Slashdot Mirror
IEEE Sets New Ethernet Standard That Brings 5X the Speed Without Cable Ripping (networkworld.com)
Reader coondoggie writes: As expected the IEEE has ratified a new Ethernet specification -- IEEE P802.3bz -- that defines 2.5GBASE-T and 5GBASE-T, boosting the current top speed of traditional Ethernet five-times without requiring the tearing out of current cabling. The Ethernet Alliance wrote that the IEEE 802.3bz Standard for Ethernet Amendment sets Media Access Control Parameters, Physical Layers and Management Parameters for 2.5G and 5Gbps Operation lets access layer bandwidth evolve incrementally beyond 1Gbps, it will help address emerging needs in a variety of settings and applications, including enterprise, wireless networks. Indeed, the wireless component may be the most significant implication of the standard as 2.5G and 5G Ethernet will allow connectivity to 802.11ac Wave 2 Access Points, considered by many to be the real driving force behind bringing up the speed of traditional NBase-T products.
until its done.
...does it just require new plugs and jacks?
This will certainly save a lot of money for enterprises. I expect it will be the RARE company that will actually need 5Gbps per workstation. Most can probably get by on 100Mbps.
Wasn't 10Gbps a thing already?
Bob Metcalf, 1976
Why not just 5 apps per page like we had back on ios 5?
Instead of this 1 app per page crap we've been stuck with since ios 6 on the iphone and ipod touch.
Minimum threshold fixed. Thanks!
Wasn't clear from TFA if this would work on Cat 5e, or if Cat 6 is required.
No.
802.11ac gets its speed via multiple spatial streams, using more than one antenna on both Tx and Rx sides.
Besides, the coax is 100% available all the time, while the wirelsss protocol is based on half duplex transmissions, and can only transmit to N clients at any given time (on each subband) when you have N antennas.
So for wired transmission you better use something else like MoCa.
This new standard is very interesting: it employs the same coding and spectral density as 10GBase-T (6.25 bps/Hz), but it employs the available bandwidth (Hz) depending on the cable category: Cat.5e (100 MHz) can provide 2.5Gbps and Cat.6 (250 MHz) can provide 5 Gbps.
Interestingly, before this standard there was no practical use for Cat.6 cabling: any speed you could obtain using Cat.6 cable (1Gbps) could be also obtained using cat.5e, and if you wanted something faster (10Gbps) you needed Cat. 6A (500 MHz BW). This newly ratified standard finally gets some use from those extra MHz you have in Cat. 6, if you have installed it. It will be interesting to know if 802.3bz ports will be able to measure link bandwidth to adapt speed accordingly to 2.5/5Gbps.
For the Price of 10GB and 40GB, so I can get rid of those pesky (and expensive) Fibre Channel links to my storage!!!!
That has been in development for ages now...
Latency not whitstanding, that's what infiniband is for anyway!
*** Suerte a todos y Feliz dia!
SInce 1990 we were promised IDF closets would be a thing of the past.
Everyone would use wifi and VOIP by the year 2000. We all have fast cell phones so why can't our offices use the same?
http://saveie6.com/
I have been working with 2.5G for around a year now using a 2.5G physical interface chip from Aquantia that seamlessly handles everything from 100Mbps to 10Gbps including 1G, 2.5G and 5G. If the cable isn't too long I've run 10G over cat 5. Hopefully the prices will drop quickly once more companies support this standard since I just bought the cheapest 2.5G switch I could find, 8 ports for around $1200 for development purposes. It also interoperates fine with standard 1G equipment.
Aquantia is also nice is that unlike many phy chip vendors their phy SDK is free as in beer and is fully GPL and BSD compatible, though it will need to be re-written for the Linux kernel to follow the guidelines. I re-wrote it for U-Boot though I won't be able to push it upstream for a while yet. The chip I'm using even supports MACsec in hardware. There were two different 2.5G proposals, one from Broadcom and one from Aquantia. The Aquantia is the one that ultimately got accepted as the standard.
This post is encrypted twice with ROT-13. Documenting or attempting to crack this encryption is illegal.
Just before they upgrade my token string and tin can coupler...
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
This is cool, but ultimately irrelevant until someone forces the ISP's to admit there is no bandwidth shortage and do some high tier interchange upgrades. The current venal ISP's have spent millions convincing the FCC and customers there is a shortage of bandwidth so they can vastly overcharge for the 'available' bandwidth. As long as the cable companies won't compete and aren't interested in resolving the situation, most of us are stuck in a hell where 25 mbps is the best we can get on a good day.
errr....umm...*whooosh* *whoosh* Is this thing on ?
I'm kind of struggling for what this is good for besides giving switch vendors a reason to push needless IDF upgrades and technology vendors yet another upcharge option.
1 gig Ethernet is already overkill for just about every desktop purpose and still has some useful life left in many data center applications, especially for lower performance areas, even in network storage.
The only place it becomes somewhat weak is in heavy use AC wireless deployments where it can be truly taxed, but most often even these deployments the vast majority of use reverts to the average of typical cabled clients.
It also feels like a reason to keep prices artificially high on 10 gig copper. 1 gig was sky high expensive when it first came out, but quickly became commoditized and very soon nearly everything came with 1 gig ports. 10 gig base T seems like it's been out for ages but prices really haven't dropped nearly as fast and I can't quite figure out why, other than it's fast enough to cut port densities by at least half while still providing 5x or greater throughput of 1 gig ports in most server deployments (ie, if you had 4x 1 gig ports and switch to 2x 10 gig ports, you have 20 gig aggregate vs. 4 gig aggregate and single stream throughput 10x the 1 gig solution).
And as usual, vendors can't stand the idea of the customer buying half of what they did before and getting 5-10x more value than they used to.
I guess the new standards will be great, but only if they replace 1 gig wherever you used to expect 1 gig, ie, everywhere. Otherwise it's either irrelevant or a new way to pay higher prices for 25-50% of the performance you should be getting out of 10 gig at the price -- or higher -- you ought to be paying for 10 gig these days.
I've fallen off your lawn, and I can't get up.
To add to what you said, it very much depends on length, and also on exactly how the termination is done (untwisted length, etc). Ten feet of CAT5 introduces less noise than 100 meters of CAT6.
Also, gigabit was actually designed to work over cat5, but barely.
If you frequently make a change to a 10GB file, check out rsync. It transfers only the bytes that changed, rather than the whole file. Basic usage:
rsync -av local/file.bin user@123.1.1.8:/home/remote/file.bin
Where the user has ssh access.
Folks,
For some more background on 2.5G/5G 802.3bz and NBASE-T, check out my blog post “What Can the NBASE-T Alliance Teach Us About the Standardization Process?” (http://www.nbaset.org/can-nbase-t-alliance-teach-us-standardization-process/)
Also look at the two white papers, “NBASE-T Ethernet Technology Basis for the IEEE 802.3bz Standard” (http://www.nbaset.org/technology/library/white-paper-1/) and “NBASE-T Performance and Cabling Guidelines” (http://www.nbaset.org/technology/library/white-paper-2/).
Cheers
Peter Jones
Chair, NBASE-T Alliance
Of course if that "large amounts of data" of data is multiple files, rsync doesn't have to read the unchanged the files. It can see by the file modification time and size that it matches the remote copy.
You mentioned ZFS, and offsite backup. For our business grade offsite backup and hot spare, we use LVM (logical volume manager). If you have a very large file, particularly a drive image, you'll get significantly better performance by creating it as a logical volume rather than as a file* on another filesystem, and LVM can easily list the blocks that have changed since the last snapshot. In fact, you don't even have to list the blocks, you can just transfer the copied-on-write parts directly because they are stored as a separate volume.
* A logical volume is of course still a file, because on *nix everything is a file. Whether you call it /home/ubuntu.img or /dev/mapper/images/ubuntu, it's a file either way, and you can do file things with it. By skipping the step of putting on ext4 filesystem underneath to hold this large file, you have less indirection and better performance, as well as enhanced data recovery options if bad things happen. (Having a filesystem inside another filesystem tends to confuse some data recovery operations).
PS:
> I've only tinkered with it in VM environments but I would like to give it a spin as an offsite backup sync solution.
In all my years on Slashdot I've never done this, but since you said you would like to give something like this a spin:
We've spent many years developing a pretty bad ass offsite backup solution based on this concept. One reason it's bad ass is that I found some cool ways to make it very efficient (cheap). You can boot up your backups live in our DC and SSH to them (or however you like to access them, they are exact replicas of the original, other than IP address). If your need is enough that you might be interested in spending about $30 / month for a really nice solution, let me know.
On a single frequency, yes. But there are lots of those.
Coax is half-duplex too
No, it's not.
With proper impedance matching networks and reasonable termination at the ends of a run you can send separate signals at the same frequency/band of frequencies down a cable in each direction. (Impedance discontinuities DO reflect some of the signal going one way back the other way, causing some interference. But even that can be "tuned out" by suitable corrections if it's too severe to just ignore.)
You can do it on a balanced pair, too. Telephones have done this with audio for more than a century, and I recall encountering a simple hack to do it all the way down to DC back in the days of discrete-transistor logic. (And it has nothing to do with two wires being involved, either. With N (= any power of 2) conductors and "phantoming" you can have up to N-1 balanced and one unbalanced two-way transmission lines on N wires.
Time Domain Reflectometry does this to FIND and MEASURE discontinuities in a cable, essentially firing a pulse down the cable and listening to the reflections, radar-style.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
And the reason you cannot do this with radio is that the noise from the transmitter is greater than the received signal.
If both the source and destination are local disks, copying the file is faster than comparing, so it copies. Over a network, comparing byes is faster (unless your network is faster than your disks), so it when used over a network it compares bytes by default.
Because twisted pair is what is already in the walls. Higher speed standards for wired internet lets you squeeze more out of the installed base of cable. Replacing the cable is far more costly than replacing the equipment at the endpoints - not because the wire is expensive (though it has become a bit more so in recent years), but because of the labor costs of installing it and repairing the damage done to walls and ceilings.
In fact it uses multiple hashes, called checksums in rsync lingo. It uses a very fast rolling checksum to quickly identify the parts that HAVE changed. It uses a more accurate, but slower checksum to verify the parts that have NOT changed.
And the reason you cannot do this with radio is that the noise from the transmitter is greater than the received signal.
Actually you CAN manage it with radio - very difficultly, with very careful antenna design.
But the combined antenna has to be far from anything that reflects, absorbs, or just phase-shifts any substantial amount of the transmitted signal energy. If not, the discontinuity destroys the careful balance that nulls out the transmitted signal at the receiver. That gets you back to the "transmitter shouts in the receiver's ear much louder than the distant communications partner" case. So it's not very practical in the real world.
Bantam Dominique roosters crow a four-note song. Once you've heard it as "Happy BIRTHday" you can't NOT hear it that way
And the reason you cannot do this with radio is that the noise from the transmitter is greater than the received signal.
Actually you CAN manage it with radio - very difficultly, with very careful antenna design.
But the combined antenna has to be far from anything that reflects, absorbs, or just phase-shifts any substantial amount of the transmitted signal energy. If not, the discontinuity destroys the careful balance that nulls out the transmitted signal at the receiver. That gets you back to the "transmitter shouts in the receiver's ear much louder than the distant communications partner" case. So it's not very practical in the real world.
I have only seen it done where the transmit and receive antenna had considerable separation. Oddly enough, I designed some yagi antennas for minimum sidelobes which could meet this requirement. They had lower gain than a normal yagi but when testing them in the field, they could *see* reflections from things like trees and bushes 100s of yards away. I never got them to be better than about 30dB because that was as good a hilltop used as an antenna test range as I could find. Using them to passively track aircraft at 10s of miles from their reflections was trivial and gave a real feel for how radar and stealth work.
Cancelling the near end crosstalk requires sampling the transmitted signal. If it is sampled before the final amplifier, then noise added by the final amplifier clobbers the received signal. If it is sampled after the final amplifier, then the dynamic range difference between the transmitted signal and transmitted noise clobbers the measured transmitted noise so that cannot be cancelled anyway. Apparently someone has solved this problem in an integrated form but I am dubious of their claims.
Back when I was first playing with SSB transmitters, I noticed the problem when keying up an SSB transmitter with no modulation which produced noise across the entire band which was more than 20dB higher than the background noise. You could still receive signals of course but maximum range was reduced by more than an order of magnitude. Every 6dB halves the distance. This neatly explains why hilltop transmitters include fixed cavity filters.