Benchmarking MoCA: Multi-stream testing pushes more TCP data through the connection 'straw'

Two weeks ago, I provided a preamble to my planned benchmarking of MoCA (Multimedia over Coax Alliance) networking adapters versus alternative Gigabit Ethernet, 5 Ghz "wide channel" 802.11n, and "500 Mbps" HomePlug AV technologies, which included operational impressions of the Actiontec ECB2200 MoCA adapters used in the study. And last week, I published a suite of initial benchmark results, conducted by passing single streams of data between two endpoint computers using Ixia's IxChariot software, and via both TCP and UDP networking protocols. At the end of that writeup, I noted:

In past benchmarking projects, I've learned that a given networking technology is sometimes capable of larger aggregate bandwidth if I simultaneously transfer multiple data streams through it, versus only one stream. In next week's concluding post in this series, therefore, I'll share the results of a repeat of the above testing, this time sending four simultaneous data streams from one endpoint node to another, using both TCP and UDP and all four networking technologies.

You'll find those exact results below, beginning with TCP data transfers:

Technology	Node Path	Average Four-Stream TCP Throughput (Mbps)
MoCA (v1.1)	1->2	93.092
MoCA (v1.1)	2->1	93.004
HomePlug ("500" Mbps AV)	1->2	16.706
HomePlug ("500" Mbps AV)	2->1	28.995
Wi-Fi ("wide" channel 5 GHz 802.11n)	1->2	130.604
Wi-Fi ("wide" channel 5 GHz 802.11n)	2->1	152.313
GbE	1->2	857.892
GbE	2->1	653.996

In comparing the above table with its equivalent from last week's article, you'll notice notably higher aggregate TCP transfer bandwidth when using MoCA, Wi-Fi and Gigabit Ethernet on four-stream transfers, versus one-stream precursors. HomePlug AV, on the other hand, doesn't seem to benefit from stream multiplication. For more details, please peruse the full suite of downloadable test reports, in Ixia's native TST format as well as CSV, HTML, PDF, and TXT alternatives. As I mentioned last week, each group-of-reports cluster is named using the following convention:

ProtocolNumberOfDataStreamsTechnologyNodetoNode.Format

For example, "TCP4WiFi1to2.html" is the HTML version of the test report for the TCP protocol over 802.11n, with four simultaneous data streams originating at IxChariot endpoint node 1 and ending at node 2.

Next, let's look at UDP transfers:

Technology	Node Path	Average Four-Stream UDP Throughput (Mbps)
MoCA (v1.1)	1->2	95.638
MoCA (v1.1)	2->1	95.608
HomePlug ("500" Mbps AV)	1->2	28.927
HomePlug ("500" Mbps AV)	2->1	26.210
Wi-Fi ("wide" channel 5 GHz 802.11n)	1->2	69.365
Wi-Fi ("wide" channel 5 GHz 802.11n)	2->1	138.624
GbE	1->2	670.565
GbE	2->1	854.929

Here, there's no tangible benefit to four-stream transfers versus single-stream equivalents, which makes sense when you think about it. As I wrote last week, "the lack of TCP's handshaking ACK typically boosts UDP's data throughput capabilities." Also, "If you peruse the resultant test reports, you'll see abundant evidence of abundant packet loss, the result of an IxChariot quirk. The program polls the originating endpoint node's network adapter and subsequently sends out data at as fast a rate as the originating adapter claims it can support, regardless of the bandwidth capabilities (or not) of the destination endpoint and any intermediary hardware (such as MoCA-to-Ethernet or HomePlug AV-to-Ethernet adapters)." Translation: IxChariot shoved as much data down each networking technology's 'pipe' as possible, regardless of whether one or four streams of data were simultaneously being sent.

Stating the perhaps obvious from the above data, combined with last week's results, MoCA seems to be a robust networking approach. It doesn't require that you string CAT5e or CAT6 Ethernet cable around, is much faster than HomePlug AV powerline networking, and is not subject to the interference that can performance-plague both HomePlug AV and 802.11n Wi-Fi. However, MoCA isn't a perfect panacea. Coax topologies in homes are primarily a North America-centric phenomenon, and aren't pervasive even here. As I mentioned in my initial writeup in this series, my longstanding desire to test MoCA was delayed until I finally moved into a residence with the requisite cable runs. And, of course, MoCA also won't be useful if the room (or wall within a room) doesn't provide the necessary coax connectivity. Conversely, the AC wall outlets used by HomePlug AV are pervasive.

Also see:

• Benchmarking MoCA: Does coax meet your bandwidth quota?
• Benchmarking MoCA: Single-stream results suggest that HomePlug AV shouldn't worry MoCA one iota