7. Performance - more

Even more distressing than the large difference between the throughput number on the box and best-case actual performance, is that throughput fell off quickly and was between 7 and 10 Mbps in most other locations. Figure 12 shows the NETGEAR XE104 pair again, but for Location 3.

Figure 12: Location 3 NETGEAR throughput
(click image to enlarge)

I don't know what the deal is with the outlet used in Location 4, since it doesn't represent the farthest distance and is only about 10 feet away from the farthest Location 5 outlet. But it looks like Location 4 is as deadly for HomePlug networking as it is for wireless, since that's where most wireless products I test show worst-case performance, too. Figure 13 shows the difficult time that the Actiontec pair of devices had maintaining a connection there.

Figure 13: Actiontec Location 4 throughput
(click image to enlarge)

Since throughput fell off so quickly, I figured that something must be wrong with my test setup or methodology. So after rerunning tests and obtaining similar results, I fired off an email to Intellon for a sanity check of the results. However, Product Manager James Mentz essentially confirmed that my results were within the expected performance for HomePlug Turbo products, saying "If you go around a house and do a random outlet-to-outlet walk you should find that over 90% of the paths get at least 10 Mbps".

Mentz also provided data from HomePlug 1.0 (non-Turbo) field trials that said that technology can provide 1.5 Mbps of application-level UDP throughput 99% of the time and 4 Mbps of application-level UDP throughput 80% of the time. Note that UDP performance is usually faster than TCP/IP due to UDP lower protocol overhead.

By the way, the lower throughput of the SMC adapters was very repeatable. But when I substituted a NETGEAR XE104 for each adapter (one at a time), throughput jumped up to the range of the Actiontec and NETGEAR pairs. This is certainly curious, especially since the Diagnostic screen of the supplied utilities reported that all adapters were using the same Intellon firmware (version 1.5).