IPOP Network Measurements Methodology
From BoykinWiki
The page aims at describing IPOP measurement techniques.
Contents |
Machine Setup Specifications
The measurements shown here are taken from two grid appliances running on VMware Server 1.0.3. One appliance ran on c8node11.acis.ufl.edu, while the other on c8node10.acis.ufl.edu.
Here is the configuration for the Cluster 8 nodes (c8node1 - c8node 14)
IBM BladeCenter with 2 gigabit switch modules 14 HS21 blades each with two Xeon (Dual-Core Woodcrest) 2.33GHz processors with 4MB L2 cache 8GB RAM 73GB SAS disk
Iperf Configurations
Before running the Iperf tests, I made sure that the machines had a chota (direct-shortcut) connection. I pinged the two machines (100 pings) which gave an average latency of 0.719 ms.
I then ran the following four tests.
Test One - Iperf TCP test without IPSec
Iperf command on server side:
iperf -s
Iperf command on client side:
iperf -c 242.188.8.37 -i 1 -f k -t 100
Results:
51.5 Mbits/sec
Test Two - Iperf TCP test with IPsec
Iperf command on server side:
iperf -s
Iperf command on client side:
iperf -c 242.188.8.37 -i 1 -f k -t 100
Results:
33.0 Mbits/sec
Test Three - Iperf UDP test without IPsec
Iperf command on server side:
iperf -s -u
Iperf command on client side:
iperf -c 242.188.8.37 -u -i -f k -b100M -t 100
Results:
45.4 Mbits/sec
Test Four - Iperf UDP test with IPsec
Iperf command on server side:
iperf -s -u
Iperf command on client side:
iperf -c 242.188.8.37 -u -i -f k -b100M -t 100
Results:
32.0 Mbits/sec
Conclusion
Overall, in TCP there is a 36 percent overhead by using IPSec. In UDP, there is a 30 percent overhead by using IPSec.
