Raritan Computer COMMANDCENTER NOC - Manual (Page 151)

APPENDIX G : NETWORK TRAFFIC OVERHEAD : NETWORK MANAGEMENT ’ S NECESSARY EVIL 139
Appendix G: Network Traffic Overhead: Network
Management’s Necessary Evil
On five-minute intervals, the CC-NOC polls services on managed nodes using Raritan's 'synthetic
transactions'. These transactions serve to better test the service's availability, as they actually
exercise the service, as opposed to simply “pinging” the box, making the leap of faith that the
services you rely on are still responding appropriately.
It's important to note that Raritan, throughout the initial development of our product, went to great
lengths to gather as much valuable information as possible without unnecessarily impacting the
network. Some overhead is necessary, but between load-leveled polling, spreading polls out over
time, and configurable concurrency, only a limited number of devices are allowed to be polled
simultaneously, the overhead appears as more of a constant "hum" in the background, as opposed
to the regular, significant spikes you may see generated by other network management tools.
On an arbitrary box, we measured the traffic generated by four different CC-NOC poll types:
• ICMP pings
• TCP socket reachability (used for monitoring database listeners)
• HTTP synthetic transaction
• SNMP data collection
As each poll happens on five-minute intervals, we'll use 300 seconds as the denominator in
calculating average bandwidth impacts. We'll also include the actual time it took to complete the
poll.
ICMP Pings
Raritan considers ICMP a service provided an interface. As such, we discover and monitor that
service independently. We also use the availability via ICMP as our "lowest common
denominator" in determining if a service outage is actually symptomatic of an interface or node
outage.
In the case of this arbitrarily chosen node:
ICMP Ping issued: 90 bytes (720 bits)
ICMP Ping response: 90 bytes (720 bits)
Total Traffic: 180 bytes (1440
bits)
Transaction time:
.000057 seconds
Average bandwidth: 4.8 bps
% of 10Mbps .00000048%
Ethernet:
% of 100Mbps .000000048%
Ethernet:
TCP Socket Reachability
As a test of a services ability to accept TCP session requests, one synthetic transaction type we
use, predominantly for database connectivity testing, is that of TCP socket connects. For those
versed in the protocol, this is the standard SYN-SYN/A