ICMP and SNMP traffic is not insignificant or negligible

I recently had the question come up about how much bandwidth SNMP monitoring consumes. I did some cursory searching the web and found that others had posed the same question. Unfortunately, many responses simply said that ICMP and SNMP traffic is insignificant or negligible.  In my case, this arose to ensure that the circuit used by the collector was adequately sized.  We were looking at thousands of devices being monitored.  On the collector I ran a packet capture to get started.  I was polling  a Cisco Catalyst 3750-X switch using SNMP v2c.

6,672 bits per interface polled

The traffic generated from polling a single interface was about 6672 bits during the span of one minute.  The first way to interpret the results is that this represents an average of 111 bps.  Another way to interpret the results is that this actually represents 6672 bps because the network will probably see that traffic in one burst.

1,000,000 bits (1Mb) divided by 111 bits is about 8,993.  Over a 1Mbps circuit, 8,993 interfaces can be monitored.  Keep in mind that this would best case, and probably and is probably unrealistic.

With 6,672 bps divide into 1,000,000, we see that only 149 interfaces that can be monitored over a 1Mbps circuit. That would be worst case where all devices send their polling data in the same second.

150 Forty-eight  Port Switches per 1Mbps

So looking at this more in terms of the problem at hand, we could monitor somewhere between 179 and 3 48 port switches over a 1Mbps circuit if polling all interfaces. That being a pretty large range, I’m thinking that a safe realistic number will fall somewhere around 150. Also keep in mind that it is often not necessary to monitor all of the interfaces on a switch or device, which will increase the number of devices possible to monitor.

So some who read this might be think “Who is using 1Mbps circuits these days?” My answer is that there are not many organizations that do, but the ones who are will certainly care about being bandwidth conscious with them. The other more influential reason was because it was a fifteen minute exercise, and I’m lazy and did not feel like doing more math.