Brian's Brain

This blog post references my cover story 'Home Transportation: Benchmarking Powerline, 802.11, and Ethernet' in EDN's August 2, 2007 edition. It's one of a series of web addendums to the print writeup.

As you read through my writeup, you may find yourself thinking 'but why didn't he test this' on at least an occasion-to-a-few. I'd encourage you to keep in mind the practical pagecount constraints of print, as well as the time and energy limitations of yours truly! With that said, I've come up with a list of additional experiments to pursue as my bandwidth allows. If you've got some extra time and motivation on your hands, I encourage you to liberally borrow from the following bullets; I'd welcome a public or private report once you've got results in-hand.

My first few topics are applicable to any of the networking technologies:

• You'll note that my Iperf testing was single-stream only. Maury Wright pointed out to me a few weeks ago that in his past powerline networking experiments, performance and reliability substantially degraded when he added a second simultaneous TCP or UPD transfer to the mix. I doubt this phenomenon is limited to powerline, and I'd like to see if Maury's stance still holds true.
• In noting that TCP and UDP transfer bandwidth varied depending on the predominant data flow direction, I commented, "Although I operated each computer as both a client and a server, I didn't swap the computers' locations, so it's difficult to determine how much of this disparity was from the computers themselves and how much was from intermediary equipment or other aspects of the link between them." A simple PC location exchange would more definitively identify the root source of the speed differential.
• I also wrote, "Dedicated software programs and entire Web sites exist to assist users in fine-tuning their network parameters, such as MTU (maximum-transmission unit), TCP-receive window, and TTL (time to live). However, I judged such tweaking to be beyond the capabilities and interests of most consumers. I also knew that just as LAN-tailored tweaking can degrade WAN performance, WAN-tailored tweaking can also degrade LAN performance. Therefore, I left these parameters at their Windows defaults." As an article I saw last night in Network World points out, Microsoft has implemented a long-overdue refresh of its TCP/IP stack for the Windows Vista (and Server 2008) generation, which supposedly precludes the need for tweaking and LAN-vs-WAN trading-off. It'd be interesting to put the theory into practice via a Windows XP-to-Vista migration to see if Microsoft's claims hold water.
• My article's introductory paragraphs pointed out the importance of both bandwidth and latency, yet I realize that I only tested the former parameter. Paraphrasing my college textbooks, "latency testing will be left as an exercise for the reader."

Next, for 1 GbE technology, you may have noted that I didn't enable Jumbo Frames support. It might be worthwhile to re-run my experiments with Jumbo Frames enabled, although from past experiments I know it would be challenging to find a Jumbo Frame setting that the PCs on both ends of the network chain, as well as all of the intermediary equipment, all support.

My next few suggestions apply to both powerline and wireless networks:

• Inject real-life interference into the environment, and see if (and if so, how much) the network experience degrades. For 2.4 GHz Wi-Fi, fire up the microwave oven, a Bluetooth headset, a cordless phone, or other nearby 802.11 access points broadcasting both on non-overlapping and overlapping channels. 5.8 GHz spectrum is comparatively uncrowded, but cordless phones still inhabit it, along with wireless headphones and speaker sets. And for powerline networking, any motor-driven piece of equipment can potentially be a power grid noise-inserting catalyst; air conditioner and heater blowers, fans, hair dryers and vacuum cleaners are all candidates. Speaking of interference, grab a shortwave radio and see if you can discern any corruption of amateur radio signal quality and strength by powerline networking equipment's EMI. If so, see whether or not enabling the powerline gear's notch filters has any effect.
• Vary the distance and complexity of the network link. Run Wi-Fi (in both 2.4 GHz and 5.8 GHz flavours) through a few walls, and across a few dozen more feet of distance. Test other power outlets, in other rooms, including ensuring that the powerline spur jumps across both phases of the 220V source within the intermediary breaker box. Etc....

For powerline networking:

• The DS2- and Intellon Home Plug AV-based equipment offers per-service QoS control. It'd be interesting to test how effective these settings are.
• Reportedly, attempting to simultaneously operate multiple '200 Mbps' powerline competitors (DS2/UPA, Intellon/HomePlug AV, and Panasonic/HD-PLC) within the same power grid disables them all. Is this true? Also, the NETGEAR DS2-based products offer an interesting 'compatibility' setting which supposedly allows them to cohabitate (albeit not communicate) with HomePlug 1.0 and HomePlug 1.0 Turbo equipment (which NETGEAR also offers). Is this true?

And finally, for the Wi-Fi gear:

• Evaluate the various encryption options, beyond the 64-bit WEP variant that I tested (i.e. disabled, 128-bit WEP, and various WPA alternatives) to see how they affect performance.
• Test wireless-to-wireless client connectivity, both via a router or access point intermediary (i.e. infrastructure mode) and in a direct (ad hoc) manner. For 802.11b and 802.11g, I have Ethernet bridges and USB adapters that'll enable me to continue using my desktop PC. For 802.11a, I'd likely need to switch from the desktop PC to a second in-house laptop (whose Wi-Fi transceiver also supports 802.11b/g). And for draft 802.11n, my testing suite is even more uncertain; my review unit of D-Link's Ubicom-based Ethernet bridge still hasn't arrived and is 5.8 GHz-only, although the Ubicom reference board I have supports both the 2.4 and 5.8 GHz bands. I also have draft 802.11n PC Cards from Buffalo (dual-band) and D-Link (2.4 GHz-only).

Any other next-step suggestions, folks?