Slashdot Mirror
IEEE Sets Sights on 100G Ethernet
coondoggie writes to mention a Network World article about the IEEE's new 100G Ethernet initiative. The organizing body's High Speed Study Group has voted to try for the 100G standard over other ideas, like 40Gbps ethernet. From the article: "The IEEE will work to standardize 100G Ethernet over distances as far as 6 miles over single-mode fiber optic cabling and 328 feet over multimode fiber. With the approval to move to 100G Ethernet, the next step is to form a 100G Ethernet Task Force to study how to achieve a standard that is technically feasible and economically viable, says John D'Ambrosia, chair of the IEEE HSSG, and scientist of components technology at Force10 Networks." With video download services and interactive media becoming ever more the focus of internet startups, the organization is eager to offer a way to aggregate pipes in the coming years. The current thinking is that achieving these speeds will be reached by advancing bonding techniques for 10G signals over multiple fibers.
As it is, your average desktop will not handle anything even close to 100G Ethernet. At that point, your bottleneck is the PCI or PCI-X bus. As the bus has been one of the slowest PC components to innovate, I see these new, ultra-high speed Ethernet standards as only benefiting backbone providers, etc., for many years to come.
My blog
- From TFA.
Which is all well and good, but for honesty, I prefer Bill Watkins' take on it.
"Let's face it, we're not changing the world. We're building a product that helps people buy more crap - and watch porn."
Bill watkins, CEO of Seagate
328 feet - it's a good standard, but I like 100 metres better.
root of all...
Chicken and egg -
Once the connectivity is there, hardware will become available and gradually more accessible as it is taken up, same goes the other way - if someone suddenly comes up with a bus and card capable of even higher speeds, it will slowly become available and more accessible until connectivity catches up and everyone wants it. Its all about getting to the point were a (potential) mass market appears and it makes the R&D viable. In the short term you will obviously see niche markets for it anyway - and they will pay buckets of cash for this kind of tech because they see a benefit from it.
What I really want to see is higher uplink ports on SMB hardware.
Right now, if I want to make a medium size network using lower cost components, it might look something like 5- 24 port, 100-meg switches with 1 GB uplink to a big GB switch.
The bottleneck here is those uplinks. Each 100meg switch has plenty of backplane, and so does the gigabit switch, but those 100 meg 24 port switches have to share 1GB each to the backbone MDF.
So I really don't care about PCs or network cards or whatever, just give me 10GB links that I can use between switches without having to pay for overpriced Cisco crap.
I've had enough abrasive sigs. Kittens are cute and fuzzy.
Not exactly but Bell Labs did something like this in March http://www.lucent.com/press/0306/060308.coi.html
Hi I'm Progress, and I'm going to guess that we haven't met. I will be forever pushing forward with faster speeds. Thought you were happy with Gigabit over copper? LUDDITE! 10Gbit is enough for all your communication needs since you can xfer the library of congress 5 times a minute? THINK AGAIN! 100G Ethernet is the natural progression and before long you WILL want it. Trust me, I have been working this way for thousands of years. Glad we could get acquainted, now excuse me I need to get back to hiding from politicians.
You are right. Here are some numbers for the curious, nothing comes close to 100 Gbit/s:
PCIe x16 (2.5 Gbit/s per lane, 8B/10B encoding): 32.0 Gbit/s bidirectional (64.0 Gbit/s of aggregated bandwidth)
PCIe x8 (2.5 Gbit/s per lane, 8B/10B encoding): 16.0 Gbit/s bidirectional (32.0 Gbit/s of aggregated bandwidth)
PCIe x4 (2.5 Gbit/s per lane, 8B/10B encoding): 8.0 Gbit/s bidirectional (16.0 Gbit/s of aggregated bandwidth)
PCIe x1 (2.5 Gbit/s per lane, 8B/10B encoding): 2.0 Gbit/s bidirectional (4.0 Gbit/s of aggregated bandwidth)
PCI-X 2.0, 533 MHz, 64-bit: 34.13 Gbit/s
PCI-X 2.0, 266 MHz, 64-bit: 17.07 Gbit/s
PCI-X, 133 MHz, 64-bit: 8.53 Gbit/s
PCI, 66 MHz, 64-bit: 4.27 Gbit/s
PCI, 66 MHz, 32-bit: 2.13 Gbit/s
PCI, 33 MHz, 32-bit: 1.06 Gbit/s
However, regarding 10G ethernet adapters, does anyone know when vendors will start making use of PCIe x8 or x16 for them ? In all those Internet2 benchmarks papers, everybody complains about PCI-X beeing too slow, but PCIe x8 or x16 would be perfect for 10G.
Glad we could get acquainted, now excuse me I need to get back to hiding from politicians.
Too late, we found you.
I want peace on earth and goodwill toward man.
We are the United States Government! We don't do that sort of thing.
Strange. The standard is to be "six miles" over single mode, and "328 feet" over multimode.
/.ers are not so stupid as to have to be fed figures fudged for obscurity!
I don't get it!
I mean, we KNOW all decent standards use metric measurements - and Americans are inclined to convert them to the National Stupid System, so 328 feet makes sense (100 metres) - but where does this "6 miles" business come from? It is only 9,660 metres (9.66 km).
Surely the standard will be 10,000 metres - ten kilometres, and the poster was lazy, and couldn't be bothered with the extra 0.2 of a mile?
My question is this: when the specification is clearly based on very simple numbers: 100 metres and 10,000 metres - why convert that into the Stupid System?
How many escape pods are there? "NONE,SIR!" You counted them? "TWICE, SIR!"
Depends on which shelf.
Seriously, a lot of folks commenting on this news item seem to be convinved that all networks have only one node. Sorry, but I'm on a university, and I think that our interbuilding connections could really saturate a 10 Gb connection in the near future. It may be a long time before one PC can make use of a 100 Gb connection, but it won't be long at all before 1000 PC's can. Deployments will start the same way that 100 Mb and 1 Gb started. Backbone switches will move to the fastest speed, feeding workstations moving at slower speeds. Some specialised equipment will be available for systems which really need to actually move that kind of data to a particular node.