherever possible, you must configure the sFlow collectors in such a way that the sFlow datagram gets routed through the same LC data (in-band) ports as described in option 1 above. If this is not possible, option 2 mentioned above may be considered as the next option. Option 3 is the least preferred in deployed systems due to the maximum CPU overhead.