Ethernet's 400-Gigabit Challenge Is a Good Problem To Have 75
alphadogg writes "As it embarks on what's likely to be a long journey to its next big increase in speed, Ethernet is in some ways a victim of its own success. Years ago, birthing a new generation of Ethernet was relatively straightforward: Enterprises wanted faster LANs, vendors figured out ways to achieve that throughput and hashed out a standard, and IT shops bought the speed boost with their next computers and switches. Now it's more complicated, with carriers, Web 2.0 giants, cloud providers, and enterprises all looking for different speeds and interfaces, some more urgently than others. ... That's what the IEEE 802.3 400Gbps Study Group faces as it tries to write the next chapter in Ethernet's history. ... 'You have a lot of different people coming in to the study group,' said John D'Ambrosia, the group's chair, in an interview at the Ethernet Alliance's Technology Exploration Forum in Santa Clara, California, on Tuesday. That can make it harder to reach consensus, with 75 percent approval required to ratify a standard, he said."
Needs more context (Score:1, Interesting)
So whats the problem? Fitting more bandwidth onto a CAT5 cable? I feel like the summary needs more context.
Re: (Score:3, Insightful)
As I understand it, the problem is more like how to fit 400 1Gbps cables into a single wrapper.
That, and too many conflicting commercial interests.
Re:Needs more context (Score:5, Funny)
But we already have 10, 40 and 100gbps cables. Why 400x1?
Because interface bonding is like herding CATs.
Re: (Score:2)
Re: (Score:2)
10, 40, and 100gbps?
Didn't you mean 10, 40, and 100 Mbps? Or I missed the joke?
Re: (Score:1)
We're going to need a bigger box.
Re:Needs more context (Score:4, Interesting)
Nobody is really even doing 10gigE over cat5 (incl 6,7,8) for more than tiny patch cables. Fiber is fairly cheap.
Your sort of right, but completely wrong. The majority of the 10G ports previously sold are SFP based, but with 10G optical SFPs running ~$80 the vast majority of those ports have direct attach copper cables (CU/CR/CX1/ whatever you want to call it).
Because of this, the growth direction seems to be towards 10Gbase-T which can do 100 meters over cat6A driving the per port costs down significantly over SFP based solutions. Frankly, with the 10G uptake as a server interconnect, 40G is a good "low cost" switch interconnect/long haul medium. Again killing the need for 10G SFP solutions.
Re: (Score:1, Insightful)
CAT6 and CAT6A are not the same, with 6A its does 100M
Not sure what you were reading on wikipedia but both
http://en.wikipedia.org/wiki/Category_6_cable
and
http://en.wikipedia.org/wiki/10-gigabit_Ethernet
as does the IEEE spec.
Re: (Score:3)
You can count the contractors that do proper CAT6A wiring on one hand. You can barely get a contractor to do CAT5E correctly (without using wire nuts to keep 2 ends together), let alone CAT6/CAT6A/CAT7, heck I have a hard enough time buying decent quality copper CAT5. CAT6 cable comes in at ~$300/1000ft, CAT7 at ~$800/1000ft. Decent quality MM fiber comes in at ~$200/1000ft.
Re: (Score:3)
I am not seeing much uptake in 10GBASE-T in the data center (yet). People seem to be doing SFP+ copper in a rack (or 10Gbps backplane copper in blade systems), and SFP+ fiber between racks.
Re: (Score:3)
Fiber can be cheap, but usually not. We're not just talking about buying actual cables here, we need to be able to buy cable by the spool and terminate it ourselves. Fiber is a bitch to do that (you need gloves, special glue, a polishing pad, and a bunch of other things I can't think of at the moment, plus about 8 minutes of time per termination if you're fast at it, not to mention the possibility of getting glass stuck in your fingers) and doing it on a regular basis eats up your time, and time is money.
Tw
Re:Needs more context (Score:5, Insightful)
The problem is that different users have different requirements: Some will want low power requirements but don't need much range. Some will want more range but don't care much about power consumption and cost of advanced signal processing. Up to 1Gbps, Ethernet was a one-size-fits-all standard, mostly because everyone needed roughly the same: cheap, fast and uses existing cabling as much as possible (implying roughly the same range). Technological advances didn't require the kinds of tradeoff that are necessary now. From 10Gbps onward, Ethernet users have become more diverse and the technical challenges have forced more tradeoffs.
Re:Needs more context (Score:4, Informative)
At the 10gigE point, things diverge. There is single mode ("don't look at laser with remaining eye") media which is great for long distances, but more expensive, multi-mode which is good enough for inside the server room, and good ol' copper. However, this is what SFP modules are for.
It would be nice if fiber optic made it to the home, other than S/PDIF connections, and preferably with a more idiot-resistant connector than what existing fiber uses, especially with fouling lightpipes due to dust and such. Copper is useful, but eventually for faster connections, we will have to jump ship completely to fiber.
Of course, once we get 400Gbps, there will be the issues of how it filters down and all the switching/routing fabric needed. Most companies were dragged kicking and screaming to 1Gbps, and might use 10gigE for their trunk, or perhaps their SAN fabric. Trying to get them to 400Gbps for anything other than maybe storage will take a very long time.
Re: (Score:3)
What most homes have isn't copper. It's copperish, with traces of iron, tin, water, corrosion, bird droppings and dead rat. Those lines were not made for data. It's a wonder engineers have managed to cram bits down them as fast as they have with DSL.
Re: (Score:2)
Re: (Score:3, Insightful)
Do you mean Al and not Au (Gold)?
Re: (Score:3)
Aluminium isn't a terrible cable - it's got about 60% the conductivity of copper. The problem is joining it. That oxide layer means that any type of twist or post connection is going to make terrible contact. You have to solder it, and it doesn't take solder at all well.
Re: (Score:2)
Re: (Score:2)
That's because it isn't soldered. It wasn't installed for ADSL, an twist-connections are good enough for voice, so why spend time doing better? No-one at the time ancitipated it would be use for broadband signals.
Re: (Score:2)
Aluminium isn't a terrible cable - it's got about 60% the conductivity of copper. The problem is joining it. That oxide layer means that any type of twist or post connection is going to make terrible contact. You have to solder it, and it doesn't take solder at all well.
The solution in house wiring is to terminate your Al with a small length of Cu. They used aluminum in mobile homes for a while to save weight. I've seen it cause an outlet fire. The solution is to get some copper ends which get attached to the aluminum wiring and then sealed on. I've never actually seen the fix done, I think they're crimped on and then either protected with a compound or with an epoxy or glue. I wonder if the same couldn't be done for existing data wiring.
Re: (Score:2)
It would be nice if fiber optic made it to the home, other than S/PDIF connections, and preferably with a more idiot-resistant connector than what existing fiber uses, especially with fouling lightpipes due to dust and such. Copper is useful, but eventually for faster connections, we will have to jump ship completely to fiber.
I doubt fiber will ever make it in the home market aside from storage attachment. The only way to persuade a typical commodity user to plug anything in these days is if they can charge their battery of it. Will likely see penetration of PoE,PoE+,etc and 10GBase-T, but not much beyond that.
Re: (Score:2)
I doubt fiber will ever make it in the home market aside from storage attachment. The only way to persuade a typical commodity user to plug anything in these days is if they can charge their battery of it. Will likely see penetration of PoE,PoE+,etc and 10GBase-T, but not much beyond that.
I doubt anything has a future in the home, to the home it'll be fiber (23% here in Norway now and rapidly rising) that plugs into a box in the closet that splits it off into TV, phone, wireless and copper wire internet service and so on. GigE over copper is plenty for in-home distribution, even for compressed 4K material unless you've got a big family all watching different things with quad-BluRay quality. Anywhere you're likely to want an Ethernet port you have wall sockets, so no point in powered varietie
Re: (Score:2)
Anywhere you're likely to want an Ethernet port you have wall sockets, so no point in powered varieties.
I think here you may be underestimating the level to which the home consumer will finick over convenience. Having one universal power adaptor (an RJ-45 cable) lying around that any gadget can charge off with the added benefit of more reliable low-latency performance beats buying a bunch of wall warts for each device or carrying said wall wart around with you. I could see inductive pads competing, but not wall outlets.
400 GB is easy (Score:2)
400 GB per small* unit of time on the other hand ...
* "small unit of time" being 1 second or less
Re: (Score:2)
Gb*
*sorry
Re: (Score:2)
Patents (Score:5, Insightful)
The problem isn't that you have a bunch of squabbling engineers who can't even figure out how to split a lunch check. It's that you have a bunch of executives and attorneys that want to get as much of their company's IP piled into the standard as possible.
Re: (Score:1)
No, the main problem is that 40Gb/s laser modulation is expensive, anything above is not out of the labs yet.
And the 25Gb/s electrical modulation which is supported by FPGAs driving those lasers will run you into the 5-figures BOM cost in a blink.
At 400Gb/s today, you need 16 x 25G links (bi-di, that's 64 critical traces), which has to be done very carefully and also dictates powerful redundant cooling.
Internally, a 400Gb/s bus at a rate supported by general silicon is pretty wide (512b in ASIC, 1024 in FPG
Re: (Score:2)
We've got 100Gbps optics live on our network. Granted, even the Coherent stuff is broken down into two polarized carriers or two different wavelength carriers that are modulated at 50Gbps using QPSK. It's expensive, but it's cheaper than the equivalent number of 10Gbps circuits.
Re: (Score:2)
The more I'm going to have to pay for the privilege.
Pretty much false. I remember in the 90s a T1 costing ~$700 for the ISP, and ~$700 for the telco fees, and only giving 1.5 Mbps. Nowadays it is common for households to have 10x that speed down stream and roughly the same speed for upstream, for ~1/20th the cost.
Re: (Score:3)
not a valid comparison, since a business grade symertric medium of course has a premium A T1 compariable line still is $300 month, and plenty of businesses have them for their mission critical traffic. Company I last worked at had 40 Mbit Comcast business line with routed subet, but they still had the T1 backup for email and main web site and replication to offsite DNS. That Comcast line went down or flapped sometimes, but the T1 was always there
Re: (Score:2)
My DSL line happily uploads 1.5Mbps for under $30 per month...
Re: (Score:2)
Until the oversubscribed DSLAM starts dropping packets at 3PM. Jitter and latency will be worse as well. T1 is dedicated bandwidth to the router or switch.
Re: (Score:2)
consumer grade, no service guarentees and not 100% reliable, that's the big difference.
I see other similar comments, it's like when people see the specs on the computer controlling a space rover and they say "but my desktop is so much faster!"
Re: (Score:2)
that's not normal, cable and DSL are very shaky compared to typical T1
Re: (Score:1)
Re: (Score:2)
Our cable network drops 2-3x per month and we have no better options. But it's about 30-35Mbps inbound and 3-5Mbps outbound, so far better bandwidth then the T1.
So we use both. The T1 is the fallback line for the cable internet and traffic automatically shifts from the cable line to the T1. It might take 2-3 minutes for the shift to happen, but its automatic and is better then a 30-240 minute downtime.
Re: (Score:2)
Yeah but you get E10 (10Mbps up/down) for less than $800/month. That's 10 times the speed for the same price at T1s in the 90s. T1s are still expensive because of their large overhead and expensive equipment.
Re: (Score:2)
E10? in the UK for ITU-T they have E1 through E4.....we're talking about business grade time division multiplex carrier lines, not DSL or cable or other consumer grade shakier and less reliable tech
Re: (Score:2)
E10? in the UK for ITU-T they have E1 through E4.....we're talking about business grade time division multiplex carrier lines, not DSL or cable or other consumer grade shakier and less reliable tech
I imagine 'E10' there is a reference to 10 Mbps metro Ethernet, something like the Ethernet in the First Mile approach. There's nothing inherently "consumer grade" about DSL itself: indeed, even E1 "leased lines" get delivered over HDSL or similar in some cases. Unlike cable, which is contended and prone to collisions, DSL gives you a constant bitrate (unless configured to vary to squeeze higher bitrates when line quality permits) point to point link, just like a conventional leased line - all the performan
Re: (Score:2)
not a valid comparison, since a business grade symertric medium of course has a premium A T1 compariable line still is $300 month, and plenty of businesses have them for their mission critical traffic. Company I last worked at had 40 Mbit Comcast business line with routed subet, but they still had the T1 backup for email and main web site and replication to offsite DNS. That Comcast line went down or flapped sometimes, but the T1 was always there
Most businesses I've seen are moving to MPLS. We have a T3 we're trying to migrate away from as we can replace it with 2 50mbit mplss and make a 70% saving, and replace the kit with something that is actually supported.
One problem we're having is we have 2 E1 lines on the barer, but you can get E1 over IP converters which look promising.
Re: (Score:2)
MPLS works over a number of carrier technologies including T1, E1, ATM, Frame Relay, and DSL.
Re: (Score:2)
MPLS works over a number of carrier technologies including T1, E1, ATM, Frame Relay, and DSL.
They do (and other technologies. I wonder if you can do MPLS over MPLS. I have 3 days put aside next month for some fiddling with MPLS and VPLS), but as all I get is an ethernet handoff on a piece of fibre, it means that old inefficient technologies like T1/DS3/etc can be bypassed.
From my point of view, I pay $xxx per month, and get a cable in my equipment room, the fibre heads off to somewhere, the ethernet packets are tagged with MPLS and shifted quickly and reliably across multiple high bandwidth links (
Re: (Score:2)
Yes, you can get a cheap 10GbE switch. No, it doesn't have nearly the management that a Cisco or HP switch has.
If you are using one or two, you probably don't care. If you are using one or two thousand of them, you VERY much care.
Microsoft (Score:2)
Before you let anyone into your standards committee, make sure they don't work for Microsoft or a Microsoft affiliate. And if they do, make sure Microsoft isn't trying to push through a competing "standard".
Be sure you learn the ISO's lessons regarding Microsoft and its henchmen stuffing standards bodies.
Re: (Score:2)
Sorta' like Micron did jamming RAMBUS-patented tech into the DDR-something definition?