IEEE turns crank on new wired Ethernet standards

Industry interest in faster enterprise Ethernet on twisted pair cables is driving a new standardisation effort in the IEEE, the chair of the Ethernet Alliance says, with a new study group formed last week to kick things along. D'Ambrosia told The Register that the industry consensus is so strong it's feasible that a task force …

[Voland's right hand]

Autonegotiate over 8+ different bandwidths?

I do not want to be the one who builds the MII for this.

Autonegotation over 10 (if supported), 100, 1000, 2500, 5000, 10000... Right... Someone is smoking something very cool and not sharing it.

[John Tserkezis]

Re: Autonegotiate over 8+ different bandwidths?

"Autonegotation over 10 (if supported)"

Auto negotiating speeds isn't the issue. It's more how vendor A handles it, vs how vendor B handles it, and how they work together (or not work as the case may be). There's not only speed, there's full/half duplex, whether one side is willing to auto negotiate, and, if there's a mix, it's up to the auto negotiator to find out what's going on. And everyone handles it a bit different.

Depending on what you read, some make it look like it's a wonder it works at all.

[Danny 14]

Re: Autonegotiate over 8+ different bandwidths?

it works because they end up settling on 10mb half duplex with "I always wondered why that was a yellow light" and all the rest were green approach to networking.

[John Tserkezis]

"The driver for this is 802.11ac, since it's silly to have multi-gigabit wireless access points served by blue cables that can only limp along at 1 Gbps."

Not really. Actually, not at all.

Once you factor in real life speeds, you're looking at 60-70% of gigabit cable capacity, as long as the ac devices are within a few feet of each other - plenty of headroom. But also note, that case would probably not be seen too often, rather than the more realistic two walls and a whole lot more feet and the associated further reduced speed.

Stating 802.11ac is capable of "multi-gigabit" speeds is as pointless as stating USB3 is 10x faster than USB2. Which has been debunked as bullshit many times over with real life tests finding closer to 3x faster.

Don't get me wrong, I welcome the faster cabled speeds. There's more than a few applications that push gigabit cable to the limit at a sustained rate, and this would help that at least a bit. But don't quote raw WiFi speeds compared with a wired medium that is clearly faster, and expect me to swallow.

[Voland's right hand]

AC is not the driver

There's more than a few applications that push gigabit cable to the limit at a sustained rate

Linux and Mac network stack speeds:

UDP and other datagram protocols, single packet per RX/TX - 1.6Gbit per core

UDP and friends, multipacket rx/tx - 3GBit per core (Linux only, Mac does not have the relevant syscalls).

TCP without any offloads: 3-5Gbit per core

TCP with offloads - 6Gbits or thereabouts per core

Not sure what Windows state is, but it should be of the same order of magnitude.

So looking at these numbers, there is plenty of "need for speed" without having AC as an excuse.

[Danny 14]

Re: AC is not the driver

iscsi point to point will benefit from cheaper faster cabled links. I'm not sure how much server 2012 lies about network speeds but our point to point links between nodes certainly max out links when live migrating. 10gb is nice but still expensive (we are only talking 3 nodes here and when the network cards and switch cost more than a node you know it isn't right...)

Cant say I would like to use wireless for iscsi although I'm sure someone does somewhere.

[Fuzz]

USB 3

"Stating 802.11ac is capable of "multi-gigabit" speeds is as pointless as stating USB3 is 10x faster than USB2. Which has been debunked as bullshit many times over with real life tests finding closer to 3x faster."

In my experience USB 3.0 is around 10x faster than 2.0. If I connect an SSD to a USB-SATA adapter I get around 22MB/s using USB 2.0 and around 250MB/s using 3.0. Now this is nowhere near the theoretical limits of either interface but it is a 10 fold increase in speed between the two specs.

[P. Lee]

>"The driver for this is 802.11ac, since it's silly to have multi-gigabit wireless access points served by blue cables that can only limp along at 1 Gbps."

It depends on your traffic pattern. Wired is 1G full duplex, radio broadcast media is... not.

Still, there isn't that much headroom.

[HMB]

USB3 is 10x Faster, 802.11ac does reach multi gigabit, Moon landing real

"Stating 802.11ac is capable of "multi-gigabit" speeds is as pointless as stating USB3 is 10x faster than USB2. Which has been debunked as bullshit many times over with real life tests finding closer to 3x faster."

While I appreciate that the £30 non branded USB 3.0 flash drive you bought off ebay might not perform with 10x the speed of your USB 2.0 stick, it's total nonsense to rubbish an entire connectivity standard due to your poor imagination. USB 3.0 is being used by SSD drives today to achieve transfer rates of 450 MB/s. Just go Google it and update yourself on the topic. I'm surprised at how often people confuse a connectivity standard with other bottlenecks.

Now with 802.11ac I thought I was going to have to explain future options of 8x8 MIMO AP's for business needs like busy conference halls and the like. I thought I was going to have to explain that just because your smartphone can only do 1x1, 228 Mbps realtime throughput (433 Mbps on the PHY link) (https://www.youtube.com/watch?v=jsEutRQ2Zyo) on 802.11ac, don't forget the other people on the other spatial channels served by the AP. So I was going to come out with all this well reasoned stuff and point out that the peak of 802.11ac capability is 6.77 Gbps aggregate capacity at the PHY level which should be about 3.48 Gbps real throughput. This meaning that when these products come out that they really will need that faster Ethernet.

I was going to say all that, then I realised that ASUS have brought out the new ASUS RT-AC87U which is capable of real throughput to another RT-AC87U at 1029.8 Mbps real throughput on cnet tests. One consumer product, not implementing 160 MHz channels and only implementing 4 spatial streams out of 8 possible has already got you beat, not by much, but enough to start bottlenecking at the gigabit ethernet.

Would you like to give us any other advice on things that aren't true?

[Anonymous Coward]

"The driver for this is 802.11ac, since it's silly to have multi-gigabit wireless access points served by blue cables that can only limp along at 1 Gbps."

Not really. Actually, not at all.

Once you factor in real life speeds, you're looking at 60-70% of gigabit cable capacity, as long as the ac devices are within a few feet of each other - plenty of headroom. But also note, that case would probably not be seen too often, rather than the more realistic two walls and a whole lot more feet and the associated further reduced speed."

True for 802.11ac Wave 1, but not true for Wave 2.

Wave 1 = 1.3Gbps

Wave 2 = 1.3Gbps, 1.73Gbps, 2.6Gbps or 3.5Gbps. So, even at 1.73Gbps, 60% would exceed 1Gbps and the rest would easily consume it.

This is why they are looking at 2.5 and 5Gbps Ethernet speeds. Even if 802.11ac Wave 2 won't consume it, future standards could.

I personally think they should just use 10Gbps over Cat6a and get PoE running under 10Gbps.

[Callam McMillan]

Why?

Instead of bringing in 2.5/5 Gbps standards, why not focus on getting the cost of 10Gbps ports down to something sensible?

[Colin Miller]

Re: Why?

10GBASE-CX4 is limited to 15metres, most places that would uses multi-gigabit WiFi (airport terminals etc), will have more than 15metres between the AP and their upstream connection.

The point of the new standard is so that the old cabling won't have to be replaced with new for > 1Gbit/sec wired.

However, biting the bullet and putting in the best fibre-optic links might be more sensible, if any APs can talk directly to fibre ethernet.

[Preston Munchensonton]

Re: Why?

10GBASE-CX4 is limited to 15m, which is why no one would use CX4 for these sorts of connections. 10GBASE-T conforms to the 100m limit of other variants, but requires either Cat6a or Cat7 cabling to support full distance (both standards can handle the 500Mhz transmissions). 10GBASE-T with Cat6 is still an improvement over CX4, however, since it will reach 55m. All of this is per IEEE 802.3an.

[Sykobee]

Re: Why?

The article indicates it's to make use of existing cabling infrastructure, rather than requiring special cables with limited lengths (15 metres as your other respondent wrote).

10GigE is more of a rack/server room technology, rather than an office LAN technology.

So we have existing hundred metre lengths of Cat 5e cable, and instead of pushing 1GigE over it (already an achievement in itself) they want to try 2.5GigE or 5GigE speeds.

[Anonymous Coward]

"D'Ambrosia said the biggest challenge in kicking wired Ethernet's speeds along is going to be handling the crosstalk in cable bundles."

That is what Cat6a is for. It has better crosstalk than Cat6.

"A second challenge will be to develop modulation techniques that can maintain 2.5 Gbps and 5 Gbps transmissions over the full hundred metres of Ethernet cables. For that, D'Ambrosia told Vulture South, “people are looking at basing it on a slowed-down 10 Gbps Ethernet technology – that had a lot of discussion last week.”"

Once again Cat6a and 10Gbps is supported over Cat6a at a full hundred meters. Now they just need to get PoE running under 10Gbps.