Networking Over Powerline: My Next Benchmarking Project Is Coming Along (Sorta) Fine
As followup to my post-CES powerline networking ‘State Of The Union’ summary, I wanted to let y’all know that my next cover story, in EDN’s June 23 edition, will be a hands-on piece devoted to the topic (since my boss concurs with my networking IC editorial interest). Here’s the promotional blurb that I wrote up on it a couple of weeks back:
Back in mid-2007, EDN Senior Technical Editor Brian Dipert exhaustively tested (among other networking technologies) three contending powerline approaches; Panasonic’s HD-PLC, Intellon-championed HomePlug AV (along with its standards-based and -augmented predecessors), and DS2-led UPA. However, a lot’s changed in less than four years:
- Intellon was acquired by Atheros
- DS2 was acquired by Marvell
- Other notable acquisitions and consolidations of powerline networking pioneers included Arkados (by STMicroelectronics), CopperGate (by Sigma Designs) and Gigle (by Broadcom)
- HomePlug AV further matured, and combined with HD-PLC to form the IEEE’s 1901 standard, and
- UPN was merged into the ITU’s G.hn standard, where DS2-now-Marvell was joined by Infineon spinoff Lantiq, by Sigma Designs (who’s striving to support both G.hn and 1901 via a common chipset) and other suppliers
As such, it’s high time for EDN to give powerline networking another look. With HomePlug AV-, 1901-, and G.hn-supportive hardware in hand from multiple vendors, Dipert will subject each technology to a battery of performance and reliability tests, with his home office’s power grid as the laboratory, and employing both synthetic and real-world benchmark schemes.
Seem straightforward? Whereas my mid-2007 tests focused on a variety of networking technologies (powerline, Cat5, and various Wi-Fi flavors), this time I’ll have a more in-depth, powerline-centric focus. For testing conventional HomePlug AV, I’ll be using NETGEAR’s single-port XAV2001 and XAV1004 four-port adapters, both based on Intellon-now-Atheros’ INT6400 MAC/PHY and INT1400 AFE/Line Driver, which have formed the foundation of my home office’s powerline topology for around a year.
To be clear, given the laundry list of optional features included in the IEEE 1901 specification, manufacturers can label conventional HomePlug AV adapters as IEEE 1901-compatible. However, I’ve got finally two manufacturers’ sets of extended-function adapters on hand, both based on Atheros’ follow-on AR7400 MAC/PHY and AR1500 AFE/Line Driver combo that (among other things) stretches HomePlug AV’s conventional 2-to-28 MHz operating frequency range to 50 Mhz.
First is TRENDnet’s two-adapter TPL-401E2K kit, two of which I possess, and which Atheros subsequently firmware-upgraded for me (along with appending a fifth reference-design adapter to round out my testing suite) to hopefully at least somewhat alleviate issues uncovered during others’ initial testing. However, NETGEAR finally also came through a couple of days ago with three sets of two XAV5001 single-port adapters (three samples of the XAV5004 four-port successor to the XAV1004 will supposedly be headed my way in mid-April, too). The XAV5001 is based on a unique hardware design that, in SmallNetBuilder’s testing, outperformed the TPL-401E even when both manufacturers’ adapters were running identical Atheros-supplied firmware releases.
What about G.hn? When I met with Lantiq back in early February, company representatives claimed that first silicon samples of its XWAY HNX transceivers introduced one month earlier would be available right about now. They showed me customers’ adapter designs that were completed and awaiting ICs to drop in them, so I’ll circle back and will hopefully have hardware in hand within a few weeks.
And then there’s Sigma Designs’ CopperGate group, whose G.hn/IEEE 1901 combo chips I wrote about late last October and who I also met with at CES. Back then, I was told to check back at the end of February for stable-firmware sample units. I’ve been remiss, but I hope that the schedules haven’t slipped too much and that Sigma Designs will also be able to participate. Stay tuned.
Last but not least, I want to say a few words about my testbench plans. For my mid-2007 study, as you may recall, I leveraged Iperf. The utility’s in a bit of flux, however; latest-generation v2.0.5 dates from mid-last summer and further development has seemingly been disbanded, while its v3 backwards-incompatible successor is still in beta. Anyway, I can’t seem to find an already-compiled Mac OS X version of v2.0.5 anywhere; granted I could do it myself, but I’d frankly prefer to not bother. And I concur with the folks at SmallNetBuilder:
Even though all the previous options are much easier to use, iperf and jperf still have their fans. Doug Reid did a good job of describing both in his iperf and jperf articles, so I’ll just point you there for the details.
Be forewarned that the defaults of both programs can seriously understate what your network is really capable of. There are lots of network parameter knobs you can twiddle, which is a strength to some, but has always put me off using them.
In fact, you might recall that I went back and reran my 2007 TCP tests at the suggestion of then-Intellon, with altered buffer length and window size settings. So I thought I’d give Ixia’s network analysis products, which I’d repeatedly noticed other publications used in their studies, a whirl instead. I requested an editorial copy of IxChariot, and the company’s marketing rep’s response indicated that such an arrangement would be possible…but the terms of the agreement were frankly too onerous for my journalistic ethics to stomach:
We actually have a program specifically tailored to IxChariot for use in editorial reviews. I’ve attached the agreement so that you can see the details involved. Basically it is Ixia agreeing to give you one 50 node license to use for the next year for your testing that is involved in your article writing. In return there are certain attribution requirements that we expect in the article in the areas describing the tools used and the setup of the test case.
What were the attribution requirements that the company expected? The form Ixia sent for my hoped-for signature included this phrase:
You agree that the terms and conditions in this letter agreement are confidential and, except as otherwise required by applicable law, will not be disclosed to any third party without Ixia’s prior written consent.
Even though I have no intention of signing the letter, thereby entering into the agreement, I might therefore still be on shaky legal ground if I quoted the requirements verbatim. Suffice it to say, though, that Ixia insisted on prominent and favorable mention of itself and its product, in multiple manifestations, complete with company-supplied logos and other artwork. Early (pre-publication) access to my editorial output for review (and presumed influence) was also alluded to in the Ixia-supplied documentation, albeit not explicitly insisted on. Nonetheless, thanks but no thanks.
Instead, I’ll be using the company’s free QCheck utility, which like IxChariot is Windows-only. QCheck doesn’t enable me to simultaneously generate multiple network streams, but I can approximate this capability via other means, and in my testing so far QCHeck otherwise seems to do everything I need. I’ll run it on my corporate laptop, a Dell Latitude D420 based on Windows XP, and on the other side of the network link will be my MacBook, running the Mac OS X version of Ixia’s Endpoint. Both the Latitude D420 and MacBook contain GbE wired Ethernet ports, necessary for logging hoped-for >100 Mbps powerline network adapter speeds, and their portability will make various-network-node testing easier than if I’d instead rely on desktop computer systems.