Testing in Heterogenous Environments
Regular Testing - Heterogenous Tuned Hosts
As in the above test case, testing between hosts that are of different capabilities (e.g. operating system, tuning, hardware configuration, connection to network) can produce different results.
We still would view the tests above as being "excellent" when we consider the environment:
- Different hardware age and capabilty
- Different operating systems and possibly different TCP tunings
- 20ms of Latency seperation
- Connected to different networks, crossing administrative domains of other networks
- A regularly configured TCP bandwidth test that runs the iperf tool for 20 seconds, and uses the "autotuning" feature of the Linux kernel (e.g. no set window size)
The disparity between directions could be related to the tuning of the NIC, and the ability of the host to serve data to the Kernel at a high rate. The drops in performance could be micro-bursts of congestion during a test. The test remains stable and consistent, and still reaches 40% of the available capacity, therefore it is not suspect as the source of a network problem.