IEEE Releases 802.3ba Standard

An anonymous reader writes "EEE announced the ratification of IEEE 802.3ba, a new standard governing 40Gbps and 100Gbps Ethernet operations. An amendment to the IEEE 802.3 Ethernet standard, IEEE 802.3ba, the first standard ever to simultaneously specify two new Ethernet speeds, paves the way for the next generation of high-rate server connectivity and core switching. The new standard will act as the catalyst needed for unlocking innovation across the greater Ethernet ecosystem. IEEE 802.3ba is expected to trigger further expansion of the 40 Gigabit and 100 Gigabit Ethernet family of technologies by driving new development efforts, as well as providing new aggregation speeds that will enable 10Gbps Ethernet network deployments."

Comments

IEEE 802.3ba code name

[geekoid]

"Pity da fool"

Re:IEEE 802.3ba code name

[cmg]

Unfortunately there is no corresponding 100000Base-MrT

Can't wait to see how my favorite vendor...

[alfredos]

...Has to use multiples of the currency unit to represent the price of the first units.

Stuck

[Scutter]

You'll still be stuck on 3Mb/512kb DSL.

Re:Stuck

[Monkeedude1212]

You know networking exists outside of the internet, right?

Re:Stuck

[masshuu]

I think this is geared for big companies LANS and Campus LANS and Datacenters as an alternative to short runs of fiber

Re:Stuck

[sortius_nod]

Tell that to my router... 20Mbps/1.3Mbps

Re:Stuck

[ePhil_One]

Tell that to my router... 20Mbps/1.3Mbps

Verizon FIOS here, my basic bundled rate is 25 Mbps/ 25Mbps. I could opt for faster

Re:Stuck

[TheThiefMaster]

I have 10 Mbps downstream, but still the same 512kbps upstream...

Re:Stuck

[Anonymous Coward]

DIE!!!

Re:Stuck

[Anonymous Coward]

Internet connection speeds....guess uptime's no longer the e-penis huh. Guess I can go restart my MULTICS box then.

This Has A Use Already

[WrongSizeGlass]

I know who can use this type of network speed: the guys trying to make a quadrillion-flop computer. What good is all that CPU horsepower if it can't be used to serve up, um, web pages?

RIP OUT THE CAT5e CABLE BOYZ !!

[Anonymous Coward]

the king is dead. Long live the KING !!

King Cat-5g (I KID YE NOT !!) Must be 100 Gb rated. There is no standard 40 Gb rating for cable.
 

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[Anonymous Coward]

you really think you're gonna be sending 40Gb over copper?

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[d3matt]

over short distances...

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[Bacon Bits]

The standard includes specifications for copper. 40GBASE-CR4 for 40GB which specifies 4 lanes of twinax cable, and 100GBASE-CR10 for 100GB which specifies 10 lanes of twinax.

Surprise, surprise. Serial too slow? Try parallel!

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[Bigjeff5]

There isn't a whole lot of difference between the raw speed of the signal in a copper line and fiber line, the electrical signal already travels at effectively the speed of light (or close enough that it doesn't really matter). It's distance that's a problem for copper. An electrical signal through copper has significantly more attenuation than an optical signal through fiber, which means right from the very start the signal is cleaner and more usable. The cleaner the signal, the easier it is to pick up small variations in the signal accurately, and the more data you can pack into the signal. Copper is also vulnerable to noise, which further reduces the signal quality, which means a less complex signal is possible. This is why copper is useful for ultra-high speed IO inside a computer covering inches or less (the IO in a CPU travels only nanometers and is obscenely fast), but once you start stretching it a few feet its effectiveness drops off dramatically. Fiber is capable of handling much longer distances before the same attenuation loss occurs.

Other than that it's just the equipment on the back end that are different, and the concepts behind both fiber and copper are the same. Only the components are different.

In other words, it's trivial to make copper just as fast or faster than fiber. In fact, the fastest copper connections are already faster than the fastest fiber connections. What isn't trivial is making copper as fast as fiber over the same distances. Fiber wins hands down on a run of any distance. Therefor copper only wins on short runs, due to the huge price difference between the two.

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[QuantumRiff]

Well, if you plan ahead, you account for the attenuation in the cable. You can then counteract it, and get much greater distances. I know the strongest electromagnetic signal in my area is a country radio station, with a tower a few miles away. To counter that, I play hip hop, all day and all night, in my server room. Keeps my distribution switch from getting slowed down and depressed from the country music.

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[Anonymous Coward]

Therefor copper only wins on short runs, due to the huge price difference between the two.

Where, for sufficiently high speed, a "shurt run" might be a few centimetres.

As information technology research moves way faster than physics research, expect everything outside the computer case to be optical fiber in a few years (for some value of "a few"), as even the 0,9m from the computer to the switch in the middle of the server rack is too long to host a multi-terabyte-per-second connection over any copper medium.

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[L4t3r4lu5]

Then why the hell is Parallel ATA giving way to Serial ATA?

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[corerunner]

because it's over short distances and serial isn't too slow

Re:RIP OUT THE CAT5e CABLE BOYZ !!

[Anonymous Coward]

Yeah, the attenuation is part of the issue. But you also have to worry about getting a modulator that is fast enough for those really high speeds over fiber, direct modulation won't cover it at the highest speeds, and even using an external modulator won't cover the very highest speeds. And well before that DWDM becomes more cost effective. So yeah, you can make a faster single lane over copper. But all too often it becomes more cost effective to multiplex and be done with it. The good thing is that we have plenty of bandwidth between the carriers when multiplex optical wavelenght (much better than multiplexing several AM carriers in copper).

Much welcomed tech

[mnmn]

It's interesting how this will increase the adoption of iSCSI storage, yet the original reason to go to iSCSI will be lost since fiber cables will have to be laid.

Either way 1Gbit Ethernet is beginning to feel a bit like a bottleneck with storage and other bottlenecks being removed.

It'll take some time between ratification and cheap D-Link switches...

Re:Much welcomed tech

[afidel]

Nah, thanks to DCE most storage will stay FC using FCoE. What this will do is eventually allow us to get to ridiculous port counts in top of rack and end of row switches and upload all that capacity without requiring a 6" diameter bundle of trunking cables. It will also allow 100Gb to be usable at metro distances since it only requires 4 pairs instead of 10.

Re:Much welcomed tech

[donkeyoverlord]

The original goal of iSCSI wasn't to avoid using Fiber, it was to avoid using Fiber Channel and requiring the creation of a second network, dedicated to storage, that is managed separately from the standard data network.

Re:Much welcomed tech

[whit3]

It's interesting how this will increase the adoption of iSCSI storage, yet the original reason to go to iSCSI will be lost since fiber cables will have to be laid.

That seems a tad disingenuous. The real reason for iSCSI was a
Microsoft price structure that made a network file service very
expensive unless it went in through the 'disk-on-SCSI-bus'
back door.

Linux and iSCSI was a way around the high cost of
a MS server/client system. None of the Linux-only or Macintosh
network systems were so encumbered, and worked
quite well without any iSCSI.

Re:Much welcomed tech

[swb]

I think 10 gig ethernet has been an option for a while now. I'm almost positive one of the sales droids spouted something about Equalogic shipping 10 gig iSCSI SANs.

AFAIK, most small-midsize organizations engaged in iSCSI SAN also do virtualization and thus don't have a ton of hosts to connect so the fiber part is less of a pain than it might seem given they can still get the "IP" part of iSCSI and leverage cheap and still useful 1 gig connectivity elsewhere.

Plus 10 gig can do copper. But there won't be any equipment reasonable priced for a year and I doubt most disk implementations would show a huge difference between 10 and 40 or 100 gig.

Re:Much welcomed tech

[Anonymous Coward]

the original reason to go to iSCSI will be lost since fiber cables will have to be laid...

I'd guess your not aware that Fibre Channel runs on copper as well. This is cheaper and, therefore, common when distances are short. Here is Apple selling some.

The appeal of iSCSI, such as it has, is that it leverages relatively inexpensive and thoroughly ubiquitous Ethernet and IP infrastructure (switches and routers.) However, don't conflate Ethernet/IP with copper; there is a lot of optical in non-edge Ethernet.

iSCSI was never about optical verses copper.

Re:Much welcomed tech

[Anonymous Coward]

Nah, thanks to DCE most storage will stay FC using FCoE.

iSCSI also benefits from DCE.

If you've got a FC infrastructure that's a "sunk cost", then FCoE makes sense. If you're a smaller company that's just starting to centralize storage then iSCSI (or even NFS) can make more sense in many situations.

While 40 and 100 GbE is barely out the door, those you really needed it could have used InfiniBand for quite a while, which has had 96 Gb/s for quite a while:

http://en.wikipedia.org/wiki/InfiniBand

Re:Much welcomed tech

[Anonymous Coward]

It's interesting how this will increase the adoption of iSCSI storage, yet the original reason to go to iSCSI will be lost since fiber cables will have to be laid.

Why? 10 GbE (copper available) from the client to the switch, and then a short run of fibre to the file server(s) where bandwidth is most needed. Hell, you can do 100 GbE over copper with either SFF-8642 (4m) or SFF-8436 (10m) cables as well.

http://www.ethernetalliance.org/files/static_page_files/83AB2F43-C299-B906-8E773A01DD8E3A04/40G_100G_Tech_overview(2).pdf

In addition you don't just lay one fibre, you generally pull like a dozen (or more) OM3 fibres in one intra-building bundle. You use a bunch for "user" data, and a bunch more for storage data.

Re:Much welcomed tech

[afidel]

There's exactly one 4Gb copper FC card and zero 8Gb, the only use I've seen for copper FC is for connecting the bays back to the controllers, from there out it's always fiber optics.

Re:Much welcomed tech

[bertok]

It's interesting how this will increase the adoption of iSCSI storage, yet the original reason to go to iSCSI will be lost since fiber cables will have to be laid.

That seems a tad disingenuous. The real reason for iSCSI was a
Microsoft price structure that made a network file service very
expensive unless it went in through the 'disk-on-SCSI-bus'
back door.

Linux and iSCSI was a way around the high cost of
a MS server/client system. None of the Linux-only or Macintosh
network systems were so encumbered, and worked
quite well without any iSCSI.

WTF are you talking about? Why was this modded up? Is it just because he's saying something negative about Microsoft?

I've worked in Microsoft Windows server environments for a decade, and I've never heard of SCSI specific MS licensing, or any kind of special licensing at all for file servers.

While it's true that a Linux server in general is cheaper from a licensing standpoint (hard to compete with free), that has nothing to do with iSCSI, SCSI, or FC.

The reason iSCSI is popular is because it's simpler to set up, halves the number of ports and switches required for a fully redundant server environment (minimum 2 ports and 2 switches vs 4 and 4), it has real authentication instead of the worthless "zones" crap in the FC world, provides user friendly names instead of numeric IDs, has encryption, 10Gb Ethernet can outperform even 8Gb FC, and even old 1GbE switches can perform adequately if port trunking is used properly.

What this all boils down to is that iSCSI is both better and cheaper than FC. Once popular SAN arrays from big vendors start to appear with 10GbE iSCSI as standard instead of an expensive "option", then FC will start to die a rapid and well deserved death.

Re:Much welcomed tech

[mnmn]

I do not believe you've actually used iSCSI, at all.

The performance numbers are very different and so are the technologies, Microsoft filesharing is file-level and iSCSI is block level. It means with an iSCSI card, the machine can treat volumes as local disks and install any OS.

Secondly, you're confusing iSCSI with NFS. NFS has been freely available even back on Windows NT4. However it was not created to counter Microsoft, it was ALREADY there.

iSCSI until recently has been the only technology that provides block-level storage access and as efficiently as possible on a routable ethernet network. The recent FCoE is even more efficient but its not so easily routable.

Re:Much welcomed tech

[Anonymous Coward]

okay, so, why does anyone want extreme latency block-level IO instead of extreme latency file-level IO?

Re:Much welcomed tech

[Bigjeff5]

It's interesting how this will increase the adoption of iSCSI storage, yet the original reason to go to iSCSI will be lost since fiber cables will have to be laid.

The spec includes 40Gbit and 100Gbit over copper via twinaxial cables, so you do have to make new runs of cable but you don't have to take the fiber hit when you do. Cat 6 definitely won't cover it though, I'm afraid.

Re:Much welcomed tech

[Bigjeff5]

Hey man, don't ruin this amature linux admin's fantasy with real world experience, that's just cruel!

MS licensing is obscene, that's for sure, but they've never tied anything to the hard disk. It's all installs, users, and CPUs, with variations for each category.

Re:Much welcomed tech

[Wyvern2005]

Dude, if you've been in tech for a decade you showed up a bit late for the original "scuzzy" party.....and get off of my lawn!

Re:Much welcomed tech

[Anonymous Coward]

What this all boils down to is that iSCSI is both better and cheaper than FC. Once popular SAN arrays from big vendors start to appear with 10GbE iSCSI as standard instead of an expensive "option", then FC will start to die a rapid and well deserved death.

Cheaper, definitely. Better? I'd debate that, probably placing them roughly on par.

I do really like iSCSI, and I actually started off with it, but I've found to get it working reliably under real stress (including Ethernet switch config) isn't a whole lot simpler than FC IMHO. I also find FC generally more robust, although admittedly I did get stung a few times by buggy implementations of the Linux IET.

FC has a few strong points too, like creating real loops without worrying about hack protocols like STP.

My real problem with FC is the stupid licensing models and of course the ridiculous pricing.

Although I haven't dabbled with it yet perhaps FCoE will bring good sides of both to the table.

Re:Much welcomed tech

[jon3k]

FWIW, you can do 10Gb on copper.

Re:Much welcomed tech

[Anpheus]

Block level makes it possible for the server to not have to know anything about the data, so the client can host encrypted data on the SAN, or boot from the SAN, or both. Or use a filesystem the SAN has no idea how to inspect, etc.

Also, it's not extreme latency. It's a few tens of microsecond latency in a small network with OK switches and 1GbE, which iSCSI targets as a market. Tens of microseconds in still two orders of magnitude less than the latency of a disk seek, so for spinning disks, you're not going to notice a difference. For good SSDs, they'll have twice the response time, but still the same overall throughput (unless you're doing a lot of synchronous IO that has to wait for those responses.)

Better switches and higher throughput connections with iSCSI offloading will get you closer to the single digit microsecond and nanosecond territory, at which point you're more likely to notice that you've saturated the bandwidth to your SAN than notice latency.

Re:Much welcomed tech

[mnmn]

Can you install Windows/Solaris/Linux/AIX on file-level storage, install Oracle/DB2/Exchange/Domino?

Block-level storage can and does completely replace local harddrives. Thats the reason for bladeservers, where blades have everything but harddisks. They're given volumes of fiber channel, iscsi or fcoe to become their local virtual disks. NFS or CIFS would be completely useless to them without first having block level volumes (except for the rare case of Linux/FreeBSD installed on NFS).

More ads faster!

[joelsanda]

What I remember most fondly about CompuServe on my 300 baud modem and Commodore 64 was the lack of ads ...

Re:More ads faster!

[MrEricSir]

Yes, but the porn was low-res and slow to download. So it's a double-edged sword.

Re:More ads faster!

[poetmatt]

and yet in probably 5-10 years even 100 gb/s will probably not be fast enough.

Re:More ads faster!

[geekoid]

I would rather have the adds and my 15Mb line, then my old 1200 baud connection to compuserve.
We only use 300 baud for internal stuff.

In 10 minutes I can down load some porn, whack off, and be a sleep in 10 minutes. In those days it was hours just to get a 5 second clip.

what..TMI?

Re:More ads faster!

[Anonymous Coward]

for me it was the 3 hours into download CARRIER LOST . . . . . . . . . .

Re:More ads faster!

[value_added]

In 10 minutes I can down load some porn, whack off, and be a sleep in 10 minutes. In those days it was hours just to get a 5 second clip.

Hmm. I could write that the above is an example of how user contributions on a site like Slashdot can offer recommendations to the average reader that are both informative and practical. On the other hand, I could write something to the effect that what you wrote provides more information than most of us asked for, or want.

I suspect both of those are too subtle, so I'll tailor my response to the wider Slashdot demographic:

"Ten minutes!??? You need to find better porn!"

Re:More ads faster!

[Surt]

For end users, 100gb/s is almost 'enough'. It's just a hair short (about 2.5x) of the speed needed to stream uncompressed video at the highest resolution anyone is likely to seriously consider, at 240hz. Once you hit that point, you just remote your applications to wherever the data is, and forget about moving data ever again, assuming, of course, that the data is close enough to you to avoid any latency issues for interactivity.

Re:More ads faster!

[afidel]

10Gb Ethernet is 8 years old and it's more than fast enough for all but a niche of applications. Heck even with high consolidation ratios most VMWare servers deployed today don't have a need for a 10Gb ethernet port. It's more useful in channelized form ala HP Flex10 or the Palo adapter in Cisco's UCS systems where you can break out specific chunks of bandwidth for various purposes.

Re:More ads faster!

[bill_mcgonigle]

What about stereo? Multiple users?

Re:More ads faster!

[Vellmont]

uncompressed video? 240 frames/second?

I'll assume you're joking and making fun of the crazy videophiles who think something like this is necessary. So well played, good sir.

Re:More ads faster!

[Surt]

Stereo is covered by the 240hz. But the multiple users .. yeah, for a family of 4, I suppose you might want to multiply by ... 4.

Re:More ads faster!

[Surt]

240hz is needed for stereo 3d at 120hz. 120hz is the upper limit of detectability for about 98% of the population. 60hz is choppy for almost 30%.

My point was just to figure out roughly where you could guarantee that not even the videophiles would be unlikely to complain about the quality, and 120hz is generally well received by videophiles, and unfortunately you do have to multiply whatever rate you pick by 2 for stereo 3d implementations.

Re:More ads faster!

[Vellmont]

My point was just to figure out roughly where you could guarantee that not even the videophiles would be unlikely to complain about the quality, and 120hz is generally well received by videophiles, and unfortunately you do have to multiply whatever rate you pick by 2 for stereo 3d implementations.

Being a videophile isn't about actual perception, it's about being superior. It's a dick measuring contest with specifications. Give them a maximum perceivable specification, and they'll imagine their way out it. You could give these people actual actors in their living room, and they'd complain about the quality of the air in the room and dust count obscuring their vision. "I think the images of the actors in the dust free room are being scattered a bit by the oxygen and nitrogen in the room. If we just had a pure vacuum that would be better. But don't interrupt the sound of course".

 

Re:More ads faster!

[bill_mcgonigle]

yeah, for a family of 4, I suppose you might want to multiply by ... 4.

I think you're still fine. It would be fair to use something like REDCODE Raw as an upper limit for home viewership at 10GB per minute. It actually costs more to process uncompressed video these days, so nobody would put money into such silicon.

Presumably for deployed stereo they'll come up with a 'joint-stereo'-like algorithm that doesn't duplicate the data-rate.

And a good LDS family can used a bonded pair.

Re:More ads faster!

[im_thatoneguy]

Even with Gigabit and 10:1 wavelet compression you're pretty much dandy. 10gb I think would be good enough with adequate compression.

Re:More ads faster!

[bbn]

Please post your math...

240 hz * 1920 px * 1080 px * 24 bit = 11,943,936,000 bit/s = 12 Gbit/s.

Seems to me that you could watch multiple 1080p uncompressed videos at 240 hz at 100 Gbit/s.

Re:More ads faster!

[poetmatt]

it's never safe to assume that there is a "good enough" ceiling of bandwidth, for personal or enterprise use.

I know you specifically went for "end users", but honestly? this stuff is also geared at corporate. Think about it. Do you really need a SC/ST when you can replace it with ethernet? Eventually as things get higher in speeds along with better storage it enables more people to host servers themselves. what is enterprise now becomes home user. Easy examples: home theatre pc's, p2p/bittorrent, more complex gaming, all things that weren't possible 10-15 years ago.

100gb ethernet cuts a lot of hardware costs out right away, and leaves the only reason to use fiber internally for security. Also it's not new, as others posted - this has been in development for a long time coming.

Re:More ads faster!

[Surt]

240hz *
http://en.wikipedia.org/wiki/Super_Hi-Vision
7,680 * 4,320
* 64 bit (not 24, you don't want to have to lose accuracy in blend)
= 509 megabit. I was actually off by a factor of 4 because I thought the 8k format was 4x2, not 8x4 before I double checked. So, sorry, it's actually worse than I thought!

Re:More ads faster!

[Surt]

Compression is horrible for a lot of content, so videophiles will insist on uncompressed for some applications.

Re:More ads faster!

[Anonymous Coward]

What if you want to stream two videos?

Re:More ads faster!

[Surt]

Yeah, others pointed that out ... as I responded there, you probably want to use some small multiple to allow for multiple streams.

Re:More ads faster!

[Surt]

My point was merely that there is a concretely definable point at which, barring latency issues, you no longer need to send data to the end user, and can instead remote the output losslessly, and do the processing wherever the data lives. This is at most one order of magnitude past 100gb/s.

Seriously?

[Pojut]

I just finally upgraded all of the connections in my house to Gigabit Ethernet, you fucking clod you!

Re:Seriously?

[broken_chaos]

Sorry to break it to you, but...

Disc speeds

[f3rret]

I can't help but wonder what you could actually use 100Gbit/s for, I mean to the best of my knowledge (which is not all that vast I admit) you'd be hard pressed to find a storage unit that can handle these sorts of speeds.

Re:Disc speeds

[afidel]

100Gb isn't for server to server or server to storage connections today, it's for network aggregation (switch to switch ISL's). 40Gb is there for server to storage on some high end configurations.

Re:Disc speeds

[Anonymous Coward]

Yeah. It's not like people have more than two machines on a network or anything.

Re:Disc speeds

[Anonymous Coward]

European internet exchange points (IXP's) such as LINX and AMS-IX are eagerly waiting 100GE. There's only so many 10GE interfaces you can aggregate together between large chasis-based switches.

Re:Disc speeds

[CyprusBlue113]

Interlinks. Router to switch, switch to switch, ISP to ISP, etc.

Re:Disc speeds

[Surt]

SSDs are going to hit 6 gbit/sec in the next year or so. Multiply by 17 devices on a SAN and you're done.

Re:Disc speeds

[dave562]

Another thing to consider is that "100Gb" is also a measure of bandwidth. While a switch might be able to handle 100Gb, it won't be able to handle 100Gb to every port at the same time. Even on a 12 port switch that is less than 10Gb per port. That is still a lot of bandwidth, but you can obviously predict how it will degrade as the port count increases.

Re:Disc speeds

[evilbessie]

You couldn't is the simple answer, unless you perhaps run the network at CERN. This is Tier 1 territory at the moment, eventually your corporate backbone, probably never on the desktop though. It just means you can run 10x10Gb networks through a single cable. So unless you have massive needs of say a large internet video hosting site you probably never need anything like this. In time I would expect it to filter down to medium sized business, by which time, no doubt, 1Tb/s will have been standardised, if not 10Tb. To keep everything smooth at the end you need some really hefty cables in the middle.

Re:Disc speeds

[Kjella]

Delivering 100 Mbit/s Internet to 1000 people before over-subscription seems like a nice application. Unless you're in the US in which case it probably covers New York.

Re:Disc speeds

[afidel]

Yes, but there are routers with multi-terrabit backplanes, 100Gb is for connecting multiple such routers together.

Re:Disc speeds

[forkazoo]

I can't help but wonder what you could actually use 100Gbit/s for, I mean to the best of my knowledge (which is not all that vast I admit) you'd be hard pressed to find a storage unit that can handle these sorts of speeds.

Just depends on what you consider a "storage unit," and what you are willing to pay for it.

If "storage unit" means a hard drive, then no.

If "storage unit" means a big box in a data center with room for hundreds of drives, then yes you will be able to find interesting uses for this speed right now.

Re:Disc speeds

[bertok]

SSDs are going to hit 6 gbit/sec in the next year or so. Multiply by 17 devices on a SAN and you're done.

Those are consumer grade devices. Many SSDs are already well above 10Gbit speeds, and I fully expect 20Gbit in a single PCI-e card this year or early next year. Just 5 of those could saturate 100Gbit!

Re:Disc speeds

[Surt]

True, though many a small business has a SAN built on consumer grade devices. My point was exactly that the low end will be pushing up against this limit all too soon.

Re:Disc speeds

[Bigjeff5]

Look up Storage Area Network, and Trunking.

There is never, ever a such thing as "too much bandwidth". You're just thinking too small, that's all.

Re:Disc speeds

[Bigjeff5]

That was his point.

Re:Disc speeds

[evilviper]

you'd be hard pressed to find a storage unit that can handle these sorts of speeds.

If it was needed "NOW" it would be getting manufactured and sold NOW. It's not. It's just now getting standardized, so the hardware can be developed, and come out at a reasonable price a few years in the future, when it will in fact be needed.

We've had 10GBit ethernet for quite some time now, and yet the cards still cost $1,000 a pop. So if you could find a use for that much speed (and many do) you might find the cost prohibitive, and would hope it had been developed a couple years earlier, so the price would have had more time to drop.

10Gbit infiniband, however, costs about 1/3rd as much, much higher speeds are available right now, and 100Gbit+ infiniband is supposedly planned for next year. So, if there's a question to be asked, it's whether ethernet is even going to be the LAN standard of the future, versus something with lower latency, processing overhead and cost.

Re:Disc speeds

[guruevi]

The interconnect will be 6Gbit/s and the highest interconnects I've seen commercially used are bonded 4 * 10Gbit/s (40Gbit/s) mainly for redundancy and latency. At 50MB/s (400Mbit/s) you'll need at least 100 of them to fill the bandwidth - enterprise SSD's don't sell for $200/32GB, try $2000.

Re:Disc speeds

[Surt]

I'm not sure who you're replying to ... I didn't say anything about pricing. And even consumer level SSDs are already at 250MB/s, not 50.

Re:Disc speeds

[Anonymous Coward]

10GbE adaptors do not cost $1000's. Infiniband is also not suitable for switched IP networks, which is where a lot of the demand for faster Ethernet comes from, and that makes it totally unsuitable as any form of LAN technology.

Re:Disc speeds

[Anonymous Coward]

Dude, have you not ever seen a non-blocking switch? There are plenty of terabit backplane switches out there right now that will handle dozens of simultaneous 10G connections without buckling under load. Non-blocking 100G is only a short matter of time.

Re:Disc speeds

[Anonymous Coward]

Large, interconnected networks of constantly streaming multicast video.

Consider, for instance, a statewide transportation network of highway cameras. With 100Gbit/s, the network designers will be able to continue to slack off on their multicast routing/design and just pump everything everywhere. ... :P

Re:Disc speeds

[evilviper]

10GbE adaptors do not cost $1000's.

Thanks for the link. That's the cheapest I've seen 10Gbit, but that still leaves infiniband cards of the same speed at 1/5th the price.

Infiniband is also not suitable for switched IP networks

I'm listening. Infiniband switches and routers are widely available, and cheaper than their ethernet equivalents. Similarly, IPoIB has been around for quite a while. In fact Infiniband was designed with substantial similarity to IPv6 in mind, and can transport IPv6 packets natively (with a few notable restrictions). So please tell me, what is it that makes you think infiniband is not suitable?

Additionally, I should say that the "IP networks" part should be removed. Part of my idea is to get away from IP on the LAN, if possible. A better IPC method would be a boon, and is long overdue. Where it's routed out to the WAN and the cloud, it can be converted into IP. On the LAN, it could reasonably be just about anything. After all, IPX networks coexisted with IP and the like for many years.

i have only 3Mbps

[stanlyb]

And i still have only 3 mega bits per seconds..........

About Time!

[Above]

I've been waiting to connect to my 8M Cable modem with 100GE for a while now. Finally, no more bottleneck!

Re:About Time!

[vlm]

I've been waiting to connect to my 8M Cable modem with 100GE for a while now. Finally, no more bottleneck!

The inter-router links that connect your CMTS back at the headend might, eventually, be 100GE. 100GE would be about 12K customers at full blast. With reasonable oversubscription ratios, figure the headend for a small city, or "a major portion" of a large city.

Re:About Time!

[Anonymous Coward]

You laugh, but my router is actually the bottleneck in my setup -- I'm still in the process of switching to GigE.

100Gb/sec

[wolrahnaes]

That's:

9102 full 3.5" floppy disks (1.44MB)
18 full CDs (700MB)
1 full DVD (8.54GB)

Every second, with room to spare (I just counted complete transfers).

Of course I'm still waiting on 10g to be affordable for LAN use and barely get 10m to the WAN, so I'm sure the various **AAs aren't afraid of this for now.

Re:100Gb/sec

[HTH NE1]

That's:

9102 full 3.5" floppy disks (1.44MB)
18 full CDs (700MB)
1 full DVD (8.54GB)

Every second, with room to spare (I just counted complete transfers).

CD and DVD capacities and transfer rates are measured in metric units, and 1.44 "MB" floppies are a combination of one metric and one binary measure (1.44 "MB" * 1024000 bytes/"MB"). Still, 8 bits per byte, so 100 Gb/s is 12.5 GB/s.

Using the correct units, I get:

1 DVD
17 CDs
8477 floppies

Consider a 1.44 "MB" floppy is defined using two different definitions for a kilobyte: a 1000 B/KB and a 1024 B/KiB factor.

(1.44 "MB"/floppy * 1024000 bytes/"MB" == 1474560 B/floppy; / 1,000,000,000 bytes/GB == .00147456 GB/floppy; 100 Gb/s * 1B/8b == 12.5 GB/s; 12.5 GB/s / .00147456 GB/floppy > 8477 floppies/sec).

In practice though, it is one floppy, one CD, and one DVD across the board, because it takes more than one second to swap each of them. And that's assuming you've pre-cached the first disk to memory since you're not going to find a drive that can read one 3.5" floppy disk that fast, and I doubt a DVD could even survive the RPMs necessary to read it that fast, let alone a CD.

Re:100Gb/sec

[MachineShedFred]

Yeah, but how many Libraries of Congress per Fortnight? (LoC/Fn)

Go for it, math boy! Show us what you got!

Re:100Gb/sec

[Bigjeff5]

6,032 LoC/Fn.

Re:100Gb/sec

[wolrahnaes]

CD and DVD capacities and transfer rates are measured in metric units, and 1.44 "MB" floppies are a combination of one metric and one binary measure (1.44 "MB" * 1024000 bytes/"MB"). Still, 8 bits per byte, so 100 Gb/s is 12.5 GB/s.

Using the correct units, I get:

1 DVD
17 CDs
8477 floppies

Consider a 1.44 "MB" floppy is defined using two different definitions for a kilobyte: a 1000 B/KB and a 1024 B/KiB factor.

(1.44 "MB"/floppy * 1024000 bytes/"MB" == 1474560 B/floppy; / 1,000,000,000 bytes/GB == .00147456 GB/floppy; 100 Gb/s * 1B/8b == 12.5 GB/s; 12.5 GB/s / .00147456 GB/floppy > 8477 floppies/sec).

I used the following capacities in my calculations:

3.5" Floppy: 1,474,560 bytes (11,796,480 bits)
80 minute CD-R: 360,000 sectors at 2,048 bytes each (Mode 1) = 737,280,000 bytes (5,898,240,000 bits)
DVD+R DL: 4,173,824 sectors at 2,048 bytes = 8,547,991,552 bytes (68,383,932,416 bits)

Our discrepancies in numbers seem to come from the fact that I did my math lazily using Google Calculator and queries such as "100 gigabits / X bytes". Google uses the binary meaning of 107,374,182,400 bits as the value for 100 gigabits, rather than the SI definition of 100,000,000,000 bits.

In practice though, it is one floppy, one CD, and one DVD across the board, because it takes more than one second to swap each of them. And that's assuming you've pre-cached the first disk to memory since you're not going to find a drive that can read one 3.5" floppy disk that fast, and I doubt a DVD could even survive the RPMs necessary to read it that fast, let alone a CD.

Of course. Just using the capacities of familiar media to put things in perspective for those who may not spend their days maxing out networks.

Shame about the MTU

[elFarto the 2nd]

The MTU is still 1500 bytes though :(

Re:Shame about the MTU

[Anonymous Coward]

Jumbo frames, dear boy. Jumbo frames.

Which have been around since 1Gb ethernet.

Re:Shame about the MTU

[Anonymous Coward]

jumbo frames are not standard even at 100gbps.

Re:Shame about the MTU

[Bigjeff5]

When are we going to get uber frames though?

Re:Shame about the MTU

[elFarto the 2nd]

Wow, 9000 bytes, that makes all the difference. And beside, Jumbo frames aren't officially recognised by the 802 spec. Just think how much work is a processor going to have to do to fill a 100Gb/s link with 1500 or 9000 byte packets?

Re:Shame about the MTU

[Anonymous Coward]

I agree. Due to the MTU size staying the same each network suffers a header tax. The optimal size of MTU for 100 GB is 6 MB and not 1500. See this link for more information: http://staff.psc.edu/mathis/MTU/

So thank you IEEE

Pardon my ignorance

[RingDev]

But are we talking about 100Gb/s over copper or fiber?

-Rick

Re:Pardon my ignorance

[Wesley Felter]

You can run it a very short distance with fat copper cables, but almost everyone will use fiber.

Re:Pardon my ignorance

[Shimbo]

But are we talking about 100Gb/s over copper or fiber?

-Rick

Fibre and short-haul (~10m) copper, at least for the current standard. Historically, there's usually a lag of several years between a new Ethernet standard and a 100m copper version.

I'm a bit sceptical about folks who say they'll never be a copper version, because I've heard that tale often enough before. I confidently predict it will be the Year of Linux on the desktop before it's the year of Fibre to the Desktop.

Re:Pardon my ignorance

[afidel]

The copper solution requires twinax, might as well run fiber as it's easier to deal with at length and can actually fit into the existing raceways (twinax is huge). There's not enough bandwidth and S/N margin in even Cat6A to do 100Gb at 100m, you need Cat7A which was just approved late last year and which requires a full plant re-work who's going to do that when a OM3 fiber installation should be good all the way to 1000Gb.

Re:Pardon my ignorance

[Anonymous Coward]

My friend works in network/video solutions for Aviation (Lufthansa) and the one network they had to design, Lufthansa required the use of Cat 7.. and man those were some funky big plugs.

Re:Pardon my ignorance

[Bigjeff5]

Twinax isn't too big, it's the bundle of 10 twinax you need to run 100gbit that are huge.

I'm a little confused, though. Cat6 is capable of 10GbE, so why not bundles of 4 and 10 Cat6 for the standard as well, instead of just twinaxial? I recognize you'd need a special port setup, but that would still be significantly smaller than twinax. They would then be capable of 100m, would they not?

Re:Pardon my ignorance

[Anonymous Coward]

Regular CAT6 is only good for about 50m in clean conditions or 37m in "dirty" (electromagnetic interference) conditions.

There's CAT6a which can be used for the longer 100m runs though.

Re:Pardon my ignorance

[Ster]

The copper solution requires twinax, might as well run fiber as it's easier to deal with at length and can actually fit into the existing raceways (twinax is huge).

I think you're thinking of CX4, which is indeed huge. 10Gb TwinAx comes in SFP+, which is the same port that you use for 10Gb fiber.

Re:Pardon my ignorance

[afidel]

OM3 has a typical jacket diameter of 2mm vs 6-8mm for twinax and 8-9mm for 7a, cat 5e is closer to 4-5mm typically. When you have hundreds of runs in a raceway it makes a big difference.

The Eighties called...

[macraig]

... and wants its A-Team back. Your meme is violating some copyright somewhere.

Re:The Eighties called...

[Tetsujin]

... and wants its A-Team back. Your meme is violating some copyright somewhere.

Was "I pity the fool" even in The A-Team? I watched like the first two or three seasons of the show and I don't recall Mr. T ever saying it. It was in Rocky III though...

Re:The Eighties called...

[macraig]

Good God... somebody modded me as Informative when I wasn't trying to be!

Re:The Eighties called...

[macraig]

No, Overrated wasn't what I had in mind, either....

Re:The Eighties called...

[Anonymous Coward]

No; He did say "shut up, fool!" and "you crazy, fool!" and a whole lot of other stuff "fool!" stuff, but the pity just wasn't there.

Re:The Eighties called...

[Bigjeff5]

It was developed later, and is now Mrs T's main catchphrase. He still says "fool" a lot, and is very accusatory about it.

Re:The Eighties called...

[Anonymous Coward]

I dunno, but the new movie shows at the beginning that BA has "Pity" tattooed on his left knuckles and "Fool" on the right ones.

Crazy speeds are playing games w/ my mind!

[shicaca]

It's amazing. I'm 26yo and started tinkering w/ Win95 when I was ?8?. The speeds are taking such huge leaps that I now have to think to myself, "OK so 100GBps is something special??" I have a 100mbps / 1Gbps router, and am so used to the 100meg cables being the "norm" that the gbps is lost on me for a while. I actually have to think of what I HAVE vs. what they're proposing! To that I say, bring it on, bring it cheap, I'm ready for integration into my next round of systems/router *huge smile*

Re:Crazy speeds are playing games w/ my mind!

[siddesu]

Well, I am a bit older than that, and I was so overwhelmed by the speeds of modern computers, that I switched to programming microcontrollers for fun. With chips running at few MHz, having just a few kb of ram and 9600 bps available to talk to them, I feel right at home, and comfortable. ;)

Hardware Vendors Giddy

[Revotron]

Do these people seriously think we're going to see 40Gbps servers any time soon?

I can't wait to see the list prices on the ensuing flood of modules.

Cisco 40Gbps SFP: $50,000/unit
Cisco switch equipped to handle them: $300,000+/chassis
Intel 40Gbps Server NIC: $10,000/unit
Server with a completely new bus technology built to handle them: If you have to ask...

Get your cost centers out, it's time to upgrade!

Re:Hardware Vendors Giddy

[afidel]

Nope, more like Intel 40Gbps NIC, $1500. They went out of their way to reuse existing technologies to bring the price per port down at launch.

Re:Hardware Vendors Giddy

[QuantumRiff]

If I have a blade server, with 10 blades, each one running several virtual machines.. Do I want to deal with a each blade having at least two nics, and all that cabling, plus either FC cards, or additional 1GB iScsi cards?

Or do I want the back of the chasis for the blade server to just have two 40GB/s network connections?

Re:Hardware Vendors Giddy

[drmerope]

These people? At the moment interest in 40G/100G is related to aggregating 10G server connections, not in running 10G to the server.

802.3ba takes on cool name

[theanyday]

I think the ba stands for badass... I'm just sayin'.

One cable to rule them all

[Citizen of Earth]

USB3, HDMI, DVI, Ethernet, DisplayPort, FireWire, eSATA, proprietary. There should be one kind cable that can be used for all of these purposes. We have the technology. Consumers will thank you.

Re:One cable to rule them all

[MachineShedFred]

They actually tried this with FireWire (IEEE-1394) in the consumer electronics industry back in 2000-ish, but then the whole HDCP thing came up, and that was that.

The idea is that you'd have a home theater receiver that just had a crapload of firewire ports on the back, and all your stuff would plug in via that, including speakers. Never happened though.

Re:One cable to rule them all

[Nemyst]

Wait, you want ONE cable to rule them all? You do realize the average consumer already has enough problems understanding how YPbPr cables work (green in green, blue in blue, red in red, no not the audio!)? Making all things unified would just mean more money for BestBuy's installation services. Either that or lots of people would be watching their music and hearing their network.

Re:One cable to rule them all

[Anonymous Coward]

Seriously, the USB-IF needs to get off their ass and put out a BSD or LGPL standard USB-over-ethernet stack and just get it over with. The necessary parts are all there in the wireless USB spec in terms of congestion and security, so all you need is the OS shim in the main USB driver.

That being said, LightPeak being the spiritual descendant of the aggregating dumb pipe concepts? Might have legs this time...

Firewire
WiMedia
Bluetooth
WiFi
60GHz stuff

Re:One cable to rule them all

[Rene S. Hollan]

I dunno. HDMI 1.4 now sports ethernet and audio return channels. About the only thing absent is USB for low/high speed data (keyboards or mice / disk drives: 100 Mb/s ethernet is a little slow for disks).

So, I suppose it will factor out to three cable types: HDMI for "media" connections that are video-centric, ethernet for long distance data and networking connections, and USB for local data and peripherals. Maybe add 1394 (firewire) for video capture and control, though GbE and even 100 Mb/s ethernet could serve that purpose (my Motorola DCH3200 has an ethernet port on it, but video is transferred over 1394).

Is four cable types too much?

Considering that each type already has a primary purpose, and that it lets us get rid of serial, mouse, keyboard, and parallel cables, I think it's still an improvement.

Re:One cable to rule them all

[aaarrrgggh]

OSI Network model. Separate the physical layer from the application layers and everything in between.

Re:One cable to rule them all

[Bigjeff5]

They actually tried this with FireWire (IEEE-1394) in the consumer electronics industry back in 2000-ish, but then the whole HDCP thing came up, and that was that.

It probably would have worked if Apple had allowed it to succeed. They always fuck that kind of shit up.

Re:One cable to rule them all

[evilviper]

USB3, HDMI, DVI, Ethernet, DisplayPort, FireWire, eSATA, proprietary. There should be one kind cable that can be used for all of these purposes.

HDMI/DVI/DisplayPort are for raw video data. They have NOTHING in common with the USB mouse/keyboard on your desk. It makes no sense to combine them.

There's some good reasons for the differences. For instance, even if I could hook up my internet access to the same port as my hard drives, I never would... One needs low-overhead, realtime and no security, while the other is high overhead, delays are better than realtime constraints, and needs much more security.

While you have a point about the display ports, it's far worse to have only one type available, which always sucks.

Re:One cable to rule them all

[ufoolme]

Hopefully one day people might see the light...

Re:One cable to rule them all

[StuffMaster]

Well, HDMI has DisplayPort as competition, but I don't mind since it's better.

C-64 porn

[Tetsujin]

Yes, but the porn was low-res and slow to download. So it's a double-edged sword.

Still, I think you're underrating the merits of the slow reveal... I mean, as the image file was loaded byte by byte onto the computer's memory, filling the display with that lustworthy graphical data, gradually revealing more and more, until you had a naked woman on your screen in 320x200 glory, 1bpp plus 4 bit colors, foreground and background, per 8x8 character cell... The five minute wait for the elusive delights to be laid plain was like a striptease...

And when I say 5-minute wait, that's how long it took to load an image from disk. Modem would take longer. :)

Re:C-64 porn

[MrEricSir]

The image format changes the strip-tease.

BMP loads from the bottom-up. It's Sir Mix-a-lot's favorite format.

Progressive JPEG gets less blurry as it loads, simulating being drunk at a strip club.

Light Peak

[bill_mcgonigle]

USB3, HDMI, DVI, Ethernet, DisplayPort, FireWire, eSATA, proprietary. There should be one kind cable that can be used for all of these purposes. We have the technology. Consumers will thank you.

Are you here from Intel marketing?

<wp:Light_Peak>

Oh, heck, that's still not working. fine:

http://en.wikipedia.org/wiki/Light_Peak

Light Peak

[Anonymous Coward]

They actually tried this with FireWire (IEEE-1394) in the consumer electronics industry back in 2000-ish, but then the whole HDCP thing came up, and that was that.

The idea is that you'd have a home theater receiver that just had a crapload of firewire ports on the back, and all your stuff would plug in via that, including speakers. Never happened though.

Well, you can do a this to a certain extent with HDMI. If you throw in CEC then all you theoretically need is one remote as well.

In theory. :/ I've actually read through the HDMI CEC spec, and it makes a lot of sense, but don't have a television so I can't really say how well it would work.

We'll see how Intel's Light Peak works out for the One True Cable(tm).

Who wrote this?

[Anonymous Coward]

"The new standard will act as the catalyst needed for unlocking innovation across the greater Ethernet ecosystem."

Wow, bullshit overflow!

Re:Who wrote this?

[CyberDragon777]

"The new standard will act as the catalyst needed for unlocking innovation across the greater Ethernet ecosystem."

Duh, Cisco marketing!