Who Says An AC Router Needs Gigabit Ports?

[Terry Kennedy]

As I understand it, Gigabit Ethernet is mature enough that it simply "comes with the silicon" on modern chips. While an OEM can certainly still buy 10/100 (only) single- and perhaps dual-port chips, most of the chips with more ports (such as the switch-on-a-chip parts used in this class of router) support Gigabit already.

When a product uses a a Gigabit-capable chip but doesn't offer Gigabit to the end user, that's a marketing decision. Either to force the user to buy the "next model up" to use the Gigabit functionality that's present but disabled in the lower model, or to save a few pennies by not using a 10/100/1000 PHY (in the fewer and fewer cases where a separate PHY is needed at all), or a couple fraction-of-a-penny passive components.

As a prime example, I give you the Cisco 2811 (and lower) vs. the 2821 (and higher):

2811#sh controllers
Interface FastEthernet0/0
Hardware is MV96340

2821#sh controllers
Interface GigabitEthernet0/0
Hardware is MV96340

Notice anything fishy?

Also, jumbo frame support is often tied to Gigabit support. Jumbos can sometimes provide a huge performance benefit, though the amount of difference (if any) depends on the protocols being used, etc.

None of this diminishes the point the article raises about many wireless routers having a cable modem as their only wired device.

However, simply using Gigabit-capable hardware (or enabling it if it is present but off for marketing reasons) would make that same product more attractive to additional customers.

I had a very difficult time finding a wireless router which had 8 Gigabit ports (1 upstream, 7 attached devices) for a small business where I was replacing a cascaded stack of 4-port 10/100 switch/hub things. I found the WD My Net N900 (which I eventually ended up using), but it has a number of software defects which make it problematic. A wider choice of units would be greatly appreciated by me.

In N years time (no pun intended), will we be having this same discussion about 10 Gigabit ports? To turn that on its head, when was the last time you saw (or needed) an AUI port? Or a MII port?

 

[stevech]

It's rare to find a WiFi router AND IT'S CLIENT DEVICE that can, in real use conditions, sustain over 100Mbps at the IP layer. Thus, a 10/100 switch port is fine for all but a few cases.

With 1000BT switch chips cheap now, it may be that using that in a WiFi router doesn't add as much cost as it used do. When prices were higher, we used separate gigE switches to interconnect wired PCs on the LAN.

Unfortunately, the typical consumer doesn't know that xxx Mbps at the WiFi air link signal level doesn't at all correspond to the IP layer bit rate, due to the nature of 802.11 with its overhead, lack of full duplex, etc.

 

[alphasparc]

I have to agree with the original poster.
Most of the time the only reason they put in FE switch instead of GE is product differentiation.
Companies which do this are usually those big brands with a big marketing department.
GE switches are dirt cheap nowadays, you can find them in many "value for money" brand at a price even lower than those 'premier' brands.
And the most important reason for GE nowadays is fibre.
With everyone going fibre, rates are going to exceed 100Mbps no point releasing FE products anymore.
Funny how a reviewer can get bought over by marketing bollocks...

 

[Terry Kennedy]

It's rare to find a WiFi router AND IT'S CLIENT DEVICE that can, in real use conditions, sustain over 100Mbps at the IP layer. Thus, a 10/100 switch port is fine for all but a few cases.

Do you mean a wireless client? If so, I agree.

For connected (wired) devices, the "IP layer" part is mostly irrelevant - on-site traffic is going through the switch-on-a-chip and never being routed. With the site I installed the My Net N900 at, they have a modern Synology and Optiplex systems which can definitely share files at faster than the 100Mbit/second (12.5Mbyte/sec, give-or-take).

I've collected hardware from the scrap pile at a colo facility I'm in which does 9.89Gbit/sec (basically wire speed) over a single TCP connection using 10GbE.

With 1000BT switch chips cheap now, it may be that using that in a WiFi router doesn't add as much cost as it used do. When prices were higher, we used separate gigE switches to interconnect wired PCs on the LAN.

Now that many router / switch manufacturers have moved to RJ45 jacks with integrated magnetics for EMI compliance, there isn't a lot between the chip and the jack. A quick search for '10/100/1000 integrated phy' turns up things like the BCM53115, which is a 5-port switch chip which has advanced features such as VLANs, Q-in-Q, built-in TDR, etc. Pricing information isn't available, as they sell by quote only in large batches. They claim it is competitive, though, and position it for home / SOHO applications.

Unfortunately, the typical consumer doesn't know that xxx Mbps at the WiFi air link signal level doesn't at all correspond to the IP layer bit rate, due to the nature of 802.11 with its overhead, lack of full duplex, etc.

Agreed 100%.

 

[RMerlin]

For me personally, having Gigabit at home is important. Makes it much faster when I backup my laptop or my HTPC to my desktop using Acronis True Image or Clonezilla, or when moving files between my HTPC and my desktop (which doubles as my server). And all of these being wired, I have no problem hitting well over 100 Mbits, so FE would definitely be a bottleneck for me.

However, I think the point in Tim's article wasn't about whether Gigabit is useful or not, but more that there is a tendency these days for people to go nearly 100% wireless in their home. For these, it makes little sense in having a Gigabit capable router. Until they decide to plug a NAS on their network.

Another thing however that manufacturers might have missed: ISPs providing Internet connection faster than 100 Mbits are starting to become more common. That means you'll need a Gigabit WAN port, and the need for that will increase in the next few years. Which means some people who bought a FE-only router today might have to buy a new router in 2-3 years when they upgrade. Which means increased sales for manufacturers...

...so maybe after all manufacturers did NOT miss that fact that it will make their product become obsolete faster.

 

[thiggins]

Thanks for the comments. A few points in response:

- My contact was talking about the U.S. market, which according to Akamai's Q1 2013 State of the Internet report, has only 25% of connections > 10 Mbps. The report also states global average connection speed at 3.1 Mbps and Global Average peak connection speed at 18.4 Mbps. So, no, "everyone" is not going fiber and certainly not in the US.

- Product manufacturing cost / margin isn't the driving factor. Of course manufacturers limit features to establish price points. This isn't unique to wireless router design. Why give something away when you can charge for it?

However, I think the point in Tim's article wasn't about whether Gigabit is useful or not, but more that there is a tendency these days for people to go nearly 100% wireless in their home.

That is precisely the main point of the article.

 

[stevech]

In my response, above, I attempted to day that it's rare to get > 100Mbps net IP layer yield on WiFi, even though it might use a 300Mbps raw bit rate on the air link frames. So a 100BT switch in a WiFi router is good enough for most all WiFi clients.

Wired clients, often PCs with gigabit ports, should use a separate switch or a WiFi router with a built-in gigabit switch. Either one is fine. And gigabit switches are amazingly low cost now. And, a dedicated outboard switch can simplify cable routing and physical access.

 

[Nerre]

The point is that even if you buy a router with WiFi you might also have wired clients (that's why the router has LAN ports!).

In our house we have at least 6 computers that have gigabit network interface (though 4 of them are HTPC related, one MythTV backend and three frontends).

We have an "open" fiber internet conenction (we can choose from a handfull of ISPs) so we will probably have gigabit internet in a few years (today we have 100/10 only because three years are included in the fiber installation cost, then I will probably upgrade to 100/100 "just because").

So gigabit WAN and LAN ports on the router is in no way overkill for us, I even got out and bought a new gigabit switch to get gigabit speed for our MythTV machines.

Now, we don't even use the WiFi of the RT-AC66U because it's in the basement, we have another old router acting access point and it only has 100 Mbit ports, but also its WiFi isn't faster than 54 Mbit. But on the other hand I don't think any of our WiFi clients is capable of higher speeds than that.

So, just because we don't use high speeds on WiFi doesn't mean we don't need high speed on wired connections. We use wired connections where we need the speed, and WiFi where we need the mobility.

 

[thiggins]

So, just because we don't use high speeds on WiFi doesn't mean we don't need high speed on wired connections. We use wired connections where we need the speed, and WiFi where we need the mobility.

Sure. That's why there are routers with Gigabit ports. One size does not fit all.

 

[Terry Kennedy]

Product manufacturing cost / margin isn't the driving factor. Of course manufacturers limit features to establish price points. This isn't unique to wireless router design. Why give something away when you can charge for it?

It seems (looking at, for example, the wireless router + 8 Gigabit ports) that manufacturers simply don't have products in that space. Are there any others besides the My Net N900, for example? The same is probably true of other combinations of hardware. If a manufacturer doesn't have a product with the more advanced feature(s), what are they upselling to? Their competitor's product?

My point is that the hardware is often in the box already. This isn't limited to Ethernet components - there's another thread here about a router that has a dual-core CPU and 256MB, but only 1 core and 128MB shows up.

As long as manufacturers continue to use GPL'd code (this is often Linux) as the base for their firmware, people will create independent firmware for the hardware if there's enough interest to justify the development. Even if features are locked out so strongly that 3rd-party firmware can't use them, the fact that the features are there but disabled will become widely known.

The alternative would be for manufacturers to admit that it is simply market positioning and come up with something like "This product has 4 10/100 ports enabled. If you wish to use 5 or more ports or if you want Gigabit functionality, please order license key <part number> for <price>." I predict that won't go over well, either.

 

[Terry Kennedy]

By the way, I'd be the first to admit I'm not in the market that home / SOHO manufacturers are targeting - with 128TB of NAS in my spare dining room, a 24-port 10GbE switch (and a pair of Catalyst 4948-10GE's) and a 2821 router, I'm definitely not the target demographic. The 4 Aironet 1250 access points just emphasize the fact.

However, that's my house. My response was based on the requirements I see from various people I consult for (paid, or friends and family mooching). [Digression - while there are lots of "No, I won't fix your computer" shirts, I've never seen a "No, I won't fix your network" one. I think there's an opportunity here, hint hint.]

The small business-and-smaller section of this ranges from somebody with a PC and a network-capable printer (in an area served by community/public WiFi) to a small business with 3 wired PCs, a NAS, a couple of network printers, the occasional WiFi or two, and a pair of phone lines on a Vonage box. I'd say that well over 2/3 of this portion of my work has a requirement for at least some Gigabit ports. Over half is also space- or outlet-constrained and has a strong preference for an integrated router / WiFi / switch solution.

 

[sfx2000]

The article was interesting in that the observation "by someone in the business, not NETGEAR" is fairly valid.

Not everyone's cup of tea - but adding GIGe is more than just the SOC - it''s the magnetics and other items, and a hell of a lot more QA/Support costs - let's say 5 bucks at the ODM/OEM level amortized across 1M unit sales - and this is non-recurring engineering costs alone - the conversion costs drive it a bit higher from there on a per-unit basis.

So for the budget AC1200 market - the approach is reasonable at the $69.95 USD pricepoint - want more speed on the wire, spend more money.

What the real tell is adding 2nd Gen 801.11ac support - this is about 30 dollar hit on 802.11n single band with current chipsets/radios. Some of that is the silicon on the SOC, and part of that is the 5GHz band support.

Just saying...

sfx

 

[Terry Kennedy]

Not everyone's cup of tea - but adding GIGe is more than just the SOC - it''s the magnetics and other items, and a hell of a lot more QA/Support costs - let's say 5 bucks at the ODM/OEM level amortized across 1M unit sales - and this is non-recurring engineering costs alone - the conversion costs drive it a bit higher from there on a per-unit basis.

As I believe I mentioned earlier, the parts that aren't moving into the chip are moving into the jack. Magnetics, LEDs, etc. are all available in jacks now.

Not that there was a lot to start with - I just checked a 7 year old 8-port switch design with a BCM5398 switch-on-a-chip. The only stuff in the path between the chip and the connectors is the magnetics (4 2-port Bi-Tek FM-3178LLF modules) and one resistor and capacitor per twisted pair.

So for the budget AC1200 market - the approach is reasonable at the $69.95 USD pricepoint - want more speed on the wire, spend more money.

The My Net N900 I keep going on about (which I know is not 801.11ac seems to still be manufactured (either that or there's a warehouse full of tens of thousands of them) shows up for $44.99 at Newegg on a regular basis ($49.99 minus $5 "limited time" discount). If it wasn't Sunday night I'd go check what Tech Data is selling them for - I'm sure TD isn't taking a loss on each one. The WD store charges full price and it is only $99.99 there.

The list price progression per the WD store is:

My Net N600 - 4x 10/100, 300+300 - $59.99
My Net N750 - 4x 10/100/1000, 300+450 - $89.99
My Net N900 - 7x 10/100/1000, 450+450 - $99.99

So you get 3 more Gigabit ports and one more radio stream for $10 (retail).

Anybody have any comparisons between otherwise-identical products with 10/100 vs. 10/100/1000? The above list has too many other variables in it to draw a useful conclusion.

 

[Razor512]

Agreed, it is horrible that some companies would cripple a router through software.

 

[alphasparc]

My Net N600 - 4x 10/100, 300+300 - $59.99
My Net N750 - 4x 10/100/1000, 300+450 - $89.99
My Net N900 - 7x 10/100/1000, 450+450 - $99.99

Not really a valid comparison N750 uses Qualcomm Atheros, N900 uses Ubicom.

 

[Terry Kennedy]

Agreed, it is horrible that some companies would cripple a router through software.

Well, they can save a few pennies by only using 10/100 magnetics. The point is that it should be a very small amount, even when multiplied by the number of units sold.

Not really a valid comparison N750 uses Qualcomm Atheros, N900 uses Ubicom.

Exactly. That's why I asked "Anybody have any comparisons between otherwise-identical products with 10/100 vs. 10/100/1000?"

 

[thiggins]

It's all a question of sales potential. The fact the WD router is so heavily discounted tells us that demand is low.

 

[Terry Kennedy]

It's all a question of sales potential. The fact the WD router is so heavily discounted tells us that demand is low.

Or that WD has decided "Why are we in this market segment? We make none of the components of this product!" I assume they got in because they wanted to sell a version integrated with one of their disk drives. It may be that they're just emptying a [big] warehouse.

The poor reviews probably have something to do with it as well.

So, let's look at popular models - are there any two that are identical other than 10/100 instead of 10/100/1000?

 

[alphasparc]

Exactly. That's why I asked "Anybody have any comparisons between otherwise-identical products with 10/100 vs. 10/100/1000?"

Yes I do.
TP-Link WDR3500 vs TP-LINK WDR3600.
Exactly the same lest the AR8327N Gigabit switch on the WDR3600.
$55 vs $63 on newegg

 

[Terry Kennedy]

Yes I do.
TP-Link WDR3500 vs TP-LINK WDR3600.
Exactly the same lest the AR8327N Gigabit switch on the WDR3600.
$55 vs $63 on newegg

Looks like the 3600 also has an additional USB port. I see pictures of the internals in the review of the 3500, but I don't see a review of the 3600 to compare with.

I would have guessed that the 3500 was an older design, as it uses a model of pulse transformers introduced in 2010, if not earlier. Is the 3600 positioned as a successor product or as an upsell? Amazon shows the 3500 first carried there in September 2012 and the 3600 in August 2012, which would make the 3600 the newer of the two?

Amazon says the 3500 is the #5286 best-seller in electronics and doesn't even rate in their router category, while the 3600 is #1050 in electronics and #43 in routers. Given that they were both added around a year ago, I'd say there is a clear preference for the 3600, at least among Amazon's customers.

In addition to (presumably) a different circuit board, the units should have different back panels (due to the 1 vs. 2 USB ports). I'm not sure if the panel is part of the overall case molding or a separate drop-in piece.

I'll assume for the sake of argument that they order enough of the cases that ordering twice as many (to have one model instead of 2) doesn't drop the per-unit cost of the case, and the mold / setup charges are spread over enough units that the extra one also doesn't affect the price they're paying for the cases.

But there's the development and certification of two sets of hardware, plus maintaining two different sets of firmware. I suspect that the firmware for both models is built from a common source tree (based on identical release dates and version numbers; the firmware itself is opaque binary blobs) but it's still 2 different compiles, 2 sets of regression tests, etc. Not to mention they're also maintaining 2 different builds of "Easy Setup Assistant" for the 2 models.

It would appear that the increased costs involved with the stuff in the preceding paragraph cost them less* than the revenue they're making by selling 3600's instead of 3500's. It would be interesting to know how close the two different profit projections are, particularly once the costs of continuing to support the extra model for a full 5 years are factored in. Of course, up-front profit is "a bird in hand", and also either represents investment capital or money that they won't have to borrow to fund future projects.

* There's always the possibility that they discovered this was a bad idea and they're actually losing money compared to giving everyone a 3600 at the 3500 price (or a price halfway between the 3500 and 3600), and that they'll never do this again. But we'd never hear about that as they'd put on happy faces and say everything is fine (particularly if they have a warehouse full of unsold units).